An Improved Multiscale Model Based on Boosting Method in HSIs Ground Object Recognition

Abstract

Ground object recognition from remote sensing images is an important hotspot of hyperspectral image recognition in recent years. Most methods mainly focus on pixel or spectral characteristics to classify hyperspectral images (HSIs), but they do not fully express the inherent spatial multi-scale features and frequency domain structural features of the data. A novel and unified model that integrates Multi-scale Gabor and LPQ features (Ms-GLPQ) is proposed, which fully extracts and fuses the multiscale structural features in the spatial and frequency domain of HSIs. Finally, a multi-feature region descriptor with stronger discrimination is obtained, and then the boosting tree methods of machine learning are carried out for recognition. Few-shot learning (about 3% for training) recognition experiments are conducted on three public HSIs datasets, and the accuracy is significantly improved. The results demonstrate the excellent performance of the proposed model compared with the traditional feature extraction methods. Besides, higher recognition accuracy is acquired by the boosting tree model classifier, which illustrates the applicability of the Ms-GLPQ model on the ground object recognition in HSIs.