pest-yolo

Pest-yolo: A lightweight YOLOv5n-based pest detection algorithm

usage

The MobileNetv3, LCnet, and Ghostnet are defined in models/common.py.

These modules are registered in yolo.py.

The lightweight backbones are configured in models/yolo5n_pest/*.py.

CBAM is defined in cbam.py

Environment Configuration

Please see requirements.txt.

It is strongly recommended to use the latest YOLOv5 repository and integrate the above files into the latest code.

YOLOv5:https://github.com/ultralytics/yolov5

Trained Weight File for detecting 8 pests

8 pests: rice leaf roller, grub, wireworm, aphids, blister beetle, Miridae, unaspis_yanonensis, and Cicadellidae.

The trained weight file is available at: https://drive.google.com/drive/folders/1lNFcXtO-4-Wrbmzxuebt8lxpbqON1DTU?usp=drive_link

pest-yolo-ip102.pt is the weight file for 102 pests.

Performance comparison of integrating CBAM

Model Precision Recall mAP50 mAP50-95 Parameters

YOLOv5n 0.895 0.91 0.927 0.525 1,769,989

YOLOv5n-M3 0.861 0.87 0.895 (-3.2%) 0.476 1,171,493 (-33.8%)

Pest-YOLO 0.864 0.874 0.903 (-2.4%) 0.478 1,185,972 (-33%)

Performance comparison of different lightweight backbones

Model Precision Recall mAP50 mAP50-95 Parameters

YOLOv5n-ghostnet 0.853 0.838 0.891 0.473 1,362,697

YOLOv5n-lcnet 0.869 0.841 0.894 0.496 1,188,349

Pest-YOLO 0.864 0.874 0.903 0.478 1,185,972

The pest-yolo-ip102 model performance

Model Precision Recall mAP50 mAP50-95

pest-yolo-ip102 0.547 0.583 0.559 0.333