Classification of spatial states of objects based on height maps using convulsive neural networks for sensory sorting systems of ores
DOI:
https://doi.org/10.34185/1562-9945-3-164-2026-13Keywords:
sensory sorting of ores, height map, laser scanning, convolutional neural network, image classification, machine learning, precision sorting, point cloud, overtraining, data augmentationAbstract
The article considers the problem of automatic classification of spatial states of objects in precision sensory sorting systems of ore raw materials. The relevance of the study is due to the limited effectiveness of traditional segmentation algorithms in conditions of overlapping and approaching pieces of ore on the conveyor. The aim of the work is to develop and study an approach to classifying three spatial states of objects (single, combined, superimposed) based on height maps obtained using laser scanning. Convolutional neural networks (CNN) are used as the research method. A comparative analysis of MLP, basic CNN, CNN with regularization and deep CNN architectures is carried out. The influence of the height map dimension (32×32, 64×64, 128×128) on the classification quality was investigated. It was found that the deep CNN demonstrates the best accuracy on real data – up to 90% on the validation sample with a dimension of 64×64. The effect of overtraining was revealed when the input data dimension increased. The obtained results confirm the promising application of CNN for the analysis of height maps in the tasks of intelligent control of the sorting process.
References
Chalapathy R., Chawla S. Deep Learning for Anomaly Detection: A Survey // ACM Com-puting Surveys. – 2019. – Vol. 51, No. 5. – P. 1–48. – DOI: 10.1145/3299869. – URL: https://arxiv.org/abs/1901.03407
Gholami A., Kim S., Dong Z., Yao Z., Mahoney M. W., Keutzer K. A Survey of Quantiza-tion Methods for Efficient Neural Network Inference // Low-Power Computer Vision. – 2021. – P. 291–326. – DOI: 10.1201/9781003162810-13. – URL: https://arxiv.org/abs/2103.13630
Parisi G. I., Kemker R., Part J. L., Kanan C., Wermter S. Continual Lifelong Learning with Neural Networks: A Review // Neural Networks. – 2019. – Vol. 113. – P. 54–71. – DOI: 10.1016/j.neunet.2019.01.012. – URL: https://arxiv.org/abs/1802.07569
Peukert S., Heinig T., Tolosana-Delgado R., Bachmann K., Gutzmer J. A Review of Sen-sor-Based Sorting in Mineral Processing: The Potential and Limitations of Sensing and Data Processing // Minerals. – 2022. – Vol. 12, No. 11. – P. 1364. – DOI: 10.3390/min12111364. URL: https://www.mdpi.com/2075-163X/12/11/1364
Qi C. R., Yi L., Su H., Guibas L. J. PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space // Proceedings of NeurIPS. – 2017. – P. 5099–5108. URL: https://arxiv.org/abs/1706.02413
Rai S., Nistala S. H., Danda A. K., Sreenivas T. Application of Machine Learning Algo-rithms for Mineral Classification from Hyperspectral Data // Minerals Engineering. – 2021. – Vol. 170. – P. 107054. – DOI: 10.1016/j.mineng.2021.107054.URL: https://www.sciencedirect.com/science/article/abs/pii/S0892687521002636
Robben C., Wotruba H. Sensor-Based Ore Sorting Technology in Mining–Past, Present and Future // Minerals. – 2019. – Vol. 9, No. 9. – P. 523. – DOI: 10.3390/min9090523. – URL: https://www.mdpi.com/2075-163X/9/9/523
Wang M., Deng W. Deep Visual Domain Adaptation: A Survey // Neurocomputing. – 2018. – Vol. 312. – P. 135–153. – DOI: 10.1016/j.neucom.2018.05.083. URL: https://arxiv.org/abs/1802.03601
Downloads
Published
Issue
Section
License
Copyright (c) 2026 System technologies
This work is licensed under a Creative Commons Attribution 4.0 International License.
