【摘要】 针对传统粮库监控效率不高、检测结果不准确的问题,提出了一种基于深度学习的粮库虫害实时监测预警系统(RMPS).该系统对粮库中常见几种害虫(米象、绣赤扁谷盗、赤拟谷盗)进行较大规模的样本采集,并使用卷积神经网络进行学习和训练,构建神经网络模型;通过新型采集器实时采集粮库内部图像信息,利用已训练的模型进行害虫种类和概率的检测;并将监测结果以web形式发布给手机客户端.在实验室条件下,搭建小型模拟粮仓,部署本系统进行试验.测试结果表明:RMPS从传统的定点定时监测转变成实时监测,并且将虫害检测准确率提升到90%;RMPS采用专门设计的采集器与移动客户端部署简单、方便,具有较高的实用性与扩展性.更多还原

【Abstract】 To solve the problems of low efficiency and inaccurate detection results of traditional grain storage monitoring, a real-time monitoring and prewarning system(RMPS) was proposed based on deep learning. The large-scale samples of several pests of sitophilus oryzae linne, cryptolestes ferrugineus and tribolium castaneum in grain depot were collected, and the samples were learned and trained by the CNN to construct a neural network model. The internal image information of grain store was collected by the new collector in real time to detect the pest species and probability. The monitoring results were published to mobile phone in the form of web. A small simulated granary with RMPS was built under laboratory conditions. The simulation results show that the RMPS can change the monitoring mode from the traditional fixed-point timing monitoring to the real-time monitoring with improved pest detection accuracy rate over 90%. The specially designed collector and the mobile client can make the deployment of RMPS more simple and convenient, which achieves high practicability and scalability.更多还原