基于改进YOLOv8的水稻病虫害检测方法研究：融合SCSA注意力机制与WiOUv3损失函数

Improved YOLOv8 for Rice Pest and Disease Detection with SCSA Attention and WiOUv3 Loss

摘要: 传统的水稻病虫害检测方法通常存在小目标识别精度低、背景干扰强以及特征提取能力有限的问题。为解决这些挑战，本研究提出了一种基于YOLOv8 (You Only Look Once version 8)的改进水稻病虫害检测框架。该方法引入SCSA (Spatial-Channel-Spatial Attention)注意力机制，用来增强特征表达能力。本研究采用WiOUv3 (Wise Intersection over Union Version 3)替代传统IoU损失。该损失函数综合考虑预测框与真实框之间的重叠区域和中心距离，引入动态梯度权重机制。本研究在公开水稻病虫害数据集上进行评估，性能指标包括mAP@0.5、mAP@0.5:0.95、Precision和Recall。结果表明，YOLOv8-SCSA-WiOU在mAP@0.5指标上达到0.834，mAP@0.5:0.95为0.689，相比基线YOLOv8n的0.800和0.655有显著提升，Precision和Recall同样有所提高。

Abstract: Traditional rice pest and disease detection methods often suffer from low accuracy in detecting small targets, strong background interference, and limited feature extraction capabilities. To address these challenges, this study proposes an improved rice pest and disease detection framework based on YOLOv8 (You Only Look Once version 8). The proposed method incorporates the SCSA (Spatial-Channel-Spatial Attention) attention mechanism to enhance feature representation. In addition, WiOUv3 (Wise Intersection over Union Version 3) is employed to replace the conventional IoU loss function. The proposed loss function jointly considers the overlap region and center distance between the predicted and ground-truth bounding boxes and introduces a dynamic gradient weighting mechanism. The proposed method is evaluated on a publicly available rice pest and disease dataset using mAP@0.5, mAP@0.5:0.95, Precision, and Recall as performance metrics. Experimental results show that YOLOv8-SCSA-WiOU achieves an mAP@0.5 of 0.834 and an mAP@0.5:0.95 of 0.689, which are significantly higher than 0.800 and 0.655, respectively, for the baseline YOLOv8n. Precision and Recall are also improved.