A Lightweight Target Detection Network for Embedded Platforms
Yinbao Niu, Zibo Lyu, Xiyang Liu, Zhaochen Lin, Fenghua He · 2020
This paper studies a target detection algorithm based on deep learning which has to be implemented on embedded platforms with limited processing capability. First, the detection requirements are given. Then, a lightweight detection network is designed based on YOLOv3 in which a streamlined MobileNetV2 is used as the feature extraction network. To shorten the training time, a pre-trained model is adopted to initialize parameters. Finally, the test experiments are performed by operating the proposed algorithm on a smartphone. The results show that the algorithm can achieve a good performance both in accuracy and speed in a given environment under limited computing resources.