Method and device for generating sample data set by online labeling remote sensing image based on crowdsourcing mode

Abstract

The invention relates to a method and a device for generating a sample data set by online labeling remote sensing images based on a crowdsourcing mode. The method comprises the following steps: dividing a remote sensing image labeling crowdsourcing task into a plurality of sub-tasks; reading remote sensing image region data corresponding to each sub-task as a remote sensing image sub-graph, and extracting to-be-labeled vector polygons from the remote sensing image sub-graph by adopting a machine learning model; distributing the sub-tasks to users participating in the crowdsourcing task, executing the sub-tasks, and obtaining semantic labeling results of the users on the to-be-labeled vector polygons in the sub-tasks; and superposing and merging a prediction result of an unlabeled region of a remote sensing image obtained by adopting the machine learning model and a labeled vector polygon region to generate a remote sensing image labeled sample data set. According to the method, the workload of crowdsourcing online labeling can be greatly reduced, it can be ensured that all sub-tasks can be executed, and the burden of online visual rendering can be relieved.

Description

technical field [0001] The invention belongs to the online labeling process of remote sensing images in the field of remote sensing images, relates to the production of remote sensing image labeling sample data sets by crowdsourcing, and is mainly used in the training of remote sensing image semantic segmentation deep learning models and other applications. Background technique [0002] The rapid development of remote sensing technology makes it more convenient to obtain images covering the whole world, and the improvement of image resolution makes the amount of image data increase geometrically. When using remote sensing images to analyze land cover, resources and environmental protection, it is necessary to interpret remote sensing images. The automatic interpretation of remote sensing images needs samples to improve the interpretation ability and accuracy, and the interpretation results also need samples to verify. The traditional method of collecting remote sensing imag...

AI Technical Summary

Problems solved by technology

This method is not only time-consuming and costly, but also limited by the time and energy of remote sensing image information extraction personnel, as well as the familiarity with the performance characteristics of different land use types in remote sensing images, it is often difficult to expand the amount of data available for training samples

Images