A Space Debris Recognition Method Based on Multimodal Feature Fusion

Aiming at the problems of traditional space debris image recognition models, including incomplete coverage of training datasets, insufficient sample size, severe noise interference in space observation images, and failure to comprehensively utilize multimodal information, this paper proposes a space debris traceability and identification method based on multimodal feature fusion.To tackle dataset-related issues, hypervelocity impact experiments are carried out to reproduce the space collision environment, and the first three-dimensional database of space debris breakup under hypervelocity impact is constructed. On this basis, a multimodal full-life-cycle dataset covering visible light, infrared, X-ray and synthetic aperture radar (SAR) is generated, and a noise-adding scheme is designed to simulate the complex space observation environment.To alleviate noise interference, the DNCNN-B blind denoising model is employed for denoising preprocessing of noisy images. Combined with the Vision Transformer model, debris feature extraction and classification are implemented. In addition, an adaptive weighted fusion algorithm called the product entropy method is proposed, which raises the recognition accuracy of 29 categories of space debris on the noisy dataset from the initial 83.2\% to 99.11\%.Experimental results show that the synergistic effect of DNCNN denoising and the multimodal fusion strategy significantly improves the accuracy and robustness of recognition in complex noise environments. This study helps to enhance the capabilities of space debris traceability, breakup inversion and collision warning, and provides innovative technical support for space security governance and space traffic management.