Land Cover Classification from Hyperspectral Images Using Regularized Hybrid CNN and ADAM

Abstract

The utilization of hyperspectral imagery offers enhanced detail and accuracy for environmental monitoring and natural resource management, particularly through land cover classification. Hyperspectral data capture spectral signatures across numerous wavelengths, allowing precise differentiation of various surface materials and land types. While numerous approaches have been proposed for hyperspectral image classification, many suffer from overly complex model structures and suboptimal performance, limiting their practical application. This study introduces a simplified yet effective architecture by implementing a Regularized Hybrid Convolutional Neural Network (CNN) optimized using Adaptive Moment Estimation (ADAM). The proposed model is evaluated on the widely used Pavia Center hyperspectral dataset to assess its performance in land cover classification tasks. The model achieves a notable Overall Accuracy of 99.25% and Average Accuracy of 97.50%, demonstrating its capability in handling high-dimensional hyperspectral data with reduced model complexity. Additionally, a comparative analysis with conventional CNN architectures is conducted, highlighting the superior performance and efficiency of the proposed approach. These findings underscore the potential of regularized hybrid CNNs as a reliable and scalable solution for hyperspectral image classification, especially in applications requiring high precision and reduced computational overhead.