分布式移动农业病虫害图像采集与诊断系统设计与试验

Design and experiment of agricultural diseases and pest image collection and diagnosis system with distributed and mobile device

为了便捷地采集和实时诊断农业病虫害图像,设计了一个分布式移动农业病虫害图像采集与诊断系统。该系统由多个便携式图像采集终端和一个图像处理服务器组成;其中,图像采集终端包括嵌入式相机、可伸缩的手持杆和装载控制App的手机;图像处理服务器包括农业病虫害诊断、信息记录和反馈模块等。手持杆可将安装在其前端的嵌入式相机送到人手或视觉难以企及的病虫害区域,手机可实时预览前端相机的拍摄画面和实现控制相机完成农业病虫害图像采集等功能;系统通过HTTP协议实现多个采集终端与图像处理服务器的数据交互,协同进行分布式计算,可以减少网络移动资费和服务器的负载。利用该系统对水稻纹枯病图像采集与诊断测试结果表明,该系统的图像采集终端可以便捷地采集到水稻纹枯病图像,手机端视频预览画面延时低,对相机控制命令无误,图像采集终端与服务器通信稳定,服务器端对水稻纹枯病图像处理和诊断实时,基于图像的水稻纹枯病为害等级诊断准确率为83.5%。如果服务器端加载不同的农业病虫害图像处理和诊断算法,该系统可广泛应用于各种农业病虫害图像的采集与诊断。

In order to easily collect images of agricultural diseases and pests and make real-time diagnose, a distributed mobile system was designed with a number of portable image collection devices and one image processing server. Each image collection device consisted of an embedded camera, a stretchable handheld pole and an Android phone equipped with an APP of control capability. The embedded camera was fixed on the end of the handheld pole via universal joints. The handheld pole could extend to about 2 m in length. The embedded camera was built upon a development board with iT OP 4412 and a set of modules, including WIFI control, camera control, image collection, H.264/JPEG coding, RTSP/RTP video transmission, GPS information collection and writing, file transfer, and image preprocessing, which were developed in Linux platform. The mobile application was developed in Android platform with a set of modules, including video streaming preview, network, image browsing and camera control. The image processing sever could receive the images from the image collection devices, record GPS information, diagnose agricultural diseases and pests, and return the diagnosis and control information of agricultural diseases and pests to the mobile phone. Among the components of this system, the handheld pole was used to deliver the embedded camera to some unreachable agricultural disease and pest area, and the mobile phone was used for browsing images and controlling camera to collect the disease and pest images. TCP/UDP protocols and SoftA p technique were used for data exchange among the embedded camera and the mobile phone, which could be independent from cable networks and wireless local area networks. HTTP protocols were used for data exchange and distributed computing among the image collection devices and the image processing server, which can reduce the mobile phone charges and the server overhead. To test the distributed mobile agricultural system, a diagnosis algorithm of damage levels of rice sheath blight was deployed to the image processing server. This algorithm mainly included image feature extraction, disease identification, disease area computation and damage level judgment. The images of rice sheath blight were collected using the image collection device in paddy fields located in China National Rice Research Institute in 2016. After the segmentation of disease area was finished in the embedded camera, the segmented images were uploaded to the image processing server. The diagnosis algorithm in the server was implemented to process these images and the diagnosis results and control information were returned to the mobile phone. The technicians or farmers could control the rice sheath blight based on the diagnosis suggestions. Our experiment indicated that the image collection device could easily collect the images of agricultural diseases and pests, especially on some places where hands and sight were hard to reach. The system could work effectively with low image browse latency, accurate camera control, reliable device-to-server communication and real-time image processing and diagnosis. The accurate rate of 83.5% was achieved to diagnose the damage levels of rice sheath blight based on our algorithm. Therefore, the system is expected to be widely applicable to agricultural disease and pest image collection and diagnosis.