Dehazing of Multispectral Remote Sensing Images Using CNN with ResNet

Satellite image utilization for data interpretation has significantly increased in the current scenario. Deep learning techniques have gained widespread popularity for analyzing satellite image data. However, a major challenge in this field is the presence of haze, which hampers accurate interpretation. Haze occurs when fine dust, smoke, or light vapors reduce the air's transparency, resulting in a lack of contrast in affected regions of satellite images. To tackle this issue, atmospheric correction is commonly employed, but it is a complex and time-consuming process that requires knowledge of various geophysical factors to achieve precise results. Hence, the objective of this research is to develop and implement image processing algorithms that can effectively restore pixels affected by haze, thereby improving the interpretability of satellite images. The proposed work consists of two main phases. Firstly, noise reduction techniques will be applied to the images to enhance their quality. In the second phase, a Convolutional Neural Network with a Residual structure will be utilized to dehaze the images. The evaluation of the proposed system will focus on the Kondapalli region and its surrounding areas in the Kondapalli dataset. This research aims to develop a robust solution that can enhance satellite images affected by haze, providing clearer and more interpretable data for various applications. The effectiveness of the proposed system will be compared with the Dark Object Subtraction method.