Hyperspectral image classification based on cross-domain adaptive broad learning system

Hyperspectral image classification is crucial for the identification of terrestrial objects in the field of remote sensing. The complexity and cost associated with annotating these images present significant challenges, as obtaining high-quality labeled samples is essential for the accuracy of classification. To address this challenge, this study proposes an innovative method for hyperspectral image classification, known as the Cross-Domain Mean Adaptive Broad Learning System (CDMABL). This approach initially employs cross-domain mean approximation within the feature mapping process of the Broad Learning System (BLS) to minimize the differences in marginal and conditional distributions between the source and target domains. To further enhance the model's generalization capabilities, domain adaptation terms are strategically integrated into the output layer of the BLS framework. Subsequently, ridge regression is utilized to optimize the model's output weights. The robustness and effectiveness of this algorithm are demonstrated through its successful application and validation on three distinct hyperspectral image datasets.