DEEP LEARNING-BASED APPROACH FOR ANNUAL CHANGE DETECTION AND OPEN PIT COAL MINE DETECTION USING SATELLITE IMAGERY

This study introduces a novel deep learning-based method for detecting open pit coal mines in medium high (MH)resolution satellite imagery and analyzing annual variations in coal mining areas. The study aims to monitor and precisely identify coal mines, given their vital function as a primary energy source and a significant contributor to greenhouse gas emissions. The proposed method employs the U-Net architecture and ResNet 34 as its backbone for accurate detection and classification. The applied dataset consists of multispectral imagery from Sentinel-2 and synthetic aperture radar (SAR) imagery from Sentinel-1. Manual labelling of known coal mine locations using mining data as a reference, subdividing Sentinel image patches for training U-Net convolutional neural networks (CNNs) to classify coal mine and non-coal mine areas, and training and testing U-Net architecture and ResNet as a backbone deep learning model are the three essential steps involved in the process. The classification accuracy of the coal mining detection deep learning model is 97%, and the kappa value is 0.91. Preliminary results indicate that the investigation demonstrates the evolution of coal mining from 2016 to 2021, with an increase of over 40% in coal mining area in 2019. In 2017 and 2020, the area mined for coal will decrease. Variations in annual coal mining variations emphasize the significance of replanting efforts. The proposed method uses deep learning and satellite imagery to provide an accurate and efficient solution for detecting and monitoring open-pit coal mines. Incorporating artificial intelligence into dynamic coal mining activities yields valuable insights that aid in making informed decisions for sustainability.