Target detection network construction method and device, target detection method

Abstract

The present application provides a method and device for constructing a target detection network, a method and device for target detection, an electronic device, and a computer-readable medium. The target detection network construction method includes: obtaining an initial network model for target detection, the initial network model including a connected backbone network and a multi-task learning module; adding a clipping module and a new classification branch network in the multi-task learning module, to get the target detection network. In this scheme, a clipping mechanism based on the anchor frame is added, and a new classification branch network is added at the same time, the classification information of the original anchor frame and the classification information obtained by the new scale anchor frame in the new classification branch network are combined to obtain Better classification results can be well applied to deep learning-based target detectors, which can better solve the multi-scale processing problem and the bottleneck problem of multi-label classification in deep target detectors, thereby improving the accuracy of target detection algorithms Spend.

Description

technical field [0001] The present application relates to the field of computer vision, in particular to a method and device for constructing a target detection network, a method and device for target detection, an electronic device, and a computer-readable medium. Background technique [0002] Object detection is an important subfield of computer vision tasks. Its task is to locate all objects from an image and accurately classify and recognize these objects. In recent years, with the rapid development of deep learning, the target detection method based on deep learning has strong performance and has become the most advanced method in the field of target detection. [0003] The existing target detection method based on deep learning is a multi-task learning algorithm, which first generates an anchor box (anchor), and then the algorithm generally has two tasks to learn, one is to accurately return the position of the anchor box, The second is to accurately classify and iden...

AI Technical Summary

Problems solved by technology

The first problem is that the performance of the algorithm will be constrained by the scale problem, because the anchor boxes are all generated on the feature map generated in the middle of the neural network convolution process, and compared with the original image, the size of the feature map has been reduced by several times, so the receptive field will be very large, so the generated anchor box will inevitably deviate from the target position of the original image; the second problem is that the relationship between targets will become a bottleneck for classification performance, because there will be many targets in the picture , the relationship between targets is very important information, but this information is ignored in the current target detection algorithm, and only a single classification label is used to classify each anchor box, which will restrict the performance improvement of the detection algorithm

Images