Image Despeckling For Synthetic Aperture Radar Using Reinforcement Learning

Synthetic aperture radar (SAR) is an imaging mechanism that provides a plethora of information for a large number of applications ranging from surveillance to research. It represents a distinct advancement over the optical spectrum which is known to be highly dependent on the meteorological conditions of the subject3. However, the SAR images are subjected to speckle noises. These noises present a challenge to the analysis and interpretation of the images6. Several despeckling methods that provide an effective solution to the speckle noises often compromise the quality of the images, especially in heterogeneous terrains. This paper lays the groundwork for a Machine learning approach to the speckle problem. The method implements a Reinforcement learning model, that works by implementing a Q-algorithm, trained to obtain the closest possible pixel value of a speckled image to its clean counterpart17. The method is oriented towards formulating a simplified and feasible approach to despeckling SAR images catering to a vast variety of speckle noises and varied speckle patterns and facilitating their applications in various fields.