Hyperspectral Target Detection Using Learned Dictionary

Target contamination is the main problem in traditional target detection (TD) methods in hyperspectral imagery when estimating the background distribution with different models. Sparse approximation is introduced to tackle the detection problem, yet while using windows to build a sparse dictionary, the contamination problem remains. In our approach, we utilize a learning method based on convex optimization to build such a dictionary. Through its application, prior information such as the size of windows can be spared while considerably reducing the occurrence of contamination. To verify the efficacy of using the learned dictionary (LD), the dictionary built through the dual window (DW) method is used as a comparison, and two sparse TD methods are employed afterward. Experimental results show that, by using the LD, a better result is obtained compared with the methods using a traditional DW background dictionary.