Feature fusion block, convolutional neural network, pedestrian re-identification method and related equipment

Abstract

The invention discloses a multi-scale feature fusion block in combination with context information. The multi-scale feature fusion block comprises a forward hierarchical connection group, a backward hierarchical connection group and a channel multi-scale selection module, wherein the forward hierarchical connection group is used for progressive inter-scale information fusion; the backward hierarchical connection group is used for information fusion between spanning scales; and the channel multi-scale selection module is used for carrying out scale feature channel selection on the backward hierarchical connection group. The invention further discloses a convolutional neural network comprising the multi-scale feature fusion block in combination with the context information, so that effectivefusion of multi-scale features is realized. The invention further discloses a pedestrian re-identification method, device and equipment based on the convolutional neural network and a storage medium,and the pedestrian re-identification accuracy can be improved.

Description

technical field [0001] This application relates to the field of computer application technology, in particular to a feature fusion block, convolutional neural network, pedestrian re-identification method and related equipment. Background technique [0002] With the rapid development of computer technology, multi-scale feature extraction algorithms have gradually been widely used, such as in person re-identification (Person ReID / person re-identification (ReID)). Pedestrian re-identification is a subdivision problem in the field of image retrieval. The identity of pedestrians is authenticated by judging whether the images of several pedestrians appearing in multiple non-overlapping surveillance cameras are the same person. [0003] At present, multi-scale feature extraction algorithms are mostly implemented by stacking multiple multi-scale convolution branches or directly stacking multiple convolution kernels. [0004] Such as figure 1 As shown, Inception has designed multip...

AI Technical Summary

Problems solved by technology

[0005] This method of simple stacking is difficult to achieve efficient fusion of multi-scale features, and the accuracy of pedestrian re-identification is low.

Images