A method for object identification in multispectral image based on tensor approach

Lefei Zhang, Xin Huang, Bo Du, Liangpei Zhang

Abstract: In this paper, we propose a new approach for multispectral image object identification using a proximal support tensor machine. The image object is represented as a 3-order tensor that encodes both the spectral and neighborhood spatial information, and then applies the proximal support tensor machine (PSTM) for object identification. Experiments demonstrate that by comparison with the support tensor machine (STM) and conventional support vector machine (SVM), the proposed PSTM reduces the false alarm rate and training time while keeps the 100% identification rate, which indicates that PSTM is an effective method for multispectral image object identification.