A depth bidirectional feature pyramid enhancement network for small scale target detection

Abstract

The invention relates to a depth bidirectional feature pyramid enhancement network for small-scale target detection, comprising: determining a backbone network of a network encoding end; determining adepth bidirectional feature pyramid enhancement network of a small-scale target detection end. Design Bottom-Up characteristic pyramid; Design Top-Down pyramid; Target Detection Subnetwork: Using strategies in detection of the two stages in faster-rcnn, wherein the two stages include candidate frame extraction and target classification respectively. In the RPN stage, regression of a target frameand prediction of whether the probability the target's are performed by use of the convolution of the convolution kernel of 3*3 on an output characteristic graph of each scale of the top-down characteristic pyramid. the screened candidate target frame and the output characteristic graph of the top-down characteristic pyramid of the corresponding scale are subjected to ROI-pooling, and finally twofull connection layers are used to adjust the frame and classify the specific categories of the target; Object Detection Results are output.

Description

technical field [0001] The invention belongs to the target detection technology in the fields of computer vision, pattern recognition, deep learning and artificial intelligence, and in particular relates to the technology of using a deep convolutional neural network in an image or video to detect a target in a scene. Background technique [0002] In the field of deep object detection, with the continuous improvement of object detection performance, the performance of small-scale object detection has become a new bottleneck, and some new network structures have been proposed to improve the problem of small-scale object detection. Feature pyramid network (feature pyramid network [1], referred to as FPN) is one of the representatives. FPN introduces the pyramid idea widely used in the traditional image processing field into the deep object detection architecture, and has achieved great improvement in large-scale object detection, especially the detection performance of small-sc...

AI Technical Summary

Problems solved by technology

However, most of the existing object detection methods have a good detection effect on large-scale objects, and the effect on small-scale object detection is not satisfactory.

Images