18-08-2023 дата публикации
Номер: CN116612004A
Принадлежит:
The invention discloses a hyperspectral image reconstruction method based on dual-path fusion. Firstly, a training set and a test set are constructed by preprocessing a data set; and then constructing a reconstruction model with a double-path structure, including shallow feature extraction, double-path depth feature extraction, and adaptive feature fusion and reconstruction. And optimizing network parameters by minimizing loss, and storing an optimal reconstruction model weight. And finally, inputting the RGB image into the optimal reconstruction model, and outputting a generated hyperspectral image. According to the method, a double-path structure is designed in feature extraction according to the features of the spatial local similarity and the spectral self-similarity of the hyperspectral image, so that targeted extraction is realized, the reconstructed image is clearer, and the texture is more obvious. Besides, the self-adaptive fusion module can measure the importance of the spectral ...
Подробнее