Sensor and method fusion in remote sensing imagery with neural networks

Abstract

The need for sensor and method fusion arises in many practical applications, one of those is extended object imaging in passive remote sensing/imaging (RSI) systems that employ different platforms of sensors. In this paper we propose a new approach to solving simultaneous image restoration problems incorporating fusion of all RSI systems by integrating these problems into one augmented inverse problem by imposing the minimum entropy (ME) image model as prior knowledge for restoration (Falkovich et al. 1989). We investigate the fine structure of a Hopfield neural network and propose a sensor fusion method that can be implemented via modification of such a network into the maximum entropy neural network (MENN) using minimum entropy regularization. It is shown that applying the proposed method, the sensor and/or method fusion tasks can be solved without principal complication of the resultant structure of the MENN independent of the number of sensor platforms or methods to be fused. The overall MENN algorithm is presented. The results are illustrated by simulation samples and compared with other high resolution image restoration techniques.