Sun Guojia, Liu Shuhua, Luo Haolun, Feng Zelin, Yang Baojun, Luo Ju, Tang Jian, Yao Qing, Xu Jiajun
School of Information Science and Technology, Zhejiang Sci-Tech University, Hangzhou, China.
State Key Laboratory of Rice Biology, China National Rice Research Institute, Hangzhou, China.
Front Plant Sci. 2022 Jun 20;13:897739. doi: 10.3389/fpls.2022.897739. eCollection 2022.
Three species of rice migratory pests (, and ) cause severe yield and economic losses to rice food every year. It is important that these pests are timely and accurately monitored for controlling them and ensuring food security. Insect radar is effective monitoring equipment for migratory pests flying at high altitude. But insect radar is costly and has not been widely used in fields. Searchlight trap is an economical device, which uses light to trap migratory pests at high altitude. But the trapped pests need to be manually identified and counted from a large number of non-target insects, which is inefficient and labor-intensive. In order to replace manual identification of migratory pests, we develop an intelligent monitoring system of migratory pests based on searchlight trap and machine vision. This system includes a searchlight trap based on machine vision, an automatic identification model of migratory pests, a Web client, and a cloud server. The searchlight trap attracts the high-altitude migratory insects through lights at night and kills them with the infrared heater. All trapped insects are dispersed through a multiple layers of insect conveyor belts and a revolving brush. The machine vision module collects the dispersed insect images and sends them to the cloud server through 4G network. The improved model YOLO-MPNet based on YOLOv4 and SENet channel attention mechanism is proposed to detect three species of migratory pests in the images. The results show that the model effectively improves the detection effect of three migratory pests. The precision is 94.14% for , 85.82% for , and 88.79% for . The recall is 91.99% for , 82.47% for , and 85.00% for . Compared with some state-of-the-art models (Faster R-CNN, YOLOv3, and YOLOv5), our model shows a low false detection and missing detection rates. The intelligent monitoring system can real-timely and automatically monitor three migratory pests instead of manually pest identification and count, which can reduce the technician workload. The trapped pest images and historical data can be visualized and traced, which provides reliable evidence for forecasting and controlling migratory pests.
三种水稻迁飞性害虫（[害虫名称1]、[害虫名称2]和[害虫名称3]）每年都会给水稻生产造成严重的产量损失和经济损失。及时、准确地监测这些害虫对于控制它们并确保粮食安全至关重要。昆虫雷达是监测高空迁飞害虫的有效设备。但昆虫雷达成本高昂，尚未在田间广泛应用。诱虫灯是一种经济的装置，利用灯光在高空诱捕迁飞害虫。但需要从大量非目标昆虫中人工识别和计数捕获的害虫，效率低下且劳动强度大。为了取代对迁飞害虫的人工识别，我们基于诱虫灯和机器视觉开发了一种迁飞害虫智能监测系统。该系统包括基于机器视觉的诱虫灯、迁飞害虫自动识别模型、Web客户端和云服务器。诱虫灯在夜间通过灯光吸引高空迁飞昆虫，并用红外加热器将其杀死。所有捕获的昆虫通过多层昆虫输送带和旋转刷进行分散。机器视觉模块收集分散的昆虫图像，并通过4G网络将其发送到云服务器。提出了基于YOLOv4和SENet通道注意力机制的改进模型YOLO-MPNet，用于检测图像中的三种迁飞害虫。结果表明，该模型有效地提高了三种迁飞害虫的检测效果。[害虫名称1]的精度为94.14%，[害虫名称2]的精度为85.82%，[害虫名称3]的精度为88.79%。[害虫名称1]的召回率为91.99%，[害虫名称2]的召回率为82.47%，[害虫名称3]的召回率为85.00%。与一些先进模型（Faster R-CNN、YOLOv3和YOLOv5）相比，我们的模型显示出较低的误检率和漏检率。该智能监测系统可以实时自动监测三种迁飞害虫，而无需人工进行害虫识别和计数，这可以减轻技术人员的工作量。捕获的害虫图像和历史数据可以可视化和追溯，为迁飞害虫的预测和控制提供可靠依据。
