Mobilenetv2-SSD Target Detection Method Based on Multi-scale Feature Fusion

Most of the deep learning networks used in target detection algorithms are very complex, which poses difficulties for edge devices with limited storage capacity and computing power, such as mobile terminals. A MobileNetv2-SSD target detection algorithm based on multi-scale feature fusion is proposed in this paper. We use MobileNetv2 to improve the SSD algorithm by replacing the original network’s backbone network VGG16 with the lightweight network MobileNetv2 to perform the model streamlining. The problem of accuracy loss due to model compression is improved by the designed feature fusion module and finally tested on the dataset. The experimental results show that the mAP and FPS are 90.37% and 50FPS, respectively, and the detection accuracy is improved by 3.24% compared with the original SSD network, the detection speed is 1.9 times of the original one, and the size of the model is only 28.5M, which is only 30.5% of the original model.