PERFORMANCE COMPARISON OF UNSUPERVISED SEGMENTATION ALGORITHMS ON RICE, GROUNDNUT, AND APPLE PLANT LEAF IMAGES

Abstract

This paper focuses on plant leaf image segmentation by considering the aspects of various unsupervised segmentation techniques for automatic plant leaf disease detection. The segmented plant leaves are crucial in the process of automatic disease detection, quantification, and classification of plant diseases. Accurate and efficient assessment of plant diseases is required to avoid economic, social, and ecological losses. This may not be easy to achieve in practice due to multiple factors. It is challenging to segment out the affected area from the images of complex background. Thus, a robust semantic segmentation for automatic recognition and analysis of plant leaf disease detection is extremely demanded in the area of precision agriculture. This breakthrough is expected to lead towards the demand for an accurate and reliable technique for plant leaf segmentation. We propose a hybrid variant that incorporates Graph Cut (GC) and Multi-Level Otsu (MOTSU) in this paper. We compare the segmentation performance implemented on rice, groundnut, and apple plant leaf images for various unsupervised segmentation algorithms. Boundary Displacement error (BDe), Global Consistency error (GCe), Variation of Information (VoI), and Probability Rand index (PRi), are the index metrics used to evaluate the performance of the proposed model. By comparing the outcomes of the simulation, it demonstrates that our proposed technique, Graph Cut based Multi-level Otsu (GCMO), provides better segmentation results as compared to other existing unsupervised algorithms.