An object detection algorithm based on the feature pyramid network and single shot multibox detector

摘要

Abstract In order to solve the problem of weak detection of small targets in traditional methods, an improved object detection algorithm is proposed. First, the six multi-scale feature maps extracted from the original SSD algorithm are fused in turn to form a new feature map with detailed information and semantic information based on the feature pyramid network and the idea of single shot multibox detector algorithm. Then, the attention model is added to the fused feature map, and the feature information of small targets can be extracted effectively. With PASCAL VOC2007 and VOC2012 as the training set, the mean average precision tested in the VOC2007 test set reached 78.3%, which is 1.1% higher than the original algorithms. In different environments, the algorithm has accurate detection effect on densely distributed small objects, and the missed detection and robustness are better than other algorithms. At the same time, the detection speed can still meet the real-time requirements.