Neural structure search weighted feature pyramid network for target detection

Abstract

The invention discloses a neural structure search weighted feature pyramid network for target detection. The weighted feature pyramid network is composed of merging units. Construction of the weighted feature pyramid network is based on FPN, the last layer in each group of feature layers is used as input of the first pyramid network, and input of the next pyramid network is output of the current pyramid network; the merging units are generated by adopting an RNN (Recurrent Neural Network), and the RNN selects any two input layers, combines the input layers to generate an output feature layer and applies the output feature layer to binary operation; the binary operation comprises weight summation and global pooling. The neural structure search weighted feature pyramid network can be stopped at any time, an optimal structure can be obtained through automatic design, feature pyramid generation quitting in advance is allowed to be used for detection at any time, and then the method can be used for connecting a classifier and frame regression to train a network, so that resource and time management is easier, and a solution is provided for dynamic resource allocation; and higher detection precision can be achieved.

Description

technical field [0001] The invention relates to the field of communication technology, in particular to a neural structure search weighted feature pyramid network for target detection. Background technique [0002] Object detection is a research hotspot in the field of computer vision. It is widely used in many fields such as face recognition, industrial inspection, and aerospace. It is of great significance to reduce the consumption of human capital through computer vision. The task of target detection is to find all the targets of interest in the image, determine their categories and positions, that is to say, it includes two subtasks of target positioning and target classification. Object detection is a fundamental problem in computer vision and a fundamental task of various video surveillance techniques. In recent years, with the rapid development of deep convolutional networks, object detection has made great progress in the design of model structures. However, due to...

AI Technical Summary

Problems solved by technology

[0010] The technical problem to be solved by the present invention is that most convolutional neural structures are manually designed so that the optimal structure cannot be obtained, and the accuracy of target detection is improved.

Images