Method for extracting video satellite data identification characteristics on line

Abstract

The invention discloses a method for extracting video satellite data identification features online, and relates to the technical field of video satellite data image processing. The method for extracting the identification features comprises the following steps: acquiring an image of video satellite data; and performing data annotation on the obtained image, then taking an annotation area after data annotation as an identification feature extraction range, training a symmetric semantic segmentation model online, and extracting identification features by adopting the trained symmetric semanticsegmentation model and an auto-encoding network model. According to the method, the pixel information loss caused in the forward propagation process is effectively reduced, and the segmentation boundary of the image is accurately positioned; meanwhile, the self-encoding network model automatically learns features from the unlabeled data; after semantic information of middle and high layers is obtained by forward propagation of video satellite data through the neural network, semantic information compression and reduction can be performed by means of the self-encoding network to obtain clearerfeature description, so that the difference between different backgrounds is expanded, and a better background segmentation effect is achieved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for online extraction of identification features of video satellite data. Background technique [0002] In the field of machine learning and computer vision applications, it is generally believed that there are two mainstream ways for machines to understand the world: "similarity" and "discrimination". Although in the prior art, deep learning theory has been used to detect and process terrestrial video images, there have been mature applications. However, due to the fundamental differences between video satellite data and terrestrial video data, including different shooting angles, different observation angles, different sensor platforms, different spatial resolutions, different imaging methods, and different image contrasts, it is impossible to use images in computer vision applications. The processing method is directly applied to the video satellite. [000...

AI Technical Summary

Problems solved by technology

[0004] The object of the present invention is to provide a kind of method of online extraction video satellite data identification feature, thus solve the foregoing problems existing in the prior art

Images