Site Classification

Resnet50 V2

Sharepoint link: model weights and source data.

• Version/date: 2 (June 10, 2025)
• Base model: ImageNet pre-trained ResNet50
• Training split: 80% training, 15% validation, 5% evaluation
• Input size: 640x640 pixels
• Classes: well_site, small_complex_site, compressor_stations, no_equipment_visible, large_complex_site, processing_facilities

Training Parameters:

• Optimizer: Adamw
• Learning rate: 0.001
• Decay rate: 0.0005
• Loss function: CrossEntropyLoss
• Batch size: 8
• Normalization: ImageNet mean/std normalization ([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
• Epochs: 50 (better performance than the 200th epoch output)