3D-MACNet: A Multiscale Asymetric Convolution Network for Feature Extraction and Classification of Hyperspectral Images

In recent years, deep learning's amazing feature extraction capabilities have led to its widespread adoption in hyperspectral image classification (HSI) applications. However, traditional deep learning methods frequently produce inferior classification results because of the low discriminatory power of the retrieved features. Furthermore, a notable area of interest in research now involves the difficulty of achieving good classification performance in HSIs with a restricted number of samples. This letter offers a unique 3D multiscale asymmetric convolution network (3D-MACNet) for the extraction of features and HSIs classification to address this issue. By adapting the concept of symmetric convolution to the construction of an asymmetric convolution with three identical branches at distinct scales that can retrieve spatial-spectral characteristics from vertical, horizontal, and all perspectives. run on both the new data sets and the freely accessible standard data sets. The provided model has been proven to exceed the other cutting-edge approaches in regard to classification accuracy, according to empirical evidence.