Multi-Dimensional Recurrent Neural Networks Hyperspectral image classification

Classifying hyperspectral images (HSIs) is a top priority for the remote sensing field. Discriminative feature extraction of HSI using conventional ML techniques is difficult because of the abundance of strongly correlated bands and the scarcity of training data. The deep learning approach has recently achieved state-of-the-art results in a number of visual identification tasks and is widely regarded as a potent tool for feature extraction. The use of deep learning in HSI categorization was spurred by these fruitful uses, and the results have been impressive. Joint spectral-spatial feature extraction for HSI classification has piqued the attention of researchers in convolutional neural network (RNN) based classification techniques among other deep learning approaches. There have been claims in the literature that convolution operations on non-informative spectral bands can cause RNN-based joint feature extraction to underutilize spectral information. In order to minimize information loss, a model was built that extracts spectral and spatial characteristics independently. Then, to take use of the semantic information, the extracted features are fused. Using the datasets from Kennedy Space Center (KSC), Indian Pines (IP), University of Pavia (UP), and Salinas (SA), our model has attained 99.17%, 98.70%, 99.18%, and 99.24% overall accuracy, respectively. For HSI’s limited training samples and high dimensional characteristics, researchers have conducted analyses on band selection and training sample quantity, respectively. In comparison to both older and more modern fusion-based approaches, the suggested model has attained state-of-the-art performance.