Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{208147,
        author = {Sanket Ramesh Dhumal and Anil R. Karwankar and Vijay G. Nemane},
        title = {RGB–NDVI Channel Fusion for Citrus Leaf Disease Classification Using a Four-Channel ResNet50 and Sentinel-2 Satellite Data},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {4},
        pages = {560-569},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=208147},
        abstract = {Fungal and bacterial diseases and nutrient deficiency reduce citrus production in India by 30–40% annually. Early and precise identification is crucial for sustainable crop management, but traditional visual diagnosis is subjective and labour-intensive at scale. This study proposes a channel-level early fusion approach combining RGB leaf images with a spatially resolved Normalized Difference Vegetation Index (NDVI) map derived from Sentinel-2 Level-2A satellite data as a fourth input channel to a modified ResNet50 convolutional neural network. The 224×224 NDVI map was computed from Sentinel-2 tile T44QKJ (2023-02-12) covering the Nagpur citrus belt, providing region-specific spectral vegetation context. The model was evaluated on a publicly available citrus leaf benchmark dataset (38,432 annotated images, five classes), with both the RGB-only baseline and the proposed RGB+NDVI fusion model trained and tested across four random seeds to assess stability. The fusion model achieved 100.00 ± 0.00% weighted test accuracy across all seeds, compared to 99.99 ± 0.02% for the baseline. In the one seed where the baseline erred, the fusion model corrected all errors, and in no seed did the fusion model underperform; however, McNemar's exact test did not reach statistical significance (p = 0.25), reflecting near-ceiling performance on this benchmark. These results indicate that the satellite-derived NDVI channel provides complementary spectral context that never degrades classification, and that the proposed open-data pipeline is reproducible and suitable for low-resource precision agriculture deployment.},
        keywords = {Channel fusion, Citrus disease detection, Convolutional neural network, Deep learning, NDVI, Precision agriculture, Remote sensing, ResNet50, Sentinel-2.},
        month = {September},
        }