Unsupervised Change Detection in High Spatial Resolution Optical Imagery Based on Modified Hopfield Neural Network

This paper addresses the problem of unsupervised change detection in high spatial resolution optical remote sensing images based on Hopfield neural network (HNN). An optimization relaxation approach based on the analysis of a modified Hopfield neural network is proposed for solving the change detection problem. The modified Hopfield neural network is designed to characterize a texture in terms of spatial-contextual information included in the neighborhood of each pixel within each color plane and interaction between different color planes. The network topology is built on the difference image so that each pixel in the RGB color planes is represented as a node in the network which is connected to its neighborhood units both in its own plane and other two planes. Each node is represented by its state which characterizes the pixel changed or unchanged, and an energy function is derived to represent the overall status of the whole network. Change detection maps are obtained by iteratively updating the output status of the neurons until the network converges. The main contribution of this paper lies in the construction of a novel continuous Hopfield-type neural network on the RGB image for solving the unsupervised image change detection problem. Experiments results obtained on two sets of remote sensing imagery confirm the effectiveness of the proposed approach.