Effects of segmentation techniques on classification algorithms for plant leaf diseases images

Abstract

With the global rise of the agricultural industry driven by a growing population, meeting the increasing demand for plant-based products has become more challenging. This difficulty arises from the complexity of ensuring that crop yield and quality are not compromised by external factors or diseases. Traditionally, visual inspection has been the primary method for identifying plant diseases. However, even experienced agriculturists and pathologists can struggle with this task due to the vast variety of diseases. This challenge has led to the development of computer vision techniques that enable us to process images and extract meaningful features. These features can then be used to classify plant diseases through machine learning algorithms such as Support Vector Machine (SVM), k-Nearest Neighbors (KNN), and Random Forest (RF), aiding in more accurate and efficient disease diagnosis. One of the key challenges in image classification is the lack of transparency in many processing methods, often referred to as the ”black box” nature of these techniques. Understanding how specific methods transform images and how these changes influence classification accuracy can be difficult. This challenge complicates efforts to optimize algorithms effectively. To address this, we develop and compare two distinct image processing techniques: K-Means-Based Segmentation and Color-Based Segmentation. After segmentation, we perform feature extraction using the Gray Level Co-occurrence Matrix (GLCM) to capture critical texture features. By varying parameters and introducing additional steps in the image processing workflows, we examine how these adjustments impact the classification accuracy. For classification, we employed SVM, KNN, and RF algorithms and compare their performance to assess the effectiveness of the proposed techniques.