Optical remote sensing image recognition method, device, equipment and storage medium

By constructing the photovoltaic panel index and cascade processing, combined with the target photovoltaic panel recognition model, the problem of low accuracy in photovoltaic panel image recognition in the existing technology is solved, and higher recognition accuracy and image integrity are achieved.

CN117671513BActive Publication Date: 2025-03-25CHINESE PEOPLE'S PUBLIC SECURITY UNIVERSITY
1 Cites 0 Cited by

Patent Information

Application Number
CN202311568844.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-11-22
Publication Date
2025-03-25
Estimated Expiration
2043-11-22

AI Technical Summary

Technical Problem

The prior art has low accuracy when identifying photovoltaic panel images, which is easy to be confused with other non-photovoltaic panel images, and deep learning algorithms may lose underlying feature information.

Method used

By acquiring the optical remote sensing source image to be identified, the photovoltaic panel index is constructed based on the optical characteristics and reflection characteristics of the photovoltaic panel in each optical band, and cascades it with the image to be identified, and the target photovoltaic panel recognition model is used for identification. The model is trained from the basic double-layer nested network model, residual convolution hybrid attention module, multi-level feature loss function, and historical photovoltaic panel sample set.

Benefits of technology

The accuracy of identifying photovoltaic panel images is effectively improved, and the extracted photovoltaic panel images have high completeness and obvious boundary properties.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The present invention discloses an optical remote sensing image recognition method, device, equipment and storage medium, the method comprising: obtaining an optical remote sensing source image to be identified; constructing a photovoltaic panel index according to the optical characteristics and reflection characteristics of the photovoltaic panel in each light band; cascading the photovoltaic panel index and the optical remote sensing source image to be identified to obtain the object to be identified; identifying the object to be identified by a target photovoltaic panel recognition model to obtain a target photovoltaic panel image, wherein the target photovoltaic panel recognition model is trained by a basic double-layer nested network model, a residual convolution mixed attention module, a multi-level feature loss function and a historical photovoltaic panel sample set; in the above manner, after the photovoltaic panel index is constructed, it is cascaded with the optical remote sensing source image to be identified, and then the object to be identified is identified by the target photovoltaic panel recognition model, thereby effectively improving the accuracy of identifying the photovoltaic panel image, and then extracting photovoltaic panels with high integrity and obvious boundaries.
Need to check novelty before this filing date? Find Prior Art