Star-ground cooperative based automatic updating method for spaceborne ground target recognition algorithm

Abstract

The application discloses a kind of star-ground cooperative based on spaceborne ground target identification algorithm automatic updating method, comprising: step 1, spaceborne computing unit obtains the multiple imaging of hot spot area, and image is sent into on-orbit data screening procedure;Step 2, using spaceborne deep learning target identification algorithm, target feature extraction is carried out to image;Step 3, after identifying the image containing suspected target, image is screened in combination with target template data statistics value;Step 4, select the target information most conducive to algorithm migration from the screened image;Step 5, the result after screening is transmitted back to ground through downlink;Step 6, ground system receives and accumulates satellite downlink data, and sends to high-performance cloud server;Step 7, on the server, execute semi-supervised learning method, use unmarked high-quality on-orbit real data and existing remote sensing data set on ground, complete algorithm model updating and upgrading under the migration learning framework of target detection algorithm model.

Images