An Improved Lightweight YOLOv5 Algorithm for Detecting Strawberry Diseases

Abstract

This paper proposes an improved lightweight YOLOv5 model for the real-time detection of strawberry diseases. The ghost convolution (GhostConv) module is incorporated into the YOLOv5 network, reducing the parameter numbers and floating-point operations (FLOPs) for extracting feature information using the backbone network. An involution operator is utilized in the backbone network to expand the receptive field, enhance the spatial information on strawberry disease characteristics, and reduce the number of FLOPs in the model. A convolutional block attention module (CBAM) is incorporated into the backbone network to enhance the network's ability to extract strawberry disease features and suppress non-critical information. The upsampling module is replaced by a lightweight upsampling operator called Content-Aware ReAssembly of Features (CARAFE), which extracts feature map information and enhances the ability to focus on strawberry disease features. The experimental results on an open-source strawberry disease dataset show that the model achieves mean average precision (mAP)@0.5 of 94.7% with 3.9 M parameters and 3.6 G FLOPs. The improved model has higher detection precision than the original one and lower hardware requirements, providing a new strategy for strawberry disease identification and control.