Application of deep learning model in brick kiln information extraction

Anam Sabir*, Anil Kumar, Prakash Chauhan

Abstract

Developing Asian nations such as India use clay bricks as basic building blocks for construction purposes. The process of firing of bricks leads to generation of exhaust gases which in turn degrade air quality and have considerable adverse effect on the flora and fauna in the vicinity. The area in north of India, especially surrounding national capital New Delhi, is highly polluted and has an abundance of brick kilns which further worsens the situation. Thus, keeping a record of brick kilns in an area can help regulate their optimal working considering both, the production and environmental impact. Satellite imagery proves to be helpful in this case as the amount of ground work is reduced up to a large extent. The dataset was formed using images from Google Earth after appropriate pre-processing. SSD (Single Shot Detector) architecture of deep learning was used to detect brick kilns and hence regulate their impact. The approach followed was used to train and test the model with enough remotely sensed data folowed by validation of results achieving an overall accuracy of 81.9% and precision of 95.06%. Furthermore, the individual brick kiln data was processed to do sub-classification in terms of location, shape and status of brick kiln. Depending upon the total number of brick kilns present in an area, the approximate amount of exhaust was estimated.

Keywords: Brick kiln, Bounding Box, Deep Learning, Single Shot Detector (SSD)

Full Text: