Investigation of classical segmentation's impact on paddy disease classification performance

Abstract

The key source of information for disease diagnosis and classification in paddy diseases is the leaves. Applying hybrid techniques, such as image processing-pattern recognition (IP-PR) and computer vision-based technologies, is the answer to assessing the health of plants. The following paddy diseases are considered in this paper: bacterial leaf blight (BLB), brown spot (BS), leaf smut (LS), and narrow brown spot (NBS) from the machine learning repository. A classical colour threshold-based segmentation method is implemented newly to separate the patterns of image pixels into the diseased part and the normal part. The human visual impression (VI), a subjective method, and a parametric-based method with an average error rate (ER) and overlap rate (OR) are used to assess the uniqueness of the suggested segmentation technique. Using a multi-class support vector machine (MSVM) classifier, the analysis yielded segmented images using the proposed method with an accuracy of 92% over the existing method with an accuracy of 76.60%. The BLB disease achieved the highest identification accuracy of 91%. Our proposed method evaluates the segmentation performance and achieved consistent accuracy higher than the previous segmentation work.