EggplantDet: An efficient lightweight model for eggplant disease detection

Abstract

To address the challenges of low accuracy and slow detection speed in eggplant disease detection within complex environments, this study proposes a lightweight algorithm based on the YOLOv8 model. Key enhancements include the integration of FasterNet for efficient feature extraction, TAM attention modules for improved feature representation, a dedicated small-object detection head, and an adaptive WIoU loss function for optimized bounding box regression. Experiments on our custom dataset demonstrate that the proposed method achieves a mean Average Precision (mAP) of 92.61 % and a detection speed of 88.39 frames per second, significantly enhancing detection performance in complex scenarios while achieving an optimal balance between accuracy and speed.

Topics & Concepts