hoàn thành cơ bản các chức năng

2025-12-14 18:26:25 +07:00
parent 174de1034b
commit 623bc9c7dd
20 changed files with 1105 additions and 168 deletions
@@ -68,11 +68,15 @@ class TrainingConfig(BaseModel):
    resolution: int = 20  # 10m hoặc 20m
    
    # Model parameters
+    model_type: str = "xgboost"  # xgboost, random_forest, decision_tree, svm, cnn
    n_estimators: int = 100
    max_depth: int = 20
    learning_rate: float = 0.1
    use_gpu: bool = True
    
+    # Cache
+    use_cache: bool = True  # Cache dataset để test nhanh hơn
+    
    # Training data
    training_shapefile: str = "train/ST_training data_updated_1130points_new.shp"

@@ -205,6 +209,90 @@ async def stop_training():
    return {"message": "Đang dừng training..."}


+@app.post("/api/cache/clear")
+async def clear_cache():
+    """Xóa cache dataset"""
+    import shutil
+    cache_dir = Path("dataset_cache")
+    
+    if not cache_dir.exists():
+        return {"message": "Không có cache để xóa", "deleted": 0}
+    
+    # Count files
+    cache_files = list(cache_dir.glob("*.joblib"))
+    count = len(cache_files)
+    
+    # Delete all cache files
+    for cache_file in cache_files:
+        try:
+            cache_file.unlink()
+        except:
+            pass
+    
+    return {"message": f"Đã xóa {count} file cache", "deleted": count}
+
+
+@app.get("/api/cache/info")
+async def get_cache_info():
+    """Lấy thông tin về cache với metadata đầy đủ"""
+    cache_dir = Path("dataset_cache")
+    
+    if not cache_dir.exists():
+        return {"exists": False, "files": [], "total_size_mb": 0}
+    
+    cache_files = []
+    total_size = 0
+    
+    for cache_file in cache_dir.glob("*.joblib"):
+        size = cache_file.stat().st_size
+        total_size += size
+        
+        # Try to load metadata from cache
+        metadata = {}
+        try:
+            cached_data = joblib.load(cache_file)
+            if isinstance(cached_data, dict):
+                metadata = {
+                    "bbox": cached_data.get("bbox", []),
+                    "time_range": cached_data.get("time_range", ""),
+                    "resolution": cached_data.get("resolution", 20),
+                    "n_samples": len(cached_data.get("features", [])),
+                    "created": cached_data.get("timestamp", "")
+                }
+                # Parse time_range to get start/end dates
+                if metadata["time_range"]:
+                    time_parts = metadata["time_range"].split("/")
+                    if len(time_parts) == 2:
+                        metadata["start_date"] = time_parts[0]
+                        metadata["end_date"] = time_parts[1]
+                # Parse bbox to get min/max lon/lat
+                if metadata["bbox"] and len(metadata["bbox"]) == 4:
+                    metadata["min_lon"] = metadata["bbox"][0]
+                    metadata["min_lat"] = metadata["bbox"][1]
+                    metadata["max_lon"] = metadata["bbox"][2]
+                    metadata["max_lat"] = metadata["bbox"][3]
+        except Exception as e:
+            print(f"Error loading cache metadata: {e}")
+        
+        cache_files.append({
+            "filename": cache_file.name,
+            "size_mb": round(size / 1024 / 1024, 2),
+            "modified": datetime.fromtimestamp(cache_file.stat().st_mtime).isoformat(),
+            "metadata": metadata
+        })
+    
+    # Sort by modified time (newest first)
+    cache_files.sort(key=lambda x: x["modified"], reverse=True)
+    
+    return {
+        "exists": True,
+        "files": cache_files,
+        "count": len(cache_files),
+        "total_size_mb": round(total_size / 1024 / 1024, 2)
+    }
+
+
+
@app.get("/api/models/list")
 async def list_models():
    """Liệt kê các model đã train"""
@@ -288,10 +376,12 @@ async def run_training(config: TrainingConfig):
            cloud_cover=config.cloud_cover,
            resolution=config.resolution,
            training_shapefile=config.training_shapefile,
+            model_type=config.model_type,
            n_estimators=config.n_estimators,
            max_depth=config.max_depth,
            learning_rate=config.learning_rate,
            use_gpu=config.use_gpu,
+            use_cache=config.use_cache,
            status_callback=lambda msg: update_progress(msg),
            cancel_check=should_cancel
        )
@@ -360,6 +450,16 @@ async def run_prediction(config: PredictionConfig):
            model = model_data
            label_encoder = None
        
+        # Check if it's a CNN model (PyTorch)
+        is_cnn_model = hasattr(model, '__class__') and 'CNN' in model.__class__.__name__
+        if is_cnn_model:
+            prediction_status["progress"] = "Phát hiện PyTorch CNN model..."
+            # Import PyTorch if needed
+            try:
+                import torch
+            except ImportError:
+                raise ImportError("PyTorch is required for CNN prediction. Install: pip install torch")
+        
        prediction_status["progress"] = "Đang kết nối Microsoft Planetary Computer..."
        
        # Import and use Microsoft Planetary Computer STAC API
@@ -447,14 +547,13 @@ async def run_prediction(config: PredictionConfig):
            s1_items = s1_items[:config.max_scenes]
            prediction_status["progress"] = f"Đang xử lý {len(s1_items)} scenes Sentinel-1..."
            
-            # Load Sentinel-1 data
+            # Load Sentinel-1 data (without like= to avoid conflict with bbox/resolution)
            s1_data = load(
                s1_items,
                bbox=bbox,
                chunks={"time": 1, "x": 2048, "y": 2048},
                groupby="sat:absolute_orbit",
-                resolution=config.resolution,
-                like=ndvi_monthly  # Align with NDVI grid
+                resolution=config.resolution
            )
            
            # Extract VH and VV bands
@@ -514,13 +613,51 @@ async def run_prediction(config: PredictionConfig):
                    vv_t = np.resize(vv_t, n_pixels)
                features = np.column_stack([features, vv_t])
        
-        # Handle NaN values in features
+        # Handle NaN values in features✓ CNN PyTorch: Mạnh nhất với ảnh vệ tinh, tự học features, tương thích GPU tốt, cần pip install torch
        features = np.nan_to_num(features, nan=0.0)
        
        # ============ BƯỚC 6: DỰ ĐOÁN ============
+        # Check model's expected feature count and adjust
+        try:
+            # Get expected number of features from model
+            if is_cnn_model:
+                # For PyTorch CNN, get n_features from model
+                expected_features = model.n_features
+            elif hasattr(model, 'n_features_in_'):
+                expected_features = model.n_features_in_
+            elif hasattr(model, 'feature_names_in_'):
+                expected_features = len(model.feature_names_in_)
+            else:
+                # Try to get from booster for XGBoost
+                try:
+                    expected_features = model.get_booster().num_features()
+                except:
+                    expected_features = features.shape[1]
+            
+            prediction_status["progress"] = f"Model cần {expected_features} features, đang có {features.shape[1]} features..."
+            
+            # Adjust features to match model
+            if features.shape[1] > expected_features:
+                # Trim to expected number (use only first N features - NDVI only)
+                prediction_status["progress"] = f"Cắt bớt features từ {features.shape[1]} xuống {expected_features}..."
+                features = features[:, :expected_features]
+            elif features.shape[1] < expected_features:
+                # Pad with zeros or repeat last features
+                prediction_status["progress"] = f"Thêm features từ {features.shape[1]} lên {expected_features}..."
+                n_missing = expected_features - features.shape[1]
+                # Repeat last feature column to fill
+                padding = np.tile(features[:, -1:], (1, n_missing))
+                features = np.column_stack([features, padding])
+        except Exception as e:
+            prediction_status["progress"] = f"Không thể xác định số features của model, tiếp tục với {features.shape[1]} features..."
+        
        prediction_status["progress"] = f"Đang dự đoán với {features.shape[1]} features..."
        
        # Make prediction
+        if is_cnn_model:
+            # PyTorch CNN prediction
+            predictions = model.predict(features)
+        else:
            predictions = model.predict(features)
        
        # Decode labels if label_encoder exists
@@ -594,6 +731,49 @@ async def run_prediction(config: PredictionConfig):
        print(traceback.format_exc())


+@app.get("/api/predictions/list")
+async def list_predictions():
+    """Lấy danh sách các file prediction đã tạo"""
+    predictions_dir = Path("predictions")
+    predictions_dir.mkdir(exist_ok=True)
+    
+    predictions = []
+    for pred_file in predictions_dir.glob("*.tif"):
+        predictions.append({
+            "filename": pred_file.name,
+            "created": datetime.fromtimestamp(pred_file.stat().st_mtime).isoformat(),
+            "size_mb": round(pred_file.stat().st_size / 1024 / 1024, 2),
+            "download_url": f"/api/predictions/download/{pred_file.name}"
+        })
+    
+    # Sort by creation time (newest first)
+    predictions.sort(key=lambda x: x["created"], reverse=True)
+    return {"predictions": predictions}
+
+
+@app.get("/api/predictions/download/{filename}")
+async def download_prediction(filename: str):
+    """Download file prediction GeoTIFF"""
+    predictions_dir = Path("predictions")
+    file_path = predictions_dir / filename
+    
+    # Security check: ensure filename doesn't contain path traversal
+    if ".." in filename or "/" in filename or "\\" in filename:
+        raise HTTPException(status_code=400, detail="Invalid filename")
+    
+    if not file_path.exists():
+        raise HTTPException(status_code=404, detail=f"File không tồn tại: {filename}")
+    
+    return FileResponse(
+        path=str(file_path),
+        filename=filename,
+        media_type="image/tiff",
+        headers={
+            "Content-Disposition": f"attachment; filename={filename}"
+        }
+    )
+
+
 if __name__ == "__main__":
    print("=" * 70)
    print("🚀 LAND CLASSIFICATION TRAINING API SERVER")
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7318566231680cb0c97883b7a5e4177aa1bfeec462a0f52ed34aa103ddec210b
+size 20903
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cabf5e0241dcc3a73133ac8ae11171f34042c491f895b268c016407619bdfe1
+size 20903
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae2f82f6c837396729cc63efa41ee3048d9a7de3197e28318dc846be830239b9
+size 41536
@@ -0,0 +1,33 @@
+{
+  "timestamp": "2025-12-14T18:04:27.406540",
+  "data_source": "Microsoft Planetary Computer STAC",
+  "collections": [
+    "sentinel-2-l2a",
+    "sentinel-1-rtc"
+  ],
+  "features": [
+    "NDVI_mean",
+    "VH_dB_mean",
+    "VV_dB_mean"
+  ],
+  "training_samples": 510,
+  "testing_samples": 128,
+  "train_accuracy": 0.515686274509804,
+  "test_accuracy": 0.5,
+  "model_type": "cnn",
+  "device": "cpu",
+  "n_estimators": 50,
+  "max_depth": null,
+  "learning_rate": null,
+  "cnn_epochs": 25,
+  "n_features": 3,
+  "n_classes": 7,
+  "bbox": [
+    105.6,
+    9.3,
+    106.2,
+    9.8
+  ],
+  "time_range": "2023-03-01/2023-05-31",
+  "resolution": 20
+}
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:abc03d62c7f620b88150a6481143026d516fd3a34bd5ab67c734adac4a8900f9
+size 41536
@@ -0,0 +1,33 @@
+{
+  "timestamp": "2025-12-14T18:17:48.399298",
+  "data_source": "Microsoft Planetary Computer STAC",
+  "collections": [
+    "sentinel-2-l2a",
+    "sentinel-1-rtc"
+  ],
+  "features": [
+    "NDVI_mean",
+    "VH_dB_mean",
+    "VV_dB_mean"
+  ],
+  "training_samples": 510,
+  "testing_samples": 128,
+  "train_accuracy": 0.4803921568627451,
+  "test_accuracy": 0.484375,
+  "model_type": "cnn",
+  "device": "cpu",
+  "n_estimators": 50,
+  "max_depth": null,
+  "learning_rate": null,
+  "cnn_epochs": 25,
+  "n_features": 3,
+  "n_classes": 7,
+  "bbox": [
+    105.6,
+    9.3,
+    106.2,
+    9.8
+  ],
+  "time_range": "2023-03-01/2023-05-25",
+  "resolution": 20
+}
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21f906aa2a61d2e793a3463df95fe3e364134934af326771efae90d80caca419
+size 41536
@@ -0,0 +1,33 @@
+{
+  "timestamp": "2025-12-14T18:23:10.713912",
+  "data_source": "Microsoft Planetary Computer STAC",
+  "collections": [
+    "sentinel-2-l2a",
+    "sentinel-1-rtc"
+  ],
+  "features": [
+    "NDVI_mean",
+    "VH_dB_mean",
+    "VV_dB_mean"
+  ],
+  "training_samples": 510,
+  "testing_samples": 128,
+  "train_accuracy": 0.46862745098039216,
+  "test_accuracy": 0.4609375,
+  "model_type": "cnn",
+  "device": "cpu",
+  "n_estimators": 50,
+  "max_depth": null,
+  "learning_rate": null,
+  "cnn_epochs": 25,
+  "n_features": 3,
+  "n_classes": 7,
+  "bbox": [
+    105.6,
+    9.3,
+    106.2,
+    9.8
+  ],
+  "time_range": "2023-03-01/2023-05-31",
+  "resolution": 20
+}
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b717f564f9413a6e5c9cd3f7011cbc18be02479691d01c554986defb400f0490
+size 1347520
@@ -0,0 +1,31 @@
+{
+  "timestamp": "2025-12-14T13:40:03.325930",
+  "data_source": "Microsoft Planetary Computer STAC",
+  "collections": [
+    "sentinel-2-l2a",
+    "sentinel-1-rtc"
+  ],
+  "features": [
+    "NDVI_mean",
+    "VH_dB_mean",
+    "VV_dB_mean"
+  ],
+  "training_samples": 510,
+  "testing_samples": 128,
+  "train_accuracy": 1.0,
+  "test_accuracy": 0.578125,
+  "model_type": "XGBClassifier",
+  "device": "cuda:0",
+  "tree_method": "hist",
+  "n_estimators": 100,
+  "max_depth": 20,
+  "learning_rate": 0.1,
+  "bbox": [
+    105.6,
+    9.3,
+    106.2,
+    9.8
+  ],
+  "time_range": "2023-03-01/2023-05-31",
+  "resolution": 20
+}
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a44efc173619782c8add224e9024f33ef0150ab192b5435d009cb119c379f03c
+size 2088632
@@ -0,0 +1,31 @@
+{
+  "timestamp": "2025-12-14T16:52:19.862770",
+  "data_source": "Microsoft Planetary Computer STAC",
+  "collections": [
+    "sentinel-2-l2a",
+    "sentinel-1-rtc"
+  ],
+  "features": [
+    "NDVI_mean",
+    "VH_dB_mean",
+    "VV_dB_mean"
+  ],
+  "training_samples": 859,
+  "testing_samples": 215,
+  "train_accuracy": 0.9976717112922002,
+  "test_accuracy": 0.6837209302325581,
+  "model_type": "XGBClassifier",
+  "device": "cuda:0",
+  "tree_method": "hist",
+  "n_estimators": 100,
+  "max_depth": 20,
+  "learning_rate": 0.1,
+  "bbox": [
+    105.6,
+    9.3,
+    106.2,
+    9.8
+  ],
+  "time_range": "2023-03-01/2023-12-31",
+  "resolution": 20
+}
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:179d3a61420c1552e9653de5c7657665c9880ad29d751e56bae646fb3b634687
+size 73272920
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8fb13c2e466b9811104cbd7747ea6ca8c14ddb44802d934aadfe52a4b5dd916
+size 73272920
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2fe092e1fd96e56da519acffe5ca5c9a8246e796bc5700cffb9694dc99f3aec
+size 73272920
@@ -9,11 +9,109 @@ import geopandas as gpd
 from sklearn.model_selection import train_test_split
 from sklearn.preprocessing import LabelEncoder
 from sklearn.metrics import classification_report, confusion_matrix
+from sklearn.ensemble import RandomForestClassifier
+from sklearn.tree import DecisionTreeClassifier
+from sklearn.svm import SVC
 from xgboost import XGBClassifier
 import joblib
 from datetime import datetime
 import json
 import os
+import warnings
+import hashlib
+from pathlib import Path
+warnings.filterwarnings('ignore')
+
+# PyTorch for CNN
+try:
+    import torch
+    import torch.nn as nn
+    import torch.nn.functional as F
+    import torch.optim as optim
+    from torch.utils.data import TensorDataset, DataLoader
+    PYTORCH_AVAILABLE = True
+except ImportError:
+    PYTORCH_AVAILABLE = False
+    print("Warning: PyTorch not available. CNN model will not work.")
+
+# Define CNN model class for PyTorch
+class CNNClassifier(nn.Module):
+    def __init__(self, n_features, n_classes):
+        super(CNNClassifier, self).__init__()
+        self.n_features = n_features
+        self.n_classes = n_classes
+        
+        # For small feature sets (like 3 features), use simpler architecture
+        if n_features < 8:
+            # Simple fully connected network for small features
+            self.use_conv = False
+            self.fc1 = nn.Linear(n_features, 64)
+            self.dropout1 = nn.Dropout(0.3)
+            self.fc2 = nn.Linear(64, 128)
+            self.dropout2 = nn.Dropout(0.5)
+            self.fc3 = nn.Linear(128, n_classes)
+        else:
+            # CNN architecture for larger feature sets
+            self.use_conv = True
+            self.conv1 = nn.Conv1d(in_channels=1, out_channels=32, kernel_size=3, padding=1)
+            self.pool1 = nn.MaxPool1d(kernel_size=2)
+            self.conv2 = nn.Conv1d(in_channels=32, out_channels=64, kernel_size=3, padding=1)
+            self.pool2 = nn.MaxPool1d(kernel_size=2)
+            
+            # Calculate size after convolutions
+            conv_output_size = (n_features // 2 // 2) * 64
+            
+            # Fully connected layers
+            self.fc1 = nn.Linear(conv_output_size, 128)
+            self.dropout = nn.Dropout(0.5)
+            self.fc2 = nn.Linear(128, n_classes)
+    
+    def forward(self, x):
+        # x shape: (batch, n_features) or (batch, 1, n_features)
+        if self.use_conv:
+            # CNN path for larger feature sets
+            if len(x.shape) == 2:
+                x = x.unsqueeze(1)  # Add channel dimension
+            x = F.relu(self.conv1(x))
+            x = self.pool1(x)
+            x = F.relu(self.conv2(x))
+            x = self.pool2(x)
+            x = x.view(x.size(0), -1)  # Flatten
+            x = F.relu(self.fc1(x))
+            x = self.dropout(x)
+            x = self.fc2(x)
+        else:
+            # Fully connected path for small feature sets
+            if len(x.shape) == 3:
+                x = x.squeeze(1)  # Remove channel dimension if present
+            x = F.relu(self.fc1(x))
+            x = self.dropout1(x)
+            x = F.relu(self.fc2(x))
+            x = self.dropout2(x)
+            x = self.fc3(x)
+        return x
+    
+    def predict(self, X):
+        """Scikit-learn style predict method"""
+        self.eval()
+        with torch.no_grad():
+            if isinstance(X, np.ndarray):
+                X = torch.FloatTensor(X)
+            # Handle both 2D and 3D inputs
+            if not self.use_conv and len(X.shape) == 3:
+                X = X.squeeze(1)
+            elif self.use_conv and len(X.shape) == 2:
+                X = X.unsqueeze(1)
+            outputs = self(X)
+            _, predicted = torch.max(outputs, 1)
+            return predicted.cpu().numpy()
+    
+    def score(self, X, y):
+        """Scikit-learn style score method"""
+        predictions = self.predict(X)
+        if isinstance(y, torch.Tensor):
+            y = y.cpu().numpy()
+        return np.mean(predictions == y)

 # Microsoft Planetary Computer imports
 import planetary_computer
@@ -28,10 +126,12 @@ def train_model(
    cloud_cover=30,
    resolution=20,
    training_shapefile='train/ST_training data_updated_1130points_new.shp',
+    model_type='xgboost',
    n_estimators=100,
    max_depth=20,
    learning_rate=0.1,
    use_gpu=True,
+    use_cache=True,
    output_model_path=None,
    status_callback=None,
    cancel_check=None
@@ -77,10 +177,39 @@ def train_model(
        # Auto-generate output path if not provided
        if output_model_path is None:
            timestamp = datetime.now().strftime('%Y%m%d_%H%M%S')
-            output_model_path = f'model_train/model_xgboost_gpu_{timestamp}.joblib'
+            output_model_path = f'model_train/model_{model_type}_{timestamp}.joblib'
+        
+        # ============ CACHE SYSTEM ============
+        # Create cache directory
+        cache_dir = Path("dataset_cache")
+        cache_dir.mkdir(exist_ok=True)
+        
+        # Generate cache key from parameters
+        cache_params = f"{bbox}_{time_range}_{max_scenes}_{cloud_cover}_{resolution}"
+        cache_key = hashlib.md5(cache_params.encode()).hexdigest()
+        cache_file = cache_dir / f"training_data_{cache_key}.joblib"
+        
+        features = None
+        labels = None
+        
+        # Try to load from cache
+        if use_cache and cache_file.exists():
+            update_status(f"📦 Loading cached dataset from {cache_file.name}...", 5)
+            try:
+                cached_data = joblib.load(cache_file)
+                features = cached_data['features']
+                labels = cached_data['labels']
+                update_status(f"✅ Loaded {len(features)} samples from cache (skipped satellite download!)", 50)
+            except Exception as e:
+                update_status(f"⚠️ Cache load failed: {str(e)}, downloading fresh data...", 10)
+                features = None
+        
+        # If no cache or cache failed, download data
+        if features is None:
+            update_status("📡 Cache not found or disabled, downloading satellite data...", 10)
            
            # Connect to Microsoft Planetary Computer
-        update_status("Connecting to Microsoft Planetary Computer...", 0)
+            update_status("Connecting to Microsoft Planetary Computer...", 12)
            catalog = Client.open("https://planetarycomputer.microsoft.com/api/stac/v1")
            check_cancellation()
            
@@ -211,6 +340,23 @@ def train_model(
            
            update_status(f"Extracted {len(features)} valid training samples", 70)
            
+            # ============ SAVE TO CACHE ============
+            if use_cache:
+                update_status(f"💾 Saving dataset to cache for future use...", 72)
+                try:
+                    cache_data = {
+                        'features': features,
+                        'labels': labels,
+                        'bbox': bbox,
+                        'time_range': time_range,
+                        'resolution': resolution,
+                        'timestamp': datetime.now().isoformat()
+                    }
+                    joblib.dump(cache_data, cache_file)
+                    update_status(f"✅ Cached to {cache_file.name}", 75)
+                except Exception as e:
+                    update_status(f"⚠️ Cache save failed: {str(e)}", 75)
+        
        # Encode labels
        label_encoder = LabelEncoder()
        labels_encoded = label_encoder.fit_transform(labels)
@@ -220,33 +366,115 @@ def train_model(
            features, labels_encoded, test_size=0.2, random_state=42, stratify=labels_encoded
        )
        
-        # Train XGBoost model
-        update_status("Training XGBoost model on GPU...", 75)
+        # Train model based on selected type
+        update_status(f"Training {model_type.upper()} model...", 75)
        
        device = 'cuda:0' if use_gpu else 'cpu'
        
-        xgb_model = XGBClassifier(
+        if model_type == 'xgboost':
+            model = XGBClassifier(
                n_estimators=n_estimators,
                max_depth=max_depth,
                learning_rate=learning_rate,
-            device=device,
+                device=device if use_gpu else 'cpu',
                tree_method='hist',
                random_state=42,
                eval_metric='mlogloss',
                verbosity=0
            )
+        elif model_type == 'random_forest':
+            model = RandomForestClassifier(
+                n_estimators=n_estimators,
+                max_depth=max_depth,
+                random_state=42,
+                n_jobs=-1,  # Use all cores
+                verbose=0
+            )
+        elif model_type == 'decision_tree':
+            model = DecisionTreeClassifier(
+                max_depth=max_depth,
+                random_state=42
+            )
+        elif model_type == 'svm':
+            model = SVC(
+                kernel='rbf',
+                random_state=42,
+                verbose=False
+            )
+        elif model_type == 'cnn':
+            if not PYTORCH_AVAILABLE:
+                raise ImportError("PyTorch is required for CNN. Install: pip install torch")
            
-        xgb_model.fit(X_train, y_train)
+            # CNN requires reshaping data
+            n_features = X_train.shape[1]
+            n_classes = len(np.unique(y_train))
+            
+            # Build PyTorch CNN model
+            device = torch.device('cuda' if torch.cuda.is_available() and use_gpu else 'cpu')
+            update_status(f"Building CNN model on {device}...", 75)
+            
+            model = CNNClassifier(n_features, n_classes).to(device)
+            
+            # Convert to PyTorch tensors
+            X_train_tensor = torch.FloatTensor(X_train).unsqueeze(1)  # Add channel dim: (N, 1, features)
+            y_train_tensor = torch.LongTensor(y_train)
+            X_test_tensor = torch.FloatTensor(X_test).unsqueeze(1)
+            y_test_tensor = torch.LongTensor(y_test)
+            
+            # Create data loaders
+            train_dataset = TensorDataset(X_train_tensor, y_train_tensor)
+            train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True)
+            
+            # Loss and optimizer
+            criterion = nn.CrossEntropyLoss()
+            optimizer = optim.Adam(model.parameters(), lr=0.001)
+            
+            # Train CNN
+            update_status("Training CNN model with PyTorch...", 80)
+            epochs = min(50, n_estimators // 2)  # Use n_estimators as epochs
+            
+            model.train()
+            for epoch in range(epochs):
+                epoch_loss = 0.0
+                for batch_X, batch_y in train_loader:
+                    batch_X, batch_y = batch_X.to(device), batch_y.to(device)
+                    
+                    optimizer.zero_grad()
+                    outputs = model(batch_X)
+                    loss = criterion(outputs, batch_y)
+                    loss.backward()
+                    optimizer.step()
+                    
+                    epoch_loss += loss.item()
+                
+                if (epoch + 1) % 10 == 0:
+                    avg_loss = epoch_loss / len(train_loader)
+                    update_status(f"CNN Epoch {epoch+1}/{epochs}, Loss: {avg_loss:.4f}", 80 + (epoch / epochs) * 10)
+            
+            # Move model to CPU for saving (compatible with non-GPU systems)
+            model = model.cpu()
+            model.device_used = str(device)
+        else:
+            raise ValueError(f"Unknown model type: {model_type}. Choose: xgboost, random_forest, decision_tree, svm, cnn")
+        
+        # Fit non-CNN models
+        if model_type != 'cnn':
+            model.fit(X_train, y_train)
        
        # Evaluate
        update_status("Evaluating model...", 90)
-        train_score = xgb_model.score(X_train, y_train)
-        test_score = xgb_model.score(X_test, y_test)
+        if model_type == 'cnn':
+            # PyTorch CNN evaluation
+            train_score = model.score(X_train, y_train)
+            test_score = model.score(X_test, y_test)
+        else:
+            train_score = model.score(X_train, y_train)
+            test_score = model.score(X_test, y_test)
        
        # Save model
        update_status("Saving model...", 95)
        os.makedirs(os.path.dirname(output_model_path), exist_ok=True)
-        joblib.dump({'model': xgb_model, 'label_encoder': label_encoder}, output_model_path)
+        joblib.dump({'model': model, 'label_encoder': label_encoder}, output_model_path)
        
        # Save model info
        info = {
@@ -258,12 +486,14 @@ def train_model(
            "testing_samples": len(X_test),
            "train_accuracy": float(train_score),
            "test_accuracy": float(test_score),
-            "model_type": "XGBClassifier",
-            "device": device,
-            "tree_method": "hist",
-            "n_estimators": n_estimators,
-            "max_depth": max_depth,
-            "learning_rate": learning_rate,
+            "model_type": model_type,
+            "device": device if model_type == 'xgboost' else 'cpu',
+            "n_estimators": n_estimators if model_type in ['xgboost', 'random_forest', 'cnn'] else None,
+            "max_depth": max_depth if model_type != 'cnn' else None,
+            "learning_rate": learning_rate if model_type == 'xgboost' else None,
+            "cnn_epochs": min(50, n_estimators // 2) if model_type == 'cnn' else None,
+            "n_features": X_train.shape[1],
+            "n_classes": len(np.unique(y_train)),
            "bbox": bbox,
            "time_range": time_range,
            "resolution": resolution
@@ -258,6 +258,29 @@
            margin: 5px 0;
            color: #666;
        }
+        
+        /* Animations for notifications */
+        @keyframes slideIn {
+            from {
+                transform: translateX(400px);
+                opacity: 0;
+            }
+            to {
+                transform: translateX(0);
+                opacity: 1;
+            }
+        }
+        
+        @keyframes slideOut {
+            from {
+                transform: translateX(0);
+                opacity: 1;
+            }
+            to {
+                transform: translateX(400px);
+                opacity: 0;
+            }
+        }
    </style>
 </head>
 <body>
@@ -369,7 +392,18 @@
                        </div>
                    </div>
                    
-                    <h3 style="margin: 20px 0 15px; color: #667eea;">🛰️ Dữ Liệu Vệ Tinh</h3>
+                    <h3 style="margin: 20px 0 15px; color: #667eea;">� Dataset Cache Preset</h3>
+                    <div class="form-group">
+                        <label>Chọn Dataset đã cache:</label>
+                        <select id="cachePreset" style="font-size: 14px;">
+                            <option value="">-- Không dùng cache preset --</option>
+                        </select>
+                        <div style="font-size: 12px; color: #666; margin-top: 5px;">
+                            💡 Chọn dataset cache có sẵn để tự động điền các thông số tương ứng
+                        </div>
+                    </div>
+                    
+                    <h3 style="margin: 20px 0 15px; color: #667eea;">�🛰️ Dữ Liệu Vệ Tinh</h3>
                    <div class="form-row">
                        <div class="form-group">
                            <label>Số scenes tối đa:</label>
@@ -390,8 +424,27 @@
                    </div>
                    
                    <h3 style="margin: 20px 0 15px; color: #667eea;">🤖 Model Parameters</h3>
-                    <div class="form-row">
+                    
+                    <!-- Model Type Selection -->
                    <div class="form-group">
+                        <label><strong>🧠 Loại Model:</strong></label>
+                        <select id="modelType" required style="font-weight: 600; font-size: 14px;">
+                            <option value="xgboost" selected>🚀 XGBoost (Nhanh, Chính xác cao, Hỗ trợ GPU)</option>
+                            <option value="random_forest">🌲 Random Forest (Ổn định, Không cần GPU)</option>
+                            <option value="decision_tree">🌳 Decision Tree (Đơn giản, Nhanh nhất)</option>
+                            <option value="svm">🎯 SVM (Chính xác, Chậm với dữ liệu lớn)</option>
+                            <option value="cnn">🧠 CNN - Deep Learning (PyTorch, Tốt với ảnh vệ tinh, Hỗ trợ GPU)</option>
+                        </select>
+                        <div style="margin-top: 8px; padding: 10px; background: #e7f3ff; border-radius: 5px; font-size: 12px;">
+                            <span id="modelTypeDesc" style="color: #1976d2;">
+                                ✓ XGBoost: Tốt nhất cho dữ liệu satellite, hỗ trợ GPU, training nhanh
+                            </span>
+                        </div>
+                    </div>
+                    
+                    <!-- Model-specific parameters -->
+                    <div class="form-row">
+                        <div class="form-group" id="nEstimatorsGroup">
                            <label>N Estimators:</label>
                            <input type="number" id="nEstimators" value="100" min="10" max="1000" required>
                        </div>
@@ -399,11 +452,11 @@
                            <label>Max Depth:</label>
                            <input type="number" id="maxDepth" value="20" min="1" max="50" required>
                        </div>
-                        <div class="form-group">
+                        <div class="form-group" id="learningRateGroup">
                            <label>Learning Rate:</label>
                            <input type="number" step="0.01" id="learningRate" value="0.1" min="0.01" max="1" required>
                        </div>
-                        <div class="form-group">
+                        <div class="form-group" id="useGpuGroup">
                            <label>Use GPU:</label>
                            <select id="useGpu" required>
                                <option value="true" selected>Có (RTX 4060)</option>
@@ -412,6 +465,26 @@
                        </div>
                    </div>
                    
+                    <!-- Cache Option -->
+                    <div class="form-group" style="margin-top: 20px; padding: 15px; background: #fff8dc; border-radius: 8px; border: 1px solid #ffc107;">
+                        <label style="display: flex; align-items: center; cursor: pointer; margin: 0;">
+                            <input type="checkbox" id="useCache" checked style="width: 18px; height: 18px; margin-right: 10px;">
+                            <span style="font-weight: 600; color: #856404;">💾 Sử dụng Cache Dataset</span>
+                        </label>
+                        <div style="font-size: 12px; color: #856404; margin-top: 8px; margin-left: 28px;">
+                            ✅ <strong>Khuyến nghị:</strong> Bật để test nhanh hơn. Lần đầu load dữ liệu sẽ chậm, nhưng các lần sau rất nhanh (không cần download lại từ satellite).<br>
+                            📊 <span id="cacheInfo">Đang kiểm tra cache...</span>
+                        </div>
+                        <div style="margin-top: 10px; margin-left: 28px;">
+                            <button type="button" class="btn btn-secondary" onclick="clearCache()" style="font-size: 12px; padding: 5px 10px; background: #dc3545;">
+                                🗑️ Xóa Cache
+                            </button>
+                            <button type="button" class="btn btn-secondary" onclick="loadCacheInfo()" style="font-size: 12px; padding: 5px 10px; background: #17a2b8; margin-left: 5px;">
+                                🔄 Refresh
+                            </button>
+                        </div>
+                    </div>
+                    
                    <div style="margin-top: 30px; text-align: center;">
                        <button type="submit" class="btn btn-primary" id="startBtn">
                            🚀 Bắt Đầu Training
@@ -526,15 +599,16 @@
                            <h3 style="color: #28a745; margin-bottom: 10px;">✅ Kết Quả Dự Đoán</h3>
                            <div style="background: linear-gradient(135deg, #d4edda 0%, #c3e6cb 100%); padding: 20px; border-radius: 8px; border: 2px solid #28a745;">
                                <div id="predResultText" style="font-size: 14px; line-height: 1.8;"></div>
-                                <div style="margin-top: 15px; text-align: center;">
-                                    <button onclick="downloadPredictionResult()" class="btn btn-secondary" style="background: #28a745;">
-                                        📥 Tải Kết Quả
-                                    </button>
-                                    <button onclick="viewPredictionResult()" class="btn btn-secondary" style="background: #17a2b8; margin-left: 10px;">
-                                        👁️ Xem Chi Tiết
-                                    </button>
+                                <div id="downloadLinkContainer" style="margin-top: 15px; text-align: center;"></div>
                            </div>
                        </div>
+                        
+                        <!-- Previous Predictions List -->
+                        <div id="predictionsListSection" style="margin-top: 20px;">
+                            <h3 style="color: #ff6b6b; margin-bottom: 10px;">📂 Các File Dự Đoán Đã Tạo</h3>
+                            <div id="predictionsList" style="background: #f8f9fa; padding: 15px; border-radius: 8px; max-height: 200px; overflow-y: auto;">
+                                <p style="color: #666; text-align: center;">Đang tải...</p>
+                            </div>
                        </div>
                    </div>
                </div>
@@ -631,10 +705,12 @@
                max_scenes: parseInt(document.getElementById('maxScenes').value),
                cloud_cover: parseInt(document.getElementById('cloudCover').value),
                resolution: parseInt(document.getElementById('resolution').value),
+                model_type: document.getElementById('modelType').value,
                n_estimators: parseInt(document.getElementById('nEstimators').value),
                max_depth: parseInt(document.getElementById('maxDepth').value),
                learning_rate: parseFloat(document.getElementById('learningRate').value),
-                use_gpu: document.getElementById('useGpu').value === 'true'
+                use_gpu: document.getElementById('useGpu').value === 'true',
+                use_cache: document.getElementById('useCache').checked
            };
            
            try {
@@ -994,10 +1070,209 @@
            }
        }
        
+        // Model type change handler
+        function updateModelTypeUI() {
+            const modelType = document.getElementById('modelType').value;
+            const desc = document.getElementById('modelTypeDesc');
+            const nEstimatorsGroup = document.getElementById('nEstimatorsGroup');
+            const learningRateGroup = document.getElementById('learningRateGroup');
+            const useGpuGroup = document.getElementById('useGpuGroup');
+            
+            const descriptions = {
+                'xgboost': '✓ XGBoost: Tốt nhất cho dữ liệu satellite, hỗ trợ GPU, training nhanh',
+                'random_forest': '✓ Random Forest: Ổn định, không overfitting, phù hợp mọi kích thước dữ liệu',
+                'decision_tree': '✓ Decision Tree: Đơn giản nhất, nhanh nhất, dễ hiểu, phù hợp để test nhanh',
+                'svm': '✓ SVM: Chính xác cao với dữ liệu nhỏ, chậm với dữ liệu lớn',
+                'cnn': '✓ CNN PyTorch: Mạnh nhất với ảnh vệ tinh, tự học features, tương thích GPU tốt, cần pip install torch'
+            };
+            
+            desc.textContent = descriptions[modelType];
+            
+            // Show/hide parameters based on model type
+            if (modelType === 'xgboost') {
+                nEstimatorsGroup.style.display = '';
+                learningRateGroup.style.display = '';
+                useGpuGroup.style.display = '';
+            } else if (modelType === 'random_forest') {
+                nEstimatorsGroup.style.display = '';
+                learningRateGroup.style.display = 'none';
+                useGpuGroup.style.display = 'none';
+            } else if (modelType === 'decision_tree') {
+                nEstimatorsGroup.style.display = 'none';
+                learningRateGroup.style.display = 'none';
+                useGpuGroup.style.display = 'none';
+            } else if (modelType === 'svm') {
+                nEstimatorsGroup.style.display = 'none';
+                learningRateGroup.style.display = 'none';
+                useGpuGroup.style.display = 'none';
+            } else if (modelType === 'cnn') {
+                // CNN uses n_estimators as epochs and supports GPU
+                nEstimatorsGroup.style.display = '';
+                document.querySelector('#nEstimatorsGroup label').textContent = 'Epochs (số lần training):';
+                document.getElementById('nEstimators').value = 50;
+                learningRateGroup.style.display = 'none';
+                useGpuGroup.style.display = '';  // Show GPU option for CNN
+            }
+            
+            // Reset n_estimators label for non-CNN
+            if (modelType !== 'cnn' && modelType !== 'decision_tree' && modelType !== 'svm') {
+                document.querySelector('#nEstimatorsGroup label').textContent = 'N Estimators:';
+            }
+        }
+        
+        // ============== CACHE MANAGEMENT ==============
+        
+        let cacheFiles = [];  // Store cache files data
+        
+        // Load cache info and populate preset dropdown
+        async function loadCacheInfo() {
+            try {
+                const response = await fetch(`${API_BASE}/cache/info`);
+                const data = await response.json();
+                
+                // Update cache info display
+                const cacheInfoSpan = document.getElementById('cacheInfo');
+                if (data.exists && data.count > 0) {
+                    cacheInfoSpan.innerHTML = `✅ Có <strong>${data.count}</strong> file cache (${data.total_size_mb} MB)`;
+                    cacheInfoSpan.style.color = '#28a745';
+                } else {
+                    cacheInfoSpan.innerHTML = '❌ Chưa có cache nào';
+                    cacheInfoSpan.style.color = '#6c757d';
+                }
+                
+                // Populate cache preset dropdown
+                const cachePreset = document.getElementById('cachePreset');
+                cachePreset.innerHTML = '<option value="">-- Không dùng cache preset --</option>';
+                
+                if (data.exists && data.files && data.files.length > 0) {
+                    cacheFiles = data.files;
+                    data.files.forEach((file, index) => {
+                        const meta = file.metadata;
+                        const option = document.createElement('option');
+                        option.value = index;
+                        
+                        // Format display text with metadata
+                        let displayText = `Cache #${index + 1}`;
+                        if (meta && meta.n_samples) {
+                            displayText += ` (${meta.n_samples} mẫu`;
+                            if (meta.start_date && meta.end_date) {
+                                displayText += `, ${meta.start_date} → ${meta.end_date}`;
+                            }
+                            if (meta.resolution) {
+                                displayText += `, ${meta.resolution}m`;
+                            }
+                            displayText += ')';
+                        } else {
+                            displayText += ` (${file.size_mb} MB, ${new Date(file.modified).toLocaleString('vi-VN')})`;
+                        }
+                        
+                        option.textContent = displayText;
+                        cachePreset.appendChild(option);
+                    });
+                }
+            } catch (error) {
+                console.error('Error loading cache info:', error);
+                document.getElementById('cacheInfo').innerHTML = '⚠️ Lỗi kiểm tra cache';
+            }
+        }
+        
+        // Apply cache preset to form inputs
+        function applyCachePreset() {
+            const selectIndex = document.getElementById('cachePreset').value;
+            
+            if (selectIndex === '') {
+                // Clear preset - no auto-fill
+                return;
+            }
+            
+            const fileData = cacheFiles[parseInt(selectIndex)];
+            if (!fileData || !fileData.metadata) {
+                alert('Không có metadata cho cache này');
+                return;
+            }
+            
+            const meta = fileData.metadata;
+            
+            // Fill bbox inputs
+            if (meta.min_lon !== undefined) {
+                document.getElementById('minLon').value = meta.min_lon;
+                document.getElementById('maxLon').value = meta.max_lon;
+                document.getElementById('minLat').value = meta.min_lat;
+                document.getElementById('maxLat').value = meta.max_lat;
+                
+                // Update map rectangle
+                if (rectangle) {
+                    drawnItems.removeLayer(rectangle);
+                }
+                const bounds = L.latLngBounds(
+                    [meta.min_lat, meta.min_lon],
+                    [meta.max_lat, meta.max_lon]
+                );
+                rectangle = L.rectangle(bounds, {color: '#3388ff', weight: 3, fillOpacity: 0.2});
+                drawnItems.addLayer(rectangle);
+                map.fitBounds(bounds);
+            }
+            
+            // Fill time range inputs
+            if (meta.start_date) {
+                document.getElementById('startDate').value = meta.start_date;
+            }
+            if (meta.end_date) {
+                document.getElementById('endDate').value = meta.end_date;
+            }
+            
+            // Fill resolution if available
+            if (meta.resolution) {
+                document.getElementById('resolution').value = meta.resolution;
+            }
+            
+            // Note to user
+            const presetNote = `📌 Đã áp dụng cache preset: ${meta.n_samples || '?'} mẫu, ${meta.start_date || '?'} → ${meta.end_date || '?'}`;
+            console.log(presetNote);
+            
+            // Show notification
+            const notification = document.createElement('div');
+            notification.style.cssText = 'position:fixed;top:20px;right:20px;background:#28a745;color:white;padding:15px 20px;border-radius:8px;box-shadow:0 4px 6px rgba(0,0,0,0.1);z-index:10000;animation:slideIn 0.3s ease-out;';
+            notification.innerHTML = `<strong>✅ Cache Preset Applied</strong><br>${presetNote}`;
+            document.body.appendChild(notification);
+            
+            setTimeout(() => {
+                notification.style.animation = 'slideOut 0.3s ease-out';
+                setTimeout(() => notification.remove(), 300);
+            }, 3000);
+        }
+        
+        // Clear cache
+        async function clearCache() {
+            if (!confirm('Bạn có chắc muốn xóa toàn bộ cache?\n\nCache giúp test nhanh hơn bằng cách lưu lại dữ liệu đã download.')) {
+                return;
+            }
+            
+            try {
+                const response = await fetch(`${API_BASE}/cache/clear`, {
+                    method: 'POST'
+                });
+                const result = await response.json();
+                alert(result.message);
+                loadCacheInfo();  // Refresh info
+            } catch (error) {
+                alert('Lỗi xóa cache: ' + error.message);
+            }
+        }
+        
        // Initialize map when page loads
        document.addEventListener('DOMContentLoaded', function() {
            initMap();
            initPredictionMap();
+            loadPredictionsList();  // Load predictions list on page load
+            loadCacheInfo();  // Load cache info
+            
+            // Add model type change listener
+            document.getElementById('modelType').addEventListener('change', updateModelTypeUI);
+            updateModelTypeUI();  // Initial update
+            
+            // Add cache preset change listener
+            document.getElementById('cachePreset').addEventListener('change', applyCachePreset);
        });
        
        // ============== PREDICTION FUNCTIONALITY ==============
@@ -1207,6 +1482,10 @@
            // Store result globally for download/view functions
            window.lastPredictionResult = result;
            
+            // Extract filename from path
+            const filename = result.output_file.split('/').pop();
+            const downloadUrl = `${API_BASE}/predictions/download/${filename}`;
+            
            resultText.innerHTML = `
                <div style="margin-bottom: 10px;">
                    <strong>📁 File kết quả:</strong><br>
@@ -1220,19 +1499,76 @@
                </div>
            `;
            
+            // Add download link
+            const downloadContainer = document.getElementById('downloadLinkContainer');
+            downloadContainer.innerHTML = `
+                <a href="${downloadUrl}" 
+                   class="btn btn-primary" 
+                   style="background: #28a745; padding: 12px 24px; text-decoration: none; display: inline-block; margin-right: 10px;"
+                   download="${filename}">
+                    📥 Tải GeoTIFF
+                </a>
+                <button onclick="viewPredictionResult()" class="btn btn-secondary" style="background: #17a2b8;">
+                    👁️ Xem Chi Tiết
+                </button>
+                <div style="margin-top: 10px; font-size: 12px; color: #666;">
+                    Hoặc copy link: <a href="${downloadUrl}" target="_blank" style="color: #28a745;">${downloadUrl}</a>
+                </div>
+            `;
+            
            resultDiv.style.display = 'block';
+            
+            // Refresh predictions list
+            loadPredictionsList();
        }
        
        // Download prediction result
        function downloadPredictionResult() {
            if (window.lastPredictionResult) {
                const result = window.lastPredictionResult;
-                alert('File kết quả: ' + result.output_file + '\n\nĐể tải file, vui lòng truy cập thư mục predictions/ trên server.');
+                const filename = result.output_file.split('/').pop();
+                const downloadUrl = `${API_BASE}/predictions/download/${filename}`;
+                window.open(downloadUrl, '_blank');
            } else {
                alert('Chưa có kết quả dự đoán nào!');
            }
        }
        
+        // Load list of previous predictions
+        async function loadPredictionsList() {
+            try {
+                const response = await fetch(`${API_BASE}/predictions/list`);
+                const data = await response.json();
+                
+                const listDiv = document.getElementById('predictionsList');
+                
+                if (data.predictions && data.predictions.length > 0) {
+                    listDiv.innerHTML = data.predictions.map(pred => `
+                        <div style="display: flex; justify-content: space-between; align-items: center; padding: 10px; margin-bottom: 8px; background: white; border-radius: 6px; border: 1px solid #ddd;">
+                            <div style="flex: 1;">
+                                <strong style="color: #333;">📄 ${pred.filename}</strong>
+                                <div style="font-size: 12px; color: #666; margin-top: 3px;">
+                                    📅 ${new Date(pred.created).toLocaleString('vi-VN')} | 💾 ${pred.size_mb} MB
+                                </div>
+                            </div>
+                            <a href="${pred.download_url}" 
+                               class="btn btn-secondary" 
+                               style="background: #28a745; padding: 6px 12px; font-size: 12px; text-decoration: none;"
+                               download="${pred.filename}">
+                                📥 Tải về
+                            </a>
+                        </div>
+                    `).join('');
+                } else {
+                    listDiv.innerHTML = '<p style="color: #666; text-align: center;">Chưa có file dự đoán nào.</p>';
+                }
+            } catch (error) {
+                console.error('Error loading predictions list:', error);
+                document.getElementById('predictionsList').innerHTML = 
+                    '<p style="color: #dc3545; text-align: center;">Lỗi tải danh sách: ' + error.message + '</p>';
+            }
+        }
+        
        // View prediction result details
        function viewPredictionResult() {
            if (window.lastPredictionResult) {