融合聚类算法与YOLO-v3网络在果蔬种植防虫害中的应用研究

Application of Clustering Algorithm and YOLO-v3 Network in Pest Control of Fruit and Vegetable Planting

采用融合ISODATA聚类算法与YOLO-v3网络构建果蔬虫害识别模型,利用预选框提取方法分辨栅格害虫目标个数,并加入空间金字塔池化结构,以提高图像特征提取的稳定性。在害虫种类识别的测试中,优化的YOLO-v3网络mAP为88.92%,比Faster-RCNN高3.7个百分点。而在果蔬图像背景测试中,优化的YOLO-v3网络mAP为87.32%,比传统YOLO-v3模型高4.4个百分点。试验表明:优化的YOLO-v3网络对于图像噪声抗干扰性更强,检测精度高的同时保持了稳定的检测效率。

The fruit and vegetable pest identification model was constructed by integrating ISODATA clustering algorithm and YOLO-v3 network.The pre selection box extraction method was used to distinguish the number of grid pest targets and add the spatial pyramid pool structure to improve the stability of image feature extraction.In the test of pest species identification,the optimized YOLO-v3 network map was 88.92%,which was 3.7 percentage points higher than Fast-RCNN.In the fruit and vegetable image background test,the optimized YOLO-v3 network map was 87.32%,which was 4.4 percentage points higher than the traditional YOLO-v3 model.Experiments showed that the optimized YOLO-v3 network had stronger resistance to image noise interference,high detection accuracy and stable inspection efficiency.