Research On Real-time Detection Of Relative Diseased Area Of Soybean Leaves Based On Deep Learning

Soybean has always been an important economic crop in China,and crop diseases seriously threaten food security.Detecting the relative diseased area of diseased leaves is an ideal index for disease resistance research.However,traditional detection methods require strict image acquisition standards or more manual processing procedures.To achieve rapid,accurate,and automated detection of the relative diseased area of soybean leaves,the main difficulty is the feature recognition and relative diseased area measurement of soybean diseased leaf images under complex background.In response to these problems,this paper proposed an improved YOLOv4 model algorithm based on deep learning.After in-depth training,it can more accurately predict the soybean leaf target and leaf diseased target in the image.A segmentation algorithm combining K-means clustering and Canny operator edge segmentation were proposed to be applied to the segmentation of soybean leaves with complex backgrounds.Experiments show that the segmentation accuracy of this algorithm can reach 99.35%.It solves the problem that the relative diseased area cannot be detected directly on the soybean leaf image with complex background.The relative lesion area calculated by this algorithm is better and more reliable,and the data is more scientific.Developed real-time detection software for relative diseased area of soybean on mobile phone.Based on the NCNN neural network computing architecture as the realization of the system bottom target detection,other functions are implemented based on the Open CV library,written in Java and C++ languages,the software can be distributed across multiple platforms.The mobile phone camera can be used for real-time shooting detection and photo album import images to detect the relative diseased area of soybean leaves and other functions.