Title: Sorting of coal and coal waste with transferred deep kernel learning

Authors: Yufan Li

Addresses: Department of Automation, Shanghai Jiao Tong University, Shanghai, China

Abstract: Gangue is difficult to sort out because it looks very close to coal, as do the rocks covered with coal dust on the surface. To simulate the sorting scenario, a dataset is constructed including coal/rock/gangue images. As the scale of our dataset is small, and the ground truth of some images cannot be observed by the naked eye, which may mislead models to learn the wrong thing, the problem is summarised as a weakly supervised learning task. A transferred deep kernel learning model is proposed, in which weights are pre-trained on ImageNet, and the feature extractor along with Gaussian process are trained together to enable end-to-end learning capabilities. The results demonstrate that it achieves high accuracy in this classification task, which might help to automate the sorting process, and might be useful in real engineering tasks that require significant resources to label images.

Keywords: coal; sorting; image classification; Gaussian process; weakly supervised learning; transferred deep kernel learning; TDKL.

DOI: 10.1504/IJSCC.2023.131970

International Journal of Systems, Control and Communications, 2023 Vol.14 No.3, pp.274 - 285

Received: 13 Jan 2023
Accepted: 31 Jan 2023
Published online: 05 Jul 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article