Fixed star classification method

Through the stellar luminosity feature fusion network, combined with the DES-SE module and CBAM attention mechanism, the luminosity data weight is dynamically adjusted, which solves the problem of low accuracy of star classification when spectral data is missing, and achieves efficient and accurate star classification.

CN120561650APending Publication Date: 2025-08-29NANJING TECH UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510699492.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-28
Publication Date
2025-08-29

AI Technical Summary

Technical Problem

Existing deep learning methods have limitations in the feature mining of photometric data, data imbalance processing, and the stability and robustness of the model, which affects the accuracy of stellar classification. Especially when spectral data are missing, the accuracy of existing methods is not high.

Method used

The stellar photometric feature fusion network is adopted, including DES-SE module, feature extraction module, feature fusion module and output module. Through star map alignment, visual enhancement and image cropping preprocessing, the photometric data is converted into a five-channel Numpy array, and the DES-SE module, depth separation convolution and CBAM attention mechanism are used to dynamically adjust the band weights, and feature extraction and fusion are combined with multiple SPFFN blocks and upsampling mechanisms.

Benefits of technology

It improves the accuracy and robustness of stellar classification, enhances the expression ability of photometric data, reduces the computational complexity, and provides an efficient and accurate stellar classification solution.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120561650A_ABST
    Figure CN120561650A_ABST
Patent Text Reader

Abstract

The invention discloses a fixed star classification method, which belongs to the technical field of fixed star classification, and comprises the following steps: acquiring fixed star luminosity data and preprocessing, and mapping luminosity information of u, g, r, i and z wavebands to a five-channel Numpy file to obtain a fixed star luminosity Numpy array; inputting the fixed star luminosity Numpy array into the trained fixed star luminosity feature fusion network, and outputting a fixed star classification result; the fixed star luminosity feature fusion network comprises a DES-SE module, a feature extraction module, a feature fusion module, a GAP layer and an output module. The fixed star luminosity feature fusion network firstly converts a fixed star luminosity Numpy array into a weighted feature map through a DES-SE module; extracting features through a feature extraction module; fusion is carried out through a feature fusion module; inputting a GAP layer to carry out global average pooling; and finally, outputting a fixed star classification result through an output module. The method can solve the problem that when spectral data are missing and luminosity data are used for fixed star classification, an existing method is not high in accuracy.
Need to check novelty before this filing date? Find Prior Art