Infrared aerial small target detection based on digital image processing

Abstract

Aiming for the problem detection of infrared imaging aerial small target under complex background, an intelligent algorithm is presented based on digital image processing which mainly makes use of the theory of contourlet transform and BP(Back propagation) neural network. Firstly, this method transforms the infrared image from space domain to contourlet domain. Then, in order to suppress most complex background, this algorithm sets lowpass coefficients to zero because it includes most gentle background information of the infrared image. Furtherly, this method constructs a novel threshold formula for bandpass coefficients which is based on the classic formula and takes the directional energies into account for restraining the remained background edges and noises. Subsequently, the reverse transform is carried out and the preprocessing result is obtained. Secondly, taking pixel’s grayscale, horizontal gradient, vertical gradient, diagonal gradient, neighborhood mean and neighborhood variance as input feature vector, a BP neural network which has three layers is constructed and trained so that the non-linear relationship between the features and the target or background’s pixel. In the end, infrared small target is detected by this BP network which has finished the procedure of training. The experimental results show that the method given by this paper can not only realize the suppression for the infrared complex background effectively, but also detect the small target whose SNR(Signal Noise Ratio) value is above 2 steadily.