Search

Improved Yolov10 for Real-Time Multi-Scale Detection of Floor Contaminants in Precision Machining Workshops

Author(s)

Juhua Chen*, Fan Chen

Affiliation(s)

Chongqing Industry and Trade Polytechnic, Chongqing, China *Corresponding Author

Abstract

Floors in precision machining workshops accumulate contaminants that differ enormously in scale, from sub-centimetre metal swarf to metre-wide patches of spilled coolant. Fine debris scratches finished workpieces carried across it, while extensive oily films create slipping hazards and contaminate guide rails, yet routine inspection still relies on the human eye, which misses small debris and reacts late. YOLOv10 offers end-to-end inference without non-maximum suppression, but its detection levels and loss function are poorly matched to millimetre-sized targets lying against a cluttered floor. This paper makes three changes to YOLOv10s: DCNv4 deformable convolution is embedded at the neck fusion nodes so that the sampling grid follows the irregular outline of stains; a shallow P2 branch preserves the high-resolution detail of minute debris; and the CIoU loss is replaced by the WIoUv3 dynamic non-monotonic focusing loss to rebalance large and small instances. On a self-collected dataset of 3746 workshop floor images the improved model reaches 92.6% mAP@0.5, 4.2 percentage points above the baseline, and runs at 66 FPS with only 0.8 M extra parameters, which meets the real-time budget of embedded monitoring devices on the shop floor.

Keywords

YOLOv10; Ground Pollutants; Multi-Scale Object Detection; Deformable Convolution; Real-Time Detection

References

[1]Wang, L., Fang, H.: 'Research on UAV detection algorithm for low-altitude small targets based on improved YOLOv10', Science and Technology Innovation, 2026, (17), pp. 120–123. [2]Shamsoshoara, A., Afghah, F., Razi, A., et al.: 'Aerial imagery pile burn detection using deep learning: the FLAME dataset', Computer Networks, 2021, 193, p. 108001. [3]Padilla, R., Passos, W.L., Dias, T.L.B., et al.: 'A comparative analysis of object detection metrics with a companion open-source toolkit', Journal of Imaging, 2021, 7, (8), p. 119. [4]Wang, C.-Y., Yeh, I.-H., Liao, H.-Y.M.: 'YOLOv9: learning what you want to learn using programmable gradient information', Proc. European Conf. Computer Vision (ECCV), 2024, pp. 1–18. [5]Xiong, Y., Liao, H., Li, X., et al.: 'Efficient deformable ConvNets: rethinking dynamic and sparse operator for vision applications', Proc. IEEE/CVF Conf. Computer Vision and Pattern Recognition (CVPR), 2024. [6]Dosovitskiy, A., Beyer, L., Kolesnikov, A., et al.: 'An image is worth 16×16 words: transformers for image recognition at scale', Proc. Int. Conf. Learning Representations (ICLR), 2021. [7]Terven, J., Cordova-Esparza, D.M., Romero-Gonzalez, J.A.: 'A comprehensive review of YOLO architectures in computer vision: from YOLOv1 to YOLOv8 and YOLO-NAS', Machine Learning and Knowledge Extraction, 2023, 5, (4), pp. 1680–1716. [8]Zhang, S., Hu, Y., Zhou, J.: 'Strip steel surface defect detection algorithm based on DN-YOLO', Shanxi Metallurgy, 2026, 49, (7), pp. 127–130. [9]Wu, S., Yu, L.: 'RD-YOLO: an improved YOLOv11 algorithm for small object detection in UAV imagery', Journal of Harbin University of Commerce (Natural Science Edition), 2026, 42, (4), pp. 419–429. [10]Nandal, P., Gaur, V., Bhutani, M., et al.: 'Automated urine sediment microscopic image analysis using YOLOv10 with chimpanzee leader election optimization', Journal of Visualized Experiments, 2026, (233). [11]Wu, Z., Li, H., Xue, R.: 'Lightweight PCB defect detection model based on improved YOLO', Journal of Southwest Minzu University (Natural Science Edition), pp. 1–13, 2026. [12]Zheng, Z., Wang, P., Liu, W., et al.: 'Distance-IoU loss: faster and better learning for bounding box regression', Proc. AAAI Conf. Artificial Intelligence, 2020, 34, (7), pp. 12993–13000. [13]Li, X., Wang, W., Wu, L., et al.: 'Generalized focal loss: learning qualified and distributed bounding boxes for dense object detection', Advances in Neural Information Processing Systems, 2020, 33, pp. 21002–21012