Affiliation:
1. College of Information Science and Engineering, Northeastern University, Shenyang, China
2. School of Astronautics, Harbin Institute of Technology, Harbin, China
Abstract
Recently, accelerating the improvement in the level of iron ore mining has become extremely important for the sustainable development of the steel industry. Various foreign objects such as steel bars, wood, and plastic pipes easily appear on iron ore belt conveyors, which cause damage to equipment and harm the personal safety of mining workers. Accordingly, a lightweight detection model based on you only look once (YOLO)v3 is proposed in this study. This study first uses the median filter method to preprocess the belt foreign body image to remove the influence of dust and improve the clarity of ore edges. Second, we train the YOLOv3 belt foreign matter detection algorithm based on the selected data set to detect belt foreign matter and evaluate the model based on mean average precision (mAP) and other indicators. Finally, after implementing the sparse training based on the batch normalization (BN) layers, the channel-pruning and layer-pruning strategies are implemented to simplify the YOLOv3 model, followed by parameter fine-tuning. When the accuracy of the model is not affected, our model realizes smaller calculations, faster processing, and a smaller size compared to the original YOLOv3 model. Hence, the model effectively achieves real-time recognition of foreign matter on the ore conveyor belt.
Funder
National Natural Science Foundation of China
National Key Research and Development Program of China
fundamental research funds for the central universities
Fundamental Research Funds for Liaoning Natural Science Foundation, China
Reference21 articles.
1. Denil M, Shakibi B, Dinh L, et al. (2013) Predicting parameters in deep learning. Available at: https://arxiv.org/abs/1306.0543
2. Moving object detection from moving platforms using Lagrange multiplier
3. Girshick R (2015) Fast R-CNN. Available at: https://arxiv.org/abs/1504.08083
4. Han S, Mao H, Dally WJ (2015) Deep compression: Compressing deep neural networks with pruning, trained quantization and Huffman coding. Available at: https://arxiv.org/abs/1510.00149
5. He K, Zhang X, Ren S, et al. (2015) Deep residual learning for image recognition. Available at: https://arxiv.org/abs/1512.03385
Cited by
4 articles.
订阅此论文施引文献
订阅此论文施引文献，注册后可以免费订阅5篇论文的施引文献，订阅后可以查看论文全部施引文献