Novel 3-D Deep Neural Network Architecture for Crop Classification Using Remote Sensing-Based Hyperspectral Images

Abstract

This research establishes a new 3-D Deep Neural Network (3D-DNN) framework that uses spatial, spectral and temporal characteristics for accurate hyperspectral image-based crop classification. The proposed model surpasses conventional 2D CNNs at processing remote sensing data, achieving 98.5% accuracy on Indian Pines and 99.2% on Salinas, which exceeds current methods by 3-5%. Model performance is optimized through advanced preprocessing methods that apply both dimensionality reduction and data augmentation strategies. The results show that the proposed model successfully recognizes different crop types across multiple agricultural areas, which proves its capability for practical precision agriculture systems. Future studies will focus on developing real-time systems and merging different data types while creating minimalistic system designs for deploying deep learning technology in resource-constrained agricultural contexts to support sustainable farming operations.