Design And Implementation Of Intelligent Acquisition And Lesion Detection System For Rice Steam Base On Disease Image

Rice pests and diseases are one of the main factors affecting rice production.Identifying pests and diseases of rice accurately as well as timely,which is the premise for pest detection and comprehensive prevention.Traditional field investigation of rice diseases and insect pests often relies agricultural technicians to estimate the situation of field diseases and insect pests by visual inspection in the paddy fields.There are many problems such as high labor intensity,strong subjectivity,low efficiency and hysteresis,which are difficult to meet the development needs of modern agricultural production.With the rapid development of image processing and communication technology,intelligent detection of agricultural pests and diseases based on image has become a research focus.In this paper,we use socket communication,video coding and deep learning object detection technology to design and establish the system of acquisition for image of diseases at the base of rice plants and detect lesion of sheath blight disease.The main research contents and results of this paper are included:(1)Design and establish an intelligent image acquisition equipment for image of diseases at the base of rice plants,which includes a development board for Android camera lenses,a mobile phone and a retractable pole.Android camera module: using Android camera2 application programming interface to capture real-time video;using Media Codec application programming interface to encode the captured real-time video frame data by H.264 hardware;the original image is transmitted by RTSP/RTP streaming protocol;using Java Socket programming and self-defined communication data structure to control command communication between Android camera and mobile phone.The mobile phone control module:using the VLC open source library to realize an RTSP session with the Android camera,and orderly accepts the RTP data packet and displays the real-time video for the base of rice plants;similarly,using Java Socket programming to realize the transmission of camera control command and accept the result of command;using Okhttp processing request framework to achieve upload the disease image to the cloud server,and then accept the processing result of deep learning object detection algorithm.(2)Study the deep learning-based detection algorithm for rice sheath blight.Design the SSD detection algorithm based on residual network for rice sheath blight lesions.In order to improve the ability of detecting network extraction features,replace the basic network VGG16 of SSD detection algorithm with Res Net50 network;the detection system has stronger fault tolerance and robustness to lesion detection,and enhances the data for image of the rice sheath blight;in order to reduce training time,the model of Image Net data is transplanted to rice sheath blight based on transfer learning method.The results show that the detection accuracy of the SSD detection algorithm is 81.2% based on residual network.In this paper,the system of image acquisition for rice agricultural disease at the base of rice plants can be used of capture the image of rice disease convenient and realize the rapid detection of lesion for rice sheath blight disease base on image.The agricultural technician previews the real-time video of the front-end Android camera through the mobile phone,controls the image of the base of the rice stem,and returns the image of the base of the rice plants to the mobile phone.At the same time,the image of the base of the rice plants can be selected and uploaded to the cloud server.The cloud server runs the lesion detection algorithm,and feedback results to the mobile phone.This program does not need a specific model of Android camera,and it is easy to operate and can be widely used in paddy field investigation and monitoring of rice pests and diseases.