A New Convolutional Kernel Classifier for Hyperspectral Image Classification

Multiple Kernel Learning (MKL) algorithms are among the most successful classification methods for hyperspectral data. Nevertheless, these suffer from two main drawbacks of computational complexity and debility to admit end-to-end learning paradigm. This paper proposed a Convolutional Classifier (CKC) remote sensing images address issues. The CKC uses Nystrm approximation method estimate low-rank basis kernels, thus solves issues associated with high dimensionality kernels. deep architecture learn optimal combination kernels task enable learning. CKCs is based on 1D-Convolutional Neural Network (CNN-1D), it kernel dropout prevent overfitting. It first instance deep-kernel in field sensing. was compared several well-known image analysis MKL algorithms, including multi-kernel variant machine optimization (M-DKMO), MKL-average, Simple-MKL, Generalize (GMKL), state-of-the-art models, Vanilla Recurrent (VanillaRNN) CNN-1D classifying four benchmark datasets. experimental results show that consistently outperforms all competitor methods, its runtime lower than algorithm counterparts Moreover, boosts accuracy. source codes available from: https://github.com/MohsenAnsari1373/A-New-Convolutional-Kernel-Classifier-for-Hyperspectral-Image-Classification .