Weak supervision target detection method based on improved deep residual network

Abstract

The invention relates to a weak supervision target detection method based on an improved deep residual network. During the construction of a network model, a deep residual network is adopted as a backbone network, and a candidate region pooling layer and a redundancy self-adaptive neck network are combined between the backbone network and a weak supervision detection head network, so that the defect that an existing deep residual network is applied to weak supervision target detection is effectively overcome, and the performance of weak supervision target detection is greatly improved.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a weakly supervised target detection method based on an improved deep residual network. Background technique [0002] Object detection is a basic problem of machine vision, and it is widely used in video surveillance, unmanned driving and other scenarios. With the rise of deep learning, a large number of excellent object detection models have emerged in recent years. However, object detection results based on strongly supervised learning heavily depend on the accuracy of object annotations, which are easily affected by subjective judgments. [0003] With the continuous development of deep learning, the cost of object labeling becomes higher and higher. Training a high-accuracy detection model requires a large amount of finely labeled image data in the form of bounding boxes as model supervision conditions, which requires a lot of manpower and material resources. How to use low-c...

AI Technical Summary

Problems solved by technology

This is also the main reason why most weakly supervised object detection methods do not use deep residual networks

Images