DEHAZING REMOTE SENSING IMAGE USING VISION TRANSFORMER WITH SALIENCY MAP TRANSMISSION TECHNIQUE

Abstract

Abstract - Remote sensing imagery often suffers from haze and atmospheric interference, impacting the quality and interpretability of captured data. This project proposes a novel approach utilizing Vision Transformer (ViT) integrated with Saliency Map Transmission (SMT) for enhancing single remote sensing image dehazing. The process involves a systematic workflow encompassing various stages. Initially, the input image undergoes preprocessing, employing a median filter to mitigate noise interference. Subsequently, Morphological Calculations are applied to extract essential features and improve image quality. The Dark of Channel In order to assist with the computation of Intensity Weights, Prior is then used to estimate the haze distribution throughout the picture. The innovation lies in the integration of Scatter Pixel Calculations and Scene Radiance Calculation, refining the dehazing process. The pivotal addition of Saliency Map Transmission optimizes the transmission map, enhancing the accuracy of dehazing outcomes. Moreover, the incorporation of Recurrent Neural Networks (RNN) in the pipeline refines the deblurring process, ensuring clearer and sharper dehazed outputs. The entire workflow is implemented using MATLAB tools, providing a comprehensive and efficient platform for image enhancement and analysis. The goal of this novel framework is to greatly enhance the quality of dehazed remote sensing imagery by combining cutting edge deep learning techniques with conventional image processing approaches. The proposed approach showcases promising potential in advancing image dehazing techniques, promising enhanced clarity and fidelity in single remote sensing images. Keywords: RNN, MATLAB, Image Processing etc.