Fusion of Dual Spatial Information for Hyperspectral Image Classification

The inclusion of spatial information into spectral classifiers for fine-resolution hyperspectral imagery has led to significant improvements in terms classification performance. task spectral-spatial image (HSI) remained challenging because high intraclass spectrum variability and low interclass variability. This fact made the extraction highly active. In this work, a novel HSI framework using fusion dual is proposed, which built by both exploiting pre-processing feature post-processing optimization. stage, an adaptive texture smoothing method proposed construct structural profile (SP), makes it possible precisely extract discriminative features from HSIs. SP used here first time remote sensing community. Then, extracted fed classifier. optimization pixel-level classifier obtain class probability followed extended random walker-based technique. Finally, decision rule utilized fuse probabilities obtained two different stages. Experiments performed on three data sets scenes illustrate that can outperform other state-of-the-art techniques. addition, method, i.e., SP, effectively improve discrimination between land covers.