GAS-YOLO: A Lightweight and Efficient Object Detection Model for Tomato Foliar Disease Detection

The continuous expansion of tomato cultivation necessitates efficient methods for detecting foliar diseases, which are critical for maintaining yield and quality. Addressing the need for computational efficiency, this paper presents GAS-YOLO, a lightweight object detection model tailored for tomato leaf disease identification. GAS-YOLO integrates GhostConv for efficient feature extraction, ASE for adaptive channel weighting, and Distance Intersection over Union (DIoU) loss for enhanced bounding box regression. Experimental results demonstrate that GAS-YOLO achieves comparable high accuracy to the baseline model while significantly reducing computational demands: Giga Floating-point Operations Per Second (GFLOPs) decreased by 12.6% and Parameters (Params) by 10.8%. Furthermore, inference speed, measured in Frames Per Second (FPS), increased by 17.0%, effectively achieving model lightweighting. Evaluation against leading models confirms GAS-YOLO's superior equilibrium between precision and computational efficacy. The proposed framework delivers a deployable methodology for instantaneous pathology identification in precision farming scenarios.