Research on the Strawberry Disease and Pest Identification Algorithm Based on the YOLOv8 Framework

Abstract

As an important economic crop, strawberries are highly susceptible to various pests and diseases during their growth, seriously affecting yield and quality. Traditional manual detection methods are time-consuming, labor-intensive, and difficult to apply promptly and comprehensively. To promote the development of intelligent pest and disease detection systems, improve strawberry cultivation efficiency, and ensure strawberry quality, a lightweight Yolo-CAD model based on Yolov8n was proposed, designed to address issues such as multi-labeling and inaccurate detection. The model uses ADown[10] as the downsampling module to reduce feature loss during pooling and optimizes Yolov8's Neck based on the Cross-Scale Feature Fusion Module (CCFM), integrating multi-scale features and contextual information to enhance the model's adaptability to scale variations and improve its ability to detect small objects. Experimental results show that the enhanced Yolo-CAD model achieves a mean accuracy (mAP50) of 76.33%, a precision of 76.19%, and a recall of 71.20% on the validation set. Compared to SSD, Retinanet, Yolov3, Yolov5n, Yolov8n, Yolov9t, and Yolov10n, the average precision (mAP50) was improved by 11.58, 6.22, 7.22, 6.25, 5.08, 5.44, and 6.21 percentage points, respectively. Moreover, the size of Yolo-CAD model is only 3.37MB, 43.65% smaller than Yolov8n, making it more suitable for deployment on mobile platforms with limited computational power. The proposed Yolo-CAD model has shown promising results in identifying and detecting strawberry pests and diseases, and can provide technical support for pest prevention and automated harvesting.