A cross-domain object detection method based on regional fully convolutional networks and self-adaptation

Abstract

The invention discloses a region full convolution network and self-adaptive cross-domain target detection method, which belongs to the technical field of computer vision. The present invention uses the deep learning target detection technology, aims at the problem of different distribution of data in the training domain and the test domain in the target detection, and uses an adaptive method to improve the cross-domain robustness of the target detection. First, a regional fully convolutional network model is constructed based on deep learning; then two corresponding domain classifiers are designed as adaptive components at the image level and target level to reduce the difference in domain transformation, and add consistency to the domain classifier Regularization; then train the network in an end-to-end manner; finally remove the adaptive component and use the network for object detection tasks. Adopting our designed cross-domain object detection method can effectively improve the average accuracy of object detection in various domain transformation scenarios.

Description

technical field [0001] The invention belongs to the technical field of computer vision, in particular to a cross-domain target detection method based on a fully convolutional network and self-adaptation. Background technique [0002] Object detection is a fundamental problem in computer vision, which aims to detect and identify all target objects in an image corresponding to a certain class. Object detection can be traced back to a long time, and there have been many classical and effective methods. Classic work usually defines object detection as a sliding window classification problem. In computer vision, the rise of deep convolutional networks originated from object detection. Driven by the rapid development of deep convolutional networks, researchers have proposed many target detection algorithms based on convolutional neural networks, which have greatly improved the performance of target detection. Among the large number of methods that have been proposed, regional f...

AI Technical Summary

Problems solved by technology

[0007] In order to solve the problem of cross-domain target detection, the present invention provides a cross-domain target detection method based on regional full convolution network and self-adaptation, using deep learning target detection technology, aiming at the problem of different distribution of data in the training domain and test domain in target detection , using an adaptive method to improve the cross-domain robustness of object detection

Images