SAR Image Despeckling Convolutional Model with Integrated Frequency Filtration Technique

The presence of unwanted multiplicative speckle noise within synthetic aperture radar (SAR) sensor data largely degrades its quality and challenges the processing of these data. Thus, in order to increase the performance of a SAR sensor data processing system, speckle removal techniques must be adopted. This paper demonstrates a novel framework that can be utilized to remove the presence of unwanted multiplicative speckle-noise within SAR images. The proposed framework is a convolutional neural network (CNN) based despeckling model with an integrated frequency-filtration based detail preservation module realized through skip connection. This configuration ensures maximal removal of speckle-noise from the raw input data with minimal compensation of sharp features present in the data. An extensive comparative analysis is carried out to study the performance of the proposed framework with other state-of-the-art despeckling models. This study has depicted the quantitative as well as visual performance superiority of the proposed model. Experimentation over real standard SAR data also demonstrates the superior performance of the proposed framework over other despeckling models present in the literature.