Single-stage target detection method using anchor-frame-free module and enhanced classifier

Abstract

The invention discloses a single-stage target detection method using an anchor-frame-free module and an enhanced classifier. The single-stage target detection method comprises the following steps: embedding the anchor-frame-free module; designing a decoupling control method; training an enhanced classifier is trained. The model is named as enhanced YOLOv3, and the enhanced YOLOv3 is established onthe basis of the YOLOv3. In order to solve the problem of low recall rate of YOLOv3, the invention provides a hybrid method, which combines an anchor-frame-free module with a YOLO prediction branch to generate more stable prediction. According to the invention, an enhanced classifier is provided, redundant bounding boxes are extracted step by step through multiple cascaded classifiers, the enhanced classifier can prevent some key bounding boxes from being eliminated before post-processing, and meanwhile error detection is restrained. On the basis, the invention provides a decoupling method tosolve the problem of inaccurate positioning, and provides a feature enhancement module to construct more stable feature representation.

Description

technical field [0001] The invention relates to the technical field of target detection, in particular to a single-stage target detection method using an anchor-free frame module and an enhanced classifier. Background technique [0002] In the past few years, object detection methods based on convolutional neural networks (CNN) have achieved remarkable progress in many applications, such as autonomous driving, face detection, and security monitoring. [0003] Previous mainstream object detection methods usually first generate anchor boxes, then perform bounding box classification and regression, and perform two-stage reasoning. Although these two-stage detectors exhibit superior performance, they still have some disadvantages. First, the properties of anchor boxes, such as size and aspect ratio, can greatly affect the accuracy of object localization. The second is that classification, regression and post-processing (NMS) in hundreds of bounding boxes incurs more computatio...

AI Technical Summary

Problems solved by technology

Although these two-stage detectors exhibit superior performance, they still have some disadvantages

First, the properties of anchor boxes, such as size and aspect ratio, can greatly affect the accuracy of object localization

The second is that classification, regression and postprocessing (NMS) in hundreds of bounding boxes incurs more computational cost

[0008] To sum up, the existing object detection techniques have the following defects: First, in the two-stage detector, the properties of the anchor box (such as size and aspect ratio) will greatly affect the accuracy of object localization sex

Moreover, classification, regression, and post-processing (NMS) in hundreds of bounding boxes will incur more computational costs; second, single-stage detectors often fail to achieve due to the extreme imbalance of positive and negative training samples. Higher accuracy; third, single-stage detectors have limited ability to handle small-sized objects, resulting in object omission in some dense scenes

Images