A Rapid Recognition Method for Rice False Smut based on HOG Features and SVM Classification. (June 2020)

• Record Type: Journal Article
• Authors: S, Naila; JJ, Yu; N, Yang; H, Kashif; J, Tang; AY, Wang
• Is Part Of: Journal of physics. Volume 1576(2020)
• Publication Date: 2020-06
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1576/1/012018
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: A rapid detection method for the detection of false smut is introduced based on HOG features and SVM classification to overcome the drawbacks of false smut detection like time wasting and inaccurate recognition. In this paper, original images of false smut used, which directly taken from the field of rice. First of all, the image false smut diseased rice acquired by image preprocessing was based on the color threshold to obtain the (ROI) region of interest from healthy and unhealthy pictures of rice. Secondly, the gamma-correction performed on the extracted ROI unhealthy and background region of healthy rice images. The HOG features extracted to form a high-dimensional HOG feature vector. Third, the PCA analysis performed on high-dimensional HOG features, and the original 1764-dimensional feature vector reduced to a two-dimensional vector that can represent most vectors to achieve the purpose of dimension reduction. Finally, the dimension-reduced HOG feature vector sent to the SVM classifier for training. The expected recognition accuracy under the binomial kernel function is 80.6%, and the actual test set recognition is 78.56%.