tahmidalbi/agroscan-efficientnet-b0

AgroScan EfficientNet-B0

This repository stores the trained model artifact for AgroScan, an Android-based plant disease screening system.

Model

• Architecture: EfficientNet-B0
• Framework: PyTorch / TorchVision
• Task: Close-up plant leaf disease classification
• Input size: 224 x 224 RGB image
• Number of classes: 39

Intended Use

This model is intended for close-up leaf-level screening of 38 PlantVillage disease/healthy classes across 14 crops, plus a Non_leaf_or_unknown class.

It is not intended for full farm-scene diagnosis, whole-plant images from far away, unsupported crops, or guaranteed agricultural diagnosis.

Test Result

Held-out PlantVillage test set:

• Top-1 accuracy: 99.56%
• Top-3 accuracy: 99.99%
• Macro F1: 99.40%
• Test images: 8215

Important Warning

This model provides an AI screening result only. It is not a guaranteed diagnosis. Users should consult an agricultural expert if the crop is valuable or symptoms spread.

Files

• agroscan_efficientnet_b0.pth: trained PyTorch checkpoint
• labels.json: class label mapping
• model_manifest.json: model metadata and intended-use documentation
• plantvillage_test_metrics.json: test metrics
• plantvillage_test_confusion_matrix.png: confusion matrix