【摘要】 林业生态环境监测建设是林业生态健康可持续发展的迫切需求，是森林资源保护、生态文明建设和林业有害生物防控体系提升的关键。快速、准确、有效地检测林业有害生物能够遏制病虫害蔓延，促进森林病虫害综合治理，减轻对林业生产和生态环境建设的危害。为此提出一种深度学习方法，利用当前强大的目标检测算法YOLOv5来实现林业有害生物的检测与识别，针对害虫图像中经常出现重叠和遮挡物体问题，采用DIoU＿NMS算法对目标框进行选择，增强被遮挡害虫的检测识别准确率。试验结果表明，YOLOv5算法模型能够有效识别数据集中包含的9种林业有害生物，精确度达到了0.973,召回率达到了0.929,均值平均精度(mean Average Precision, mAP)达到了0.942。与YOLOv3和Faster-RCNN相比，mAP比YOLOv3高0.04,比Faster-RCNN高0.087,充分显现出该模型的识别精度高，且实时性好，鲁棒性强。更多还原

【Abstract】 The construction of forestry ecological environment monitoring is an urgent need for the healthy and sustainable development of forestry ecology. It is also the key to the protection of forest resources, the construction of ecological civilization and the improvement of forestry pest control system. Rapid, accurate and effective identification of forest pests can curb the spread of pests and diseases, promote the comprehensive management of forest pests and diseases, and reduce the harm to forestry production and ecological environment construction. In this paper, a deep learning method is proposed. Using the current powerful object detection algorithm YOLOv5 to achieve the detection and identification of forest pests. Overlapping and occluded objects often appear in pest images, so DIoU＿NMS algorithm is used to select the target box to enhance the detection and identification accuracy of sheltered pests. Experimental results show that the proposed model can effectively identify nine forest pests in the dataset, with a precision of 0.973, recall of 0.929 and mean Average Precision(mAP) of 0.942.Compared with YOLOv3 and Faster-RCNN, mAP is 0.04 higher than YOLOv3 and 0.087 higher than Faster-RCNN. It shows that the model has high recognition accuracy, good real-time performance and strong robustness.更多还原