@article {10.3844/jcssp.2025.2065.2073,
article_type = {journal},
title = {Early Plant Disease Detection Using Graph Isomorphic Networks: Enhancing Crop Yield Through Leaf Analysis},
author = {Sumathi, D. and Ankam, Sreejyothsna and Adivarekar, Pravin Prakash and Sandeep, Kasturi Sai and R., Gomathi and Shobarani, R. and Iswarya, S. Karpaga and Bhoopathy, V.},
volume = {21},
number = {9},
year = {2025},
month = {Oct},
pages = {2065-2073},
doi = {10.3844/jcssp.2025.2065.2073},
url = {https://thescipub.com/abstract/jcssp.2025.2065.2073},
abstract = {The economy of Tanzania is mostly driven by agriculture. Disease is one of the reasons that contributes to the low production of staple foods like cassava and maize, alongside climate change. Loss of income and food security are the results. In order to detect the diseases early, preventative measures are required. A potential option for farmers could be the use of image processing tools to identify plant diseases on leaves. Implementing the existing method of disease detection, which involves an expert using their naked eyes, on a large farm is a laborious and time consuming process. This study provides a comprehensive overview of recent research in image processing by reviewing methods for identifying plant diseases in their leaves or fruits and the corresponding machine learning models for disease classification. This study examines issues in the identification of plant diseases, pertinent to agriculture-dependent nations like Tanzania and India. Presenting the present state of the art, elucidating the steps done during the image processing stage, and assessing the pros and cons of each technique as well as the effectiveness of the machine learning model used for disease classification are the primary goals of the work. Among the preprocessing and resampling techniques, the evaluation's results show that GIN-based approach for resampling, in conjunction with contrast limited adaptive histogram equalization (CLAHE), achieved the best results, with an average F1-score of 95.65% and a classification accuracy of 95.62%. The study concludes with a generic process for a disease detection system, which may be broken down into individual components as needed.},
journal = {Journal of Computer Science},
publisher = {Science Publications}
}