Impact of Vegetation Indices on Crop Stage Classification using Deep Learning

This research work focuses on computing different vegetation indices and on image fusion techniques to support deep learning-based vegetation analysis. The proposed approach computes multiple vegetation indices, including the Normalized Difference Vegetation Index (NDVI) and Green Normalized Difference Vegetation Index (GNDVI), from multispectral Sentinel-2 imagery to capture essential spectral characteristics of vegetation. Two image fusion configurations, RGNDVI and RGGNDVI, are generated by replacing the blue channel of the RGB imagery with vegetation index information, producing three-channel composites suitable for deep learning models while preserving the spatial characteristics of the original imagery. The generated fused datasets are subsequently used for training deep learning models for vegetation-related tasks. Experimental results demonstrate that the proposed RGNDVI fusion significantly outperforms both the RGGNDVI fusion and conventional RGB inputs, achieving a classification accuracy of 82%, compared with 40% and 42%, respectively, together with a macro Precision of 0.84, Recall of 0.86, and F1-score of 0.84. The proposed pipeline provides a structured, scalable, and efficient method for transforming raw satellite data into enhanced feature representations for vegetation monitoring, crop analysis, and AI-driven remote sensing applications. The findings highlight the effectiveness of integrating vegetation indices with RGB imagery to improve feature representation, classification reliability, and the overall suitability of satellite data for AI-based crop monitoring