Work

YOLO-LSD: A lightweight scale dynamic model for tiny pest detection in tea plantations

Year: 2025

Type: article

Abstract: • This study effectively solved the problem of identification and monitoring of tea green leafhoppers in tea tree. • The improved YOLOv8n exhibits excellent performance in the detection of tea green leafhoppers. • The YOLO-LSD based model has achieved impressive lightweight and accuracy. • The proposed method can accurately monitor the other tiny pests. The tea green leafhopper ( Empoasca pirisuga Matumura), a tiny pest of tea plantations, is a small target in the board's trapping image, while the complex environment of the tea plantation and the changing light conditions easily interfere with the model's judgment. This study presents YOLO-LSD, a lightweight scale dynamic feature detection model based on an improved YOLOv8 algorithm. Firstly, a Scale Dynamic Attention Module (SDAM) is incorporated into the backbone network. By calculating the standard deviation of the input feature map to measure its complexity, SDAM dynamically assigns weights to different convolutional kernel branches. This mechanism focuses on the key features and suppresses redundant information generated by lighting changes, thereby improving detection accuracy. Secondly, to improve the capture detail of tea green leafhopper, the detection head is optimized with a P2 head replacing the P5 head and integrating the SEAM attention mechanism. Finally, the Wavelet Pool module substitutes some convolution and sampling operations, reducing aliasing, preserving high-frequency details, and decreasing the computational cost. The experimental results demonstrate that the enhanced model attains mAP, F1, Precision, and Recall scores of 94.3%, 92.1%, 91.9%, and 92.4%, respectively. When evaluated in comparison with the original YOLOv8n model, these represent enhancements of 2.2%, 1.9%, 0.2%, and 3.7%. The model parameters and FLOPs were reduced by 1.32 million and 0.1 gigaflops, respectively, resulting in a model size of only 3.8 MB. This research provides a valuable reference for the precise identification of pests in tea plantations.

Source: Smart Agricultural Technology

DOI: 10.1016/j.atech.2025.101474