Hyperspectral Image Classification for Vegetation Detection Using Lightweight Cascaded Deep Convolutional Neural Network

Deep learning frameworks have been applied to hyperspectral image classification (HIC) problems more and more in recent years, with excellent results. Existing network models, on the other hand, have a higher model complexity and take more time. The link between local spatial features is often overlooked in traditional HIC algorithms. This article presents a novel Lightweight Cascaded Deep Convolutional Neural Network (LC-DCNN) that describes the spatial as well as spectral characteristics of the hyperspectral images. The performance of the proposed LC-DCNN is validated for different spectral band reductions techniques to minimize the computational complexity of HIC such as Principle Component Analysis (PCA) and Linear Discriminant Analysis (LDA). The effectiveness of the proposed algorithm is validated on the Indian Pines and Salinas dataset for the vegetation and agriculture detection field based on accuracy, recall, precision and F1-score. The proposed approach provides 99% and 99.63% accuracy over Indian Pines and Salinas datasets, respectively, over the traditional state of arts employed previously for the HIC. The performance of the suggested LC-DCNN decreases for a larger number of classes. In the future, the performance can be improved for the real time dataset by considering a larger dataset consisting of larger objects.