Lake Ice Infrared Image De-noising using Deep Learning Convolutional Neural Network

Elimination of noise from the acquired images is one of the fundamental tasks so as to restore the high quality of images to increase the results of segmentation, classification, recognition, detection etc. In this research work, Infrared imaging is used to capture the images with various temperature ranges during day and night time. Infrared thermography is used as a nondestructive image acquisition sensor. So infrared imaging sensors are used to identify and classify different ice types formed on the rivers during winter season in western countries. However, you can observe blur and degradation in the acquired images through infrared imaging, as the climatic nature is involved in the process. These factors affect the quality of the image and also the quantitative analyses of the images. Especially, when there are deeper faults which are found in an under laying substance and high thermal conductivity materials can also be studied. Acquired infrared images are usually affected with noises. In this paper, de-noising of infrared images is carried out using convolution neural network model in deep learning. In this work Gaussian white noise is added to the infrared images with varying noise intensities from 1% to 10% and then the convolution neural network model is applied to de-noise the infrared images. In the qualitative analysis of the infrared images edge factor, uniform region, texture, smoothness, non-uniform region, structure of the region of interest is considered. For quantitative analysis, mean square error, peak-signal- noise-ratio and structure similarity index measure is used. The results of traditional methods of image de-noising are compared with convolution neural network based method. It is concluded from the experimental work that the convolution neural network based method proves to be better for removal of Gaussian white noise when compared with traditional methods.