Unmanned aerial vehicle image labeling method assisted by deep learning target detection algorithm

Abstract

The invention provides an unmanned aerial vehicle image labeling method based on deep learning target detection algorithm assistance, and belongs to the field of unmanned aerial vehicle image processing. For two scenes of unmanned image annotation, different schemes are provided. A full-graph annotation scene includes: performing initial training on the detection network by using a small number ofother public unmanned aerial vehicle image data sets; inputting the unlabeled images into the network according to the sequence of the target numbers from small to large for forward deduction, automatic processing and manual correction, adding the group of images into the original data set for training again, so as to achieve better detection performance for the next group of images. An image partial region labeling scene includes: training the detection network by utilizing sub-images which are randomly cut from a labeled region in a data set and have similar sizes with an unlabeled region,and further labeling the unlabeled region. According to the method, manpower and material resources required by unmanned aerial vehicle image annotation are greatly reduced, and the annotation speed and precision are improved.

Description

technical field [0001] The invention belongs to the field of UAV image processing, and in particular relates to an UAV image tagging method assisted by a deep learning target detection algorithm. Background technique [0002] UAV image refers to the ground remote sensing image collected by the imaging platform on the UAV. A wide range of applications, including traffic flow monitoring based on deep learning and moving target tracking, etc. This also creates a broader need for annotated drone imagery datasets. [0003] Different from conventional scene images, UAV images are often large in size, cover a wide range of land surfaces, and have a large number of typical targets. Manually labeling a UAV image often takes several minutes, which requires a large amount of labeled training data. The time cost required in the task is unbearable. [0004] In recent years, the target detection method based on deep learning has gradually matured, and the detection accuracy has been gr...

AI Technical Summary

Problems solved by technology

At present, most of the annotations for UAV image data sets are manually annotated based on relevant annotation software. The existing annotation methods lack the use of a small amount of public UAV image data to train a detection network with good generalization and then assist unlabeled Annotating technology for UAV images

Images