Recognition Method of Cucumber Leaves Diseases Based on Visual Spectrum and Support Vector Machine
Li Xin-xing,Zhu Chen-guang,Bai Xue-bing,Mao Fu-huan,Fu Ze-tian,Zhang Ling-xian
DOI: https://doi.org/10.3964/j.issn.1000-0593(2019)07-2250-07
2019-01-01
Spectroscopy and spectral analysis
Abstract:In this paper, we used cucumber leaves disease as the research object, and identified cucumber leaves disease based on the difference of visible spectral reflectance. The support vector machine recognition is an efficient recognition method, which is always used as identification model. For cucumber leaves diseases, if we constructed the support vector machine based on digital image process, we can get accurate and efficient recognition. Consequently, this paper studied the cucumber leaves disease recognition method based on support vector machine. Firstly, the method of wavelet domain denoising was applied to image denoising. The segmentation results were compared with K mean clustering, OTSU and edge segmentation. The results showed that K-means clustering method was more accurate. We extracted texture, color and shape feature parameters, 15 feature parameters. Then, the optimal parameters of c and g were selected by cross-validation, and the parameters of the kernel function were optimized and using RBF kernel to construct SVM classifier. By comparing the linearity kernel, polynomial kernel and RBF kernel of the SVM recognition's correct rate, we got that the RBF kernel is most accurate for the recognition of the cucumber leaf disease, Therefore, we used RBF kernel to construct SVM classifier. Finally, there was an identification model of cucumber leaf disease which was based on SVM classifier, and two other efficient identification models, back Propagation neural network, fuzzy clustering identification model. We constructed three kinds of identification models through comparing the correct recognition rate and running time. The results of the test showed that the cucumber downy mildew's correct recognition rate based on SVM classifier was 95%. The correct recognition rate of cucumber powdery mildew and brown spot was 90%, and the average diagnosis accuracy was 92%. In addition, the method running time was the shortest. In summary, the results show that, among the three recognition methods, cucumber leaves disease recognition based on the SVM classifier is the most suitable, demonstrating that the method can be used to rapidly identify cucumber leaves diagnosis based on visual spectrum.