A Method of Generating Samples for Remote Sensing Image Target Detection Based on Traversing Source Targets

Abstract

The invention discloses a remote sensing image target detection sample generation method based on traversing source targets, using web crawler technology or artificial punctuation to obtain coordinate information of object elements to be detected within a certain range, and then generating corresponding sizes according to different remote sensing image resolutions The label frame of the target is used as the target label information, combined with the corresponding remote sensing image, by traversing the target one by one, the training sample pair of the remote sensing image target is generated with the size range of the training sample, and at the same time, the target in the sample is evaluated and screened by GIS spatial analysis, and finally automatically completed The training sample set of the object element to be detected is generated. The size of the sample set generated by the present invention is positively correlated with the number of marker elements, and does not depend on the size of the remote sensing image, so as to solve the problem of fewer samples generated by traversing the overall image, and, based on each target point, it can be more Conveniently combined with GIS spatial analysis and enhanced algorithms, it has better integrity guarantee for the targets in a single sample.

Description

A sample generation method for remote sensing image target detection based on traversing source targets technical field The invention belongs to the remote sensing image deep learning target detection field, be specifically related to a kind of remote sensing based on traversing the source target. Sensing image target detection sample generation method. Background technique [0002] With the rapid development of deep learning target detection technology, it has been successfully applied in the target detection of natural images. technologies such as face recognition, vehicle recognition, and pedestrian detection are improving day by day. The rapid development of remote sensing image technology for Earth observation It provides an important data source, and remote sensing image target detection is an important part of national defense security, precision agriculture, surveying and mapping, urban planning and other fields. The key technology of the deep learning target ...

AI Technical Summary

Problems solved by technology

Since remote sensing images are affected by factors such as resolution, band number, width range, and spectral range, the above open source remote sensing sample sets cannot cover all business application requirements in the field of remote sensing target detection. For specific business target detection, it is necessary to establish a corresponding Remote sensing image training set of

[0004] The production process of target detection training samples is usually based on manual labeling, which not only consumes a lot of human resources, but also is time-consuming and inefficient. It not only cannot meet the business applications of remote sensing target detection in time, but also increases the investment in the operation of remote sensing image target detection business. cost

Although a certain amount of training samples can be automatically generated by traversing the remote sensing images through the sliding window combined with the target vector data, this method will lead to training problems when targeting a small range of remote sensing images and fewer targets on the images. The number of sample sets generated is too small, and it will cause false samples containing a lot of background information, and the proportion of valid samples is low. Therefore, a more effective and feasible remote sensing target detection training sample generation algorithm is needed.

Images