Target Detection Based on The Artificial Neural Network Technology

As it is difficult to detect small ship targets from complex background in an IR image because of the impact of kinds of noises, complex background and the smallness of the target, this paper puts forward a method of background prediction utilizing the neural network. The method estimates the background of the input image nonlinearly, gains the residual error image, and detects the target. By comparison, NARX whose hidden layer transfer function utilizing L-M algorithm can achieve more excellent target detection than other neural networks like BP in training speed and stability, thus it is research valuable in the field of small target detection in complex background