Research on Real-Time Detection Method of Beet Damage Based on Improved YOLOv8s
Abstract：To address the issue of insufficient recognition accuracy and poor real-time performance in mechanized harvesting operations, which affects the accuracy of damage rate evaluation, this study proposes an improved lightweight real-time object detection model called LB-YOLOv8s. The model is based on YOLOv8s, with a lightweight backbone network named BS-GhostNet designed, and it integrates a recursive dilated self-attention mechanism to enhance the model's feature extraction capabilities under complex backgrounds and occlusion conditions, thereby improving the recognition accuracy of intact and damaged beets. Experimental results show that LB-YOLOv8s achieved an average precision of 92.6% on the constructed beet dataset, with the model parameter count controlled at 9.8×10^6. It achieves a推理 speed of 145 FPS on a desktop computer and 55 FPS on an embedded platform, representing a 7.0 percentage point improvement over the baseline YOLOv8s. The research conclusion provides high-precision technical support for damage rate evaluation and mechanized harvesting efficiency enhancement in beet harvesting.
Keywords：Deep LearningObject RecognitionBeetAttention MechanismYOLOv8s
Publication Date：2025-06-25
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：4( 45-48 )