Automated Detection of Crop Diseases Using K-Means Clustering and Image Segmentation

Abstract

In this paper, we propose a fully automated and efficient method for detecting diseases in crop leaves using modern image processing techniques. The health of plant leaves plays a vital role in determining the overall condition of a crop, and early identification of any disease can prevent large-scale damage. Traditionally, disease identification is done manually by experts or farmers, which is not only time-consuming but also prone to human error due to fatigue or limited experience. To overcome these challenges, we have developed a computerized system that can automatically analyse images of crop leaves and detect signs of disease. Our approach is based on an improved version of the K-Means clustering algorithm, which is commonly used in image segmentation tasks. This algorithm groups pixels in the image based on their colour similarity, allowing it to separate the healthy parts of the leaf from the potentially infected areas. To make the detection more accurate, we have combined K-Means clustering with Otsu’s thresholding method, which automatically finds the best value to separate important areas in the image. In addition, we apply a colour space transformation to convert images from RGB (Red, Green, Blue) format to HSI (Hue, Saturation, Intensity), which makes it easier to identify subtle differences in colour and texture that are often caused by disease. The result of this combined method is a clear and precise identification of the infected regions on the leaf surface. Our system removes unnecessary background and green (healthy) areas to focus only on the sections that matter for disease detection. This greatly reduces the need for human involvement and minimizes the chances of subjective errors. Through extensive experiments on different types of diseased leaf images, we observed high accuracy in detecting and separating diseased parts from healthy ones. The system performed well in identifying even the small and earlystage infections, which are usually hard to notice with the naked eye. These positive results show that the proposed method is not only technically sound but also very practical for use in smart agriculture, where digital tools are used to help farmers monitor their crops more effectively and make better decisions in managing plant health.