Search

Synthetic aperture radar (SAR) image interpretation is a great scientific application challenge. The classification of SAR image targets has become one of the main research directions for SAR image interpretation. Therefore, achieving fast and accurate SAR image target classification has always been a research hotspot in this field. Here, the authors propose a classification method based on a regularised stochastic configuration network (SCN), which randomly assigns the input weights and biases with constraint and finds out the output weights all together by solving a global least squares problem. Experimental results on the moving and stationary target acquisition and recognition benchmark dataset illustrate that the regularised SCN classifies ten-class targets to achieve an accuracy of 94.6%. It is significantly superior to the traditional SCN model and effectively improves the generalisation ability of the network.

Synthetic aperture radar (SAR) image interpretation is a great scientific application challenge. The classification of SAR image targets has become one of the main research directions for SAR image interpretation. Therefore, achieving fast and accurate SAR image target classification has always been a research hotspot in this field. Here, the authors propose a classification method based on a regularised stochastic configuration network (SCN), which randomly assigns the input weights and biases with constraint and finds out the output weights all together by solving a global least squares problem. Experimental results on the moving and stationary target acquisition and recognition benchmark dataset illustrate that the regularised SCN classifies ten-class targets to achieve an accuracy of 94.6%. It is significantly superior to the traditional SCN model and effectively improves the generalisation ability of the network. [ABSTRACT FROM AUTHOR]