Research On Identification Method Of Wheat Leaf Diseases And Development Of Smartphone Diagnosis System

Plant diseases and insect pests are the important factors affecting the yield and quality of crops, and how to monitor them real-time and distinguish them fast and accurately is important for the management of crops production. Traditional monitoring and differentiation method was finished through the sample survey and artificial distinguish by the plant protection experts, which was time consuming and laborious, and it is difficult to meet the needs of large area survey. With the rapid development of science and technology, image processing technology, pattern recognition technique, remote sensing technique, etc. were used to monitor and recognize the plant diseases and insect pests, and achieved remarkable results. However, the existing technology and sensor were still short of practical, low cost and portable. In this paper, the wheat leaf diseases (stripe rust, powdery mildews) were treated as the research objects, the technology of image processing and pattern recognition were combined to explore the method of fast recognition of wheat diseases for instrument development, and developed a diagnosed system based on Android smartphone. The main research contents, innovations and results were as follow:1. In order to reduce the influence caused by the image acquisition environment, we study the image enhancement algorithm, including high pass filtering, median filtering and neighborhood averaging method. Three kinds of image segmentation algorithm, including watershed segmentation, automatic threshold segmentation and level set segmentation, were used to separate the lesion from the wheat leaves for the analysis of the lesion characteristics. And the lesion characteristics from color, texture and shape, a total of 23 features, were described. The results found that, single image enhancement algorithm cannot achieve ideal effect, and a single image segmentation algorithm was not good enough to separate the target area, so they should be optimized to improve the effect of image enhancement and segmentation.2. Relevance vector machine (RVM), support vector machine (SVM) and back propagation neural network (BPNN) were studied for finding a better method to recognize wheat diseases. Total 150 samples in different severities were selected as the study objects, which focused on the mild-to-moderate diseases.68 samples were selected as the training samples, and twenty larger weight features, which were in color, texture and shape features, were selected by algorithm Relief to generate SVM, RVM and BPNN models. Finally, we validated them with another 68 testing samples of two testing group. The results told that the overall accuracy of SVM, BPNN and RVM were 86.76%,91.17% and 89.71%, respectively. While the recognition accuracies of SVM, RVM and BPNN models for mild-to-moderate disease were 86.67&、90.00% and 88.83%, respectively. And the prediction time of RVM was less than those of SVM and BP neural network, with differences as large as 7.96 and 31.68 times, respectively.3. A system for diagnose of wheat leaf diseases based on Android smartphone was developed. A diseases recognition method was combined with android smartphone, the Sony DSC-H9 and SAMSUNG GT-N7100 mobile phone camera were used to collect different severity of stripe rust and powdery samples 66, respectively, which include 33 yellow rust and 33 powdery mildews. Where 48 samples (24 yellow rust and 24 powdery mildews) treated as the training samples and others were treated as testing samples. And changed the pixel of the samples collected by SAMSUNG GT-N7100 mobile phone as a control group to study the relationship between the pixel and the recognition rate. The results showed that, the average recognition accuracy of RVM was 88.89%, and the correct recognition rate is proportional to the pixel of the acquisition tool. Finally, the application test discovered that the system could identify the diseases fast and accurately, and it was finished in 50s, so it could provide important technical support for the plant protection personnel in field real-time investigation.