A Framework for Guava Wilt Disease Segmentation Using K-Means Clustering and Neural Network Techniques

Abstract

Guava fruit production is influenced by myriad of external and internal and the rapid decline in production is often attributed to various diseases. Among these, ‘Wilt’ disease stands out as a significant contributor to the decrease in guava fruit yield. This paper proposes and implement a framework leveraging advanced technologies, including image processing and machine learning techniques, to detect wilt disease at its early stages. To facilitate this, a database comprising 1420 images of guava plant leaves affected by wilt disease is created. The database is further categorized into three datasets based on the level of noise present in the images: fully noisy images, partially noisy images and noise-free-images. The segmentation of guava wilt disease is achieved through the application of K-means clustering sand Convolutional Neural Network techniques on the guava images. Remarkably, the proposed framework demonstrates a cumulative accuracy of 93.82% and 95.17% using K-means clustering and neural network algorithms, respectively. This innovative approach hold s promise for effectively managing and increasing guava fruit production by identifying and addressing wilt disease in its early stages.