Target detection method and device based on BSSD

Abstract

The invention discloses a target detection method and device based on a BSSD ( Bidirection Single Shot Multibox Detector). The BSSD network model is constructed based on the SSD network model; firstly, a relatively low high-level feature layer is fused with a feature layer, used for detecting a minimum target, of previous SSD through linear interpolation, and then a relatively high low-level feature layer is spliced with the fused feature layer through a passthrough method to obtain a feature layer used for detecting a small target; according to the improved BSSD, feature information extractedby each network layer can be fully utilized; the problem that the SSD does not use high-level feature semantic information is effectively improved, and the setting mode of the SSD default box and thesampling proportion in the data augmentation strategy during model training are further modified so as to improve the detection capacity of the model for small targets. Therefore, compared with the SSD, the improved BSSD has a better detection effect on small target detection and has better robustness.

Description

technical field [0001] The present invention relates to a target detection method and device based on BSSD (Bidirection Single Shot Multibox Detector). By fusing the high-feature layer and low-feature layer in SSD (Single Shot Multibox Detector), the setting method of SSD default frame is improved. Modification, and modifying the data augmentation strategy in the SSD training program can effectively improve the detection ability of the model for small targets, and belongs to the field of computer vision processing technology. Background technique [0002] Object detection is one of the hottest and most challenging research tasks in the field of computer vision. It has very important applications in real-world scenarios, such as face recognition, behavior recognition, automatic driving, intelligent video surveillance, etc. With the vigorous development of deep learning, the target detection algorithm based on deep learning has become the mainstream. It has the advantages of h...

AI Technical Summary

Problems solved by technology

[0005] Purpose of the invention: Aiming at the problems and deficiencies in the prior art, the present invention provides a BSSD-based target detection method and device, which makes full use of the position information of low-level features and the semantic information of high-level features, and can effectively improve the performance of SSD algorithms for small targets. The problem of insufficient detection capacity

Images