Pyramid Frequency Feature Fusion Object Detection Networks

摘要

<p indent=0mm>For the problem of the absence of detail texture and other high-frequency features in the feature extraction process of deep learning network employing the up-sampling operation, a pyramid frequency feature fusion object detection network is proposed with three deep learning pyramid networks, to balance the high and low frequency feature information and improve the detection accuracy. The deep feature of the input image is extracted from the primary pyramid, different frequency characteristics are formed respectively by the high and low frequencies enhancement pyramid. In the process of information transmission, feature fusion is used to highlight the detail information protection ability of deep learning network and improve the object detection capability. After the simulation test based on CornerNet algorithm framework, the detection effect of proposed algorithm on vague objects, overlapping objects and low contrast between the objects and the background is significantly increased. The detection results on COCO dataset are more than 1% higher than CornerNet algorithm, so the proposed algorithm has a good performance in detecting pedestrians, vehicles and other objects, which is able to application autonomous vehicle systems and smart robots.