Study on identification of Rice False Smut based on CNN in natural environment

摘要

The widely distributed Rice False Smut(RFS) is one of the most harmful viruses for rice. However, it's identification methods are based on subjective judgment. In this paper, two methods are put forward to identify the RSF under natural conditions. One is the traditional ML(machine learning) classification method SVM (Support Vector Machine) combined with the feature extraction method HOG (Histogram of Oriented Gradient), the second is to build a new CNN(Convolutional Neural Networks) architecture. For the SVM-HOG identification method, firstly, the input images are pre-processed by the image processing algorithms, and a method of dividing rice panicle is proposed. Secondly, extracting the HOG features from the pre-processed images. Finally, the features are classified by SVM. For the CNN identification method, it has built a new CNN architecture, compared with the classic AlexNet and VGGNet-11, and analyze its advantages. The purpose of using SVM is to highlight the advantages of CNN through comparison. The results of experiment show that the new CNN is highly accurate and effective in the identification of RFS.