Date of Award
8-17-2025
Document Type
Masters Project
Abstract
Spontaneous combustion in coal mines have been a concern for miners all over the world, particularly in China, the USA and India. If allowed to establish over a large area it can become a serious environmental concern. It is important to detect it at initial stages and isolate affected coal seam to quench fire. Monitoring combustion in spoil piles of abandoned and artisanal coal mines is even more difficult due to lack of access to regulating authorities. By using Deep Learning Neural Networks, Convolutional Neural Network models can be trained to detect spontaneous combustion. These scans can be made more frequently, and detection can be done at early stages. Training dataset consists of over 5000 images processed from different mines at different conditions. The dataset is processed using Python Libraries such as TensorFlow, NumPy and Pandas to develop three different models i.e. simple CNN, LeNet and AlexNet. Accuracy comparison shows that LeNet is the most suitable model giving accuracy of 97%. It was observed that selection of an appropriate dataset is more critical than selecting model architecture.
Recommended Citation
Juneja, Jatin, "Developing a deep learning model for detection of spontaneous combustion in open pit coal mines" (2025). Engineering . 549.
https://ualaska.researchcommons.org/uaf_grad_engineering/549
Handle
http://hdl.handle.net/11122/16330