CNN Based Restoration of Remote Sensing Imagery through Denoising Schemes
Sujatha Canavoy Narahari, Sivaneasan Bala Krishnan, Prąsun Chakrabarti · 2024
Image restoration through denoising is a technique used to remove noise from noisy images generated by sensor malfunctions, environmental factors, lossy compression algorithms, electrical interference, and so on. To improve the image quality, we present a novel approach using a deep convolutional neural network, which involves of two convolution layers with a ReLU activation function, one convolution layer with no activation function, and a convolution layer with a sigmoid function to limit the mapping value to a range between 0 and 1. To evaluate the customized model architecture, we generated some noisy images using salt and pepper noise and fed them into the neural network as noisy input, with the final output generated after the fourth convolution layer with the sigmoid function. The experimental results are astonishing and outperform most of the deep CNN models over the past decade.