Pulse-coupled neural network and its optimization for segmentation of electrical faults with infrared thermography

Abstract

This paper describes a method for segmenting electrical equipment fault with infrared thermography by using pulse-coupled neural networks. The pulse coupled neuron model used in PCNN is an optimization of the original neural model, in order to easily control and alter the behavior of neuron activity. In the receptive field, the image and its gradient information are regarded as the inputs of PCNN. The method for adjusting the value of linking coefficient is then derived from the principle of maximum likelihood estimate in the region of temporal pulse outputs, thus building the inner relationship between the parameters and image statistics. Besides, the edge constraint method, which is integrated into the pulse generator, is designed to alter the behavior of the neighboring neurons to be captured. It is shown that the addition of the constraint to the model increases the possibility of desired fault region segmentation. Finally, several experimental results, especially performed on the electrical equipment fault images, show that the proposed model has better performance than some existing PCNN-based models in the performance of image segmentation.