Target Detection Method Based on Improved Quadratic Feature Fusion

摘要

For the requirement of accuracy and speed in target detection, an improved secondary feature fusion based on SSD target detection method is proposed. On the basis of SSD network, FPNSSD network is constructed to enhance the semantic information of feature map. Secondary feature fusion is performed on the foundation of FPNSSD network. TFFSSD (twice feature fused SSD) network is proposed to further enhance the ability of low-level semantic information extraction and enrich semantic information. By introducing dichotomous K-means to cluster the selection of default box in TFFSSD network, optimizing the number of channels in prediction layer and using top-down modulation mode of TDM to improve the network, the final improved SSD network model is obtained. The experimental results show that the improved SSD network has better performance in detection accuracy and detection time.