Research On Recognition Of Plant Leaf Diseases Based On Computer Vision

The traditional disease identification method has some defects,such as low efficiency,high time-consuming and low recognition rate,which would seriously affect the quality and yield.In order to solve these problems,disease monitoring,identification and diagnosis technology are developed all the time.They have become research hotspots in monitoring of plant diseases in agriculture and have showed wide applications.The main research of this paper is to achieve remote video monitoring and detection,and to take grape leaf diseases.First,the overall system layout was designed and the software and hardware were selected,as well as the embedded Linux application platform was constructed based on the discussion of video monitoring and analysis of the status of machine vision in agriculture.Then,the development process of the video monitoring system was studied,mainly including data acquisition based on the research of V4L2 and USB camera,x264 video codec module and the Socket network programming of video transmission and receiving module.Furthermore,the development of plant diseases and insect pests recognition algorithm based on OpenCV were designed.The effective preprocessing and segmentation methods were chosen based on original image processing algorithms through a large number of tests.In preprocess,median filter was selected for image denoising.Image gray processing used weighted average method.In image segmentation,B component in RGB color space model of color image was selected for OTSU threshold segmentation and mathematical morphology processing to achieve the extraction of the lesions.Feature extraction was extracted 15 characteristic parameters in 3 aspects: color,shape and texture.Classification recognition utilized SVM to test recognition rate on the randomly selected training and test set.Finally,RBF was selected as kernel function of SVM.When the penalty parameter C was 64 and RBF parameter gamma was 1,the average recognition accuracy in 15 features extracted from grape leaves for disease recognition become the highest.Finally,the Qt programming design of the plant leaf disease monitoring and recognition system was completed.The final result of the whole system test is as follows: monitoring system application software can achieve real-time monitoring of plant growth in both embedded client and remote PC server.At the same time,the fluency of video transmission and display resolution were in accordance with the basic requirements;the recognition system application software in embedded client and remote PC servers both can identify grape leaf diseases.