Research on Target Detection Technology Based on SSD

As a single-stage object detection model, SSD(Single Shot MultiBox Detector) has been widely used for its advantages such as fast speed and good scalability. However, after multi-layer feature extraction, the deep network contains less small target information, resulting in the absence of model detection. When training deep networks, it is easy to have the problem of gradient disappearing and gradient explosion. Therefore, this article will use the ResNet(Residual Network) network instead of the VGG-16(Visual Geometry Group) network in the model to solve these problems. At the same time, through the method of feature fusion, the shallow features and deep features are fused, which enriches the information of different levels and improves the detection accuracy of the model. The experimental results show that the accuracy is improved by 2.2% after replacing VGG with ResNet network, and the accuracy of the model after feature fusion is improved by 4.95% compared with ResNet network.