Publication

Abstract : Remote sensing data provides satellite imagery which is specific in nature. Compared to the normal photography, these have much more pixels, about tens of thousands. Obtaining meaningful information from remotely sensed images is a collective process involving the expertise in areas like pattern recognition, image analysis, and machine learning. The use of right learning algorithm is the key to obtain detailed and accurate results which match the ground truth. Some of the traditional supervised classification techniques for satellite images have been mahalanobis, maximum likelihood classifier, and support vector machines . The proposed work aims at using the latest upcoming classification techniques in machine learning. Deep learning, a subfield of machine learning has slowly found its place in many domains due to its nature of continuous analysis of the data to draw conclusions. The dense layers of the deep neural network can give promising results with remotely sensed data as has been experimented in this work. During the months of October and November, around 7 to 8 million metric tons of paddy residue are burned openly throughout Punjab each year. The present study is the identification of these burnt paddy fields in Patiala and surrounding areas The data used is Sentinel 2A,2B. A case study of the burnt fields is done using possibilistic c-means classifier to understand the application of remote sensing with regards to image classification . The results have been promising as verified with the previous published work cited in the paper.