江苏农业学报2025，Vol.41Issue(5)：905-915,11.DOI:10.3969/j.issn.1000-4440.2025.05.009
基于改进的LSN-YOLOv8模型和无人机遥感图像的水稻稻曲病检测方法
A method for rice false smut detection based on improved LSN-YOLOv8 model and unmanned aerial vehicle remote sensing images
摘要
Abstract
To address the challenges of complex backgrounds,small lesion targets,and the similarity between lesion targets and background features in rice false smut images collected by unmanned aerial vehicles(UAVs),we proposed the LSN-YOLOv8 detection model.The model was based on the YOLOv8 framework,and the large selective kernel network(LSKNet)was incorporated into the backbone network.By dynamically adjusting the receptive field range,the model en-hanced its ability to extract features of small targets.Additionally,a coordinate attention mechanism(CA)module was inte-grated into the backbone network to combine the spatial location information of lesions with channel attention,thereby enhancing the model's focus on key regions while reducing background interference.The detection process was visualized and analyzed using the gradient-weighted class activation mapping(Grad-CAM)technique,thereby providing intuitive explanations for the model's decision-making.To verify the model's performance,rice false smut images captured by UAVs at different disease stages and under various background conditions were used to construct a rice false smut dataset.This dataset was utilized for model training and testing.The experimental results indicated that the precision,recall,and mean average precision at an intersection over union threshold of 0.50(mAP50)of the LSN-YOLOv8 model pro-posed in this study were 94.8%,87.3%,and 92.3%,respectively.These indices were all higher than those of classic ob-ject detection models such as YOLOv5,YOLOv7,YOLOv8 and Faster R-CNN.The visualization analysis results using Grad-CAM technology indicated that the LSN-YOLOv8 model was capable of more accurately focusing on the diseased re-gions in the images.The LSN-YOLOv8 model proposed in this study can provide technical support for the monitoring of rice false smut,disease control and prevention,and the identification of rice disease resistance.关键词
稻曲病/病害识别/无人机/YOLOv8模型/大选择性核网络(LSKNet)/坐标注意力机制(CA)Key words
rice false smut/disease identification/unmanned aerial vehicle/YOLOv8 model/large selective ker-nel network(LSKNet)/coordinate attention mechanism(CA)分类
农业科技引用本文复制引用
杨玉青,朱德泉,刘凯旋,严从宽,孟凡凯,唐七星,廖娟..基于改进的LSN-YOLOv8模型和无人机遥感图像的水稻稻曲病检测方法[J].江苏农业学报,2025,41(5):905-915,11.基金项目
国家重点研发计划项目(2022YFD2001801-3) （2022YFD2001801-3）
国家自然科学基金项目(32201665) （32201665）
