A method based on stacked contractive autoencoder for hyperspectral image classification

Aiming at the problem of low accuracy, poor model stability, time consumption of deep learning model and high computational cost of traditional hyperspectral image classification methods, fully considers the spectral information and spatial information of HSI, a method is proposed based on the stacked contractive autoencoder (SCAE). Extracted spatial information is fused with the spectral information of the unclassified pixel. SCAE was used to extract the deep features of the fused information. Logistic regression (LR) was used to determine the categories of each pixel in the HIS. Compared with the existing methods on Indian Pines, Pavia University datasets, the results show that the method proposed in the paper has achieved overall accuracy of 89.943%, 93.949% on the two datasets respectively, and with better classification performance.