基于机器视觉和深度学习的稻纵卷叶螟性诱智能监测系统

【目的】为减轻基层测报人员工作量，提高稻纵卷叶螟Cnaphalocrocis medinalis性诱测报的准确率和实时性，实现监测数据可追溯，建立了基于机器视觉的稻纵卷叶螟性诱智能监测系统。【方法】稻纵卷叶螟性诱智能监测系统包括基于机器视觉的智能性诱捕器、基于深度学习的稻纵卷叶螟检测模型、系统Web前端和服务器端。利用工业相机、光源和Android平板搭建了智能性诱捕器的机器视觉系统；建立了基于改进的YOLOv3和DBTNet-101双层网络的稻纵卷叶螟检测模型；利用HTML, CSS, JavaScript和Vue搭建系统Web前端展示稻纵卷叶螟检测与计数结果；使用Django框架搭建服务器端，对来自智能性诱捕器通过4G网络上传的图像进行接收与结果反馈；采用MySQL数据库保存图像和模型检测结果等信息。【结果】基于机器视觉的稻纵卷叶螟性诱智能监测系统利用智能性诱捕器自动定期上传稻纵卷叶螟图像至服务器，部署在服务器上的目标检测模型对稻纵卷叶螟成虫进行实时自动检测，精确率和召回率分别达97.6%和98.6%；用户可通过Web前端查看稻纵卷叶螟检测结果图。【结论】基于机器视觉的稻纵卷叶螟性诱智能监测系统实现了图像的定时自动采集、稻纵卷叶螟成虫的准确检测与计数，实现了稻纵卷叶螟性诱监测的智能化和实时性，减轻了测报人员的工作量，监测数据可追溯。

Pheromone baited intelligent monitoring system ofCnaphalocrocis medinalis(Lepidoptera:Pyralidae) based on machine vision and deep learning

【Aim】 In order to reduce the workload of forecasting technicians, improve theprecision and the real-time of Cnaphalocrocis medinalis forecasting and realize thetraceability of monitoring pest data, a pheromone-baited intelligent monitoring system ofC. medinalis based on machine vision was established. 【Methods】 The pheromone-baitedintelligent monitoring system of C. medinalis includes an intelligent pest trap based onmachine vision, a detection model of C. medinalis based on deep learning, a system Webfront-end and a server. Several devices including an industrial camera, a light source andan Android pad were integrated into the machine vision system of pheromone-basedintelligent trap. A two-layer network detection model based on improved YOLOv3 and DBTNet-101 was developed. HTML, CSS, JavaScript and Vue were adopted to build the Web front-endfor displaying the results of detecting and counting the pests in the trap. Djangoframework was used to build a server to receive the images from the intelligent trapsuploaded through 4G network and provide feedback. MySQL database was used to store theimages, model detection results and other information.【Results】 The pheromone-baitedintelligent monitoring system of C. medinalis based on machine vision used the intelligenttrap to automatically upload the images of C. medinalis to the server on a regular time.The object detection model deployed on the server performs could automatically detect C.medinalis adults in real time, with the precision rate and recall rate of 97.6% and 98.6%,respectively. Users could check the detection results of C. medinalis images through theWeb front-end. 【Conclusion】 The pheromone-baited intelligent monitoring system ofC. medinalis can automatically capture the images, and accurately detect and count C.medinalis adults. This system can realize the real-time and intelligentized monitoring ofC. medinalis by pheromone baited trap, reduce the workload of forecasting technicians, andtrace back the data easily.