With the rapid development of deep learning in computer vision, object detection algorithms are now widely used in industrial inspection, security surveillance, and embedded intelligent systems. However, deep learning models typically rely on high-performance CPUs or GPUs, making direct deployment on resource-constrained embedded platforms challenging due to limited compute capability and power constraints. This paper takes the YOLOv5 object detection model as the research object and focuses on porting and deploying it on the domestic MLU220 AI accelerator chip platform. By analyzing the YOLOv5 network structure and leveraging the hardware characteristics of the MLU220 platform, we completed model quantization, offline compilation, cross-compilation, and the construction of a multi-threaded inference system. On this basis, we conducted comparative experiments to evaluate the end-to-end inference performance on both a PC platform and the MLU220 platform. Experimental results show that under the premise of nearly identical detection results, the MLU220 platform effectively reduces overall inference time, verifying the feasibility and engineering application value of deploying object detection models on domestic AI accelerator hardware.

Jingshan Zeng. Porting and Performance Analysis of a YOLOv5-Based Object Detection Model on the MLU220 Platform. Academic Journal of Computing & Information Science (2026), Vol. 9, Issue 1: 87-93. https://doi.org/10.25236/AJCIS.2026.090111.

[1] Redmon J., Divvala S., Girshick R., et al. You Only Look Once: Unified, Real-Time Object Detection. In Proc. of IEEE CVPR, 2016.

[2] Redmon J., Farhadi A. YOLO9000: Better, Faster, Stronger. In Proc. of IEEE CVPR, 2017.

[3] Redmon J., Farhadi A. YOLOv3: An Incremental Improvement. [Online] arXiv:1804.02767, 2018.

[4] Bochkovskiy A., Wang C.Y., Liao H.Y.M. YOLOv4: Optimal Speed and Accuracy of Object Detection. [Online] arXiv:2004.10934, 2020.

[5] Jocher G. YOLOv5 [EB/OL]. Ultralytics, 2020. (GitHub repository: https://github.com/ultralytics/ yolov5).

[6] Wang C.Y., et al. CSPNet: A New Backbone that Can Enhance Learning Capability of CNN. In Proc. of IEEE CVPR Workshops, 2020.

[7] Liu S., Qi L., Qin H., et al. Path Aggregation Network for Instance Segmentation. In Proc. of IEEE CVPR, 2018.

[8] Cambricon Technology. Cambricon releases edge AI chip “Siyuan 220/MLU220” [EB/OL]. Company News, 2019-11-14.

[9] Jacob B., Kligys S., Chen B., et al. Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference. In Proc. of IEEE CVPR, 2018.

[10] Cambricon Technology. MagicMind Developer Guide [EB/OL]. 2021. (Cambricon Developer Documentation for MLU220).