With the rapid development of convolutional neural networks, object detection technology has been revitalized, finding broader applications in robots, cars, cameras, and more. Many object detection algorithms can efficiently annotate anchor boxes for objects within the field of view, determining the specific location of the detection target. However, as the performance of detection algorithms improves, the issue of energy consumption arises, as high performance requires the sacrifice of high energy. Yet, most mobile devices have limited resources and cannot bear such a high load. The goal of this paper is to construct a low-energy consumption object detection algorithm, providing insights for the continuous efficient operation of mobile devices. YOLOv5s is selected as the base model, and its structure is improved. In neural networks, convolution operations are the main part of energy consumption. To reduce convolution operations, the backbone network is changed to the ShuffleNetv2 module, and DWConv is used to replace the C3 structure in the detection head. To improve model accuracy, a channel attention mechanism is introduced, and the loss function is modified. Ultimately, the new model significantly reduces the model’s parameter quantity and size, thereby reducing its energy consumption, while maintaining essentially unchanged accuracy.

Xiaokun Qi, Tian He. Low-power neural network based on YOLOv5s. International Journal of Frontiers in Engineering Technology (2024), Vol. 6, Issue 2: 64-72. https://doi.org/10.25236/IJFET.2024.060209.

[1] Hu L, Li Y. Micro-YOLO: Exploring Efficient Methods to Compress CNN based Object Detection Model; proceedings of the International Conference on Agents and Artificial Intelligence, F, 2021 [C].

[2] Zhang Y, Lee C-C, Hsieh J-W, et al. CSL-YOLO: A New Lightweight Object Detection System for Edge Computing [J]. ArXiv, 2021, abs/2107.04829.

[3] Yujie Luo, Jian Zhang, Liang Chen, et al. Lightweight target detection algorithm based on adaptive spatial feature fusion [J]. Laser & Optoelectronics Progress, 2022, 59(4): 0415004--11.

[4] Chen Y, Chen X, Chen L, et al. UAV Lightweight Object Detection Based on the Improved YOLO Algorithm [J]. Proceedings of the 2021 5th International Conference on Electronic Information Technology and Computer Engineering, 2021.

[5] Gao P, Ji C-L, Yu T, et al. YOLO-TLA: An Efficient and Lightweight Small Object Detection Model based on YOLOv5 [J]. ArXiv, 2024, abs/2402.14309.

[6] Gong W. Lightweight Object Detection: A Study Based on YOLOv7 Integrated with ShuffleNetv2 and Vision Transformer [J]. ArXiv, 2024, abs/2403.01736.

[7] Xiong Y, Liu H, Gupta S, et al. MobileDets: Searching for Object Detection Architectures for Mobile Accelerators [J]. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020: 3824-33.

[8] Yu G, Chang Q, Lv W, et al. PP-PicoDet: A Better Real-Time Object Detector on Mobile Devices [J]. ArXiv, 2021, abs/2111.00902.

[9] Gao Y, Zhuang J-X, Li K, et al. DisCo: Remedy Self-supervised Learning on Lightweight Models with Distilled Contrastive Learning [J]. ArXiv, 2021, abs/2104.09124.

[10] Jiang Y, Tan Z, Wang J, et al. GiraffeDet: A Heavy-Neck Paradigm for Object Detection [J]. ArXiv, 2022, abs/2202.04256.