Method for constructing target detection adaptive model based on CycleGAN and pseudo tag

Abstract

The invention discloses a method for constructing a target detection adaptive model based on a CycleGAN and a pseudo label. The method comprises the following steps: S1, preprocessing a source domaindata set and a target domain data set; S2, converting the source domain data set into an intermediate domain data set close to a target domain data set by using a CycleGAN network, and inputting the intermediate domain data set into a Faster RCNN network for training to obtain a preliminary domain adaptive model Q; re-inputting the target domain data set into the model Q to obtain a target domaindata set with a pseudo tag; and S3, inputting the intermediate domain data set and the target domain data set with the pseudo tag into the model Q in turn for iterative updating and optimization, andfinally obtaining a target detection domain adaptive model based on the CycleGAN and the pseudo tag. According to the method, the confidence is used to improve the target detection total loss functionof the Faster RCNN network to train the obtained target detection domain adaptive model, and the problem of domain drift of target detection due to distribution difference between two domains can besolved.

Description

technical field [0001] The invention relates to the technical field of deep learning, in particular to a method for constructing an adaptive model for target detection based on CycleGAN and pseudo-labels. Background technique [0002] Existing target detection methods based on deep neural networks (such as AlexNet, VGGNet, GoogleNet, and ResNet, etc.) can apply the learned model to the test set under the condition that the data distribution of the training set and the test set are strictly consistent. obtain higher detection accuracy. However, deploying the model trained by the training set in the actual natural scene, because the actual natural scene environment is often uncontrollable, such as the huge difference in object appearance, background, lighting, climate, image quality, etc., makes the difference between the two Differences in data distribution lead to a significant drop in detection accuracy when the model trained through the training set is used in the real wo...

AI Technical Summary

Problems solved by technology

[0003] The technical problem to be solved by the present invention is to provide a method for building an adaptive model for target detection based on CycleGAN and pseudo-labels for the problem of domain drift in target detection due to distribution differences between the two domains. Improve the target detection total loss function of the Faster R-CNN network to train a target detection domain adaptive model based on CycleGAN and pseudo-labels

Images