CLASSIFICATION OF SPATIAL OBJECT STATES BASED ON HEIGHT MAPS USING CONVOLUTIONAL NEURAL NETWORKS FOR ORE SENSOR SORTING SYSTEMS
DOI:
https://doi.org/10.34185/1991-7848.itmm.2026.01.008Keywords:
sensor ore sorting, height map, laser scanning, convolutional neural network, machine learning, classification, point cloud, overfitting, data augmentationAbstract
This paper addresses the problem of automatic classification of spatial object states – single, united (close but non-overlapping), and overlapped – in precision sensor-based ore sorting systems. A method for transforming 3D point clouds obtained from a laser scanner into 2D height maps is proposed, followed by classification using convolutional neural networks (CNN). The influence of grid resolution (32×32, 64×64, 128×128) and model architecture on classification quality is investigated. A comparative analysis of five architectures is conducted: MLP, basic CNN, CNN+Dropout, CNN+L1, and Deep CNN. Results demonstrate that the Deep CNN achieves up to 92% validation accuracy on real data at 64×64 resolution, outperforming the MLP baseline by 29 percentage points.
References
Chalapathy R., Chawla S. Deep Learning for Anomaly Detection: A Survey // ACM Computing Surveys. – 2019. – Vol. 51, No. 5. – DOI: 10.1145/3299869.
Gholami A. et al. A Survey of Quantization Methods for Efficient Neural Network Inference // Low-Power Computer Vision. – 2021. – DOI: 10.1201/9781003162810-13.
Parisi G. I. et al. Continual Lifelong Learning with Neural Networks: A Review // Neural Networks. – 2019. – Vol. 113. – DOI: 10.1016/j.neunet.2019.01.012.
Peukert S. et al. A Review of Sensor-Based Sorting in Mineral Processing // Minerals. – 2022. – Vol. 12, No. 11. – DOI: 10.3390/min12111364.
Qi C. R. et al. PointNet++: Deep Hierarchical Feature Learning on Point Sets // NeurIPS. – 2017. – P. 5099–5108.
Rai S. et al. Application of Machine Learning Algorithms for Mineral Classification // Minerals Engineering. – 2021. – Vol. 170. – DOI: 10.1016/j.mineng.2021.107054.
Robben C., Wotruba H. Sensor-Based Ore Sorting Technology in Mining // Minerals. – 2019. – Vol. 9, No. 9. – DOI: 10.3390/min9090523.
Wang M., Deng W. Deep Visual Domain Adaptation: A Survey // Neurocomputing. – 2018. – Vol. 312. – DOI: 10.1016/j.neucom.2018.05.083.
