松弛图嵌入的判别宽度学习系统以及在视觉识别中的应用

The broad learning system, as a lightweight network, achieves a good balance between efficiency and accuracy. However, it primarily relies on strict binary labels for supervision and neglects local structural information during data transformation, limiting the model＇s performance. To address this issue, this paper proposes a relaxed-graph embedding discriminative broad learning system model and applies it to visual recognition, with the goal of enhancing model performance through the introduction of flexible labels and a relaxed-graph structure. The innovations of this paper are as follows: 1) We innovatively use double transformation matrices to construct the relaxed-graph, separating the responsibilities of the transformation matrix. This reduces the burden on the transformation matrix and allows for the learning of more flexible transformation matrix, thereby mitigating the overfitting problem; 2) We introduce a flexible label strategy that increases the distance between different categories labels, addressing the issue of strict binary labels, and thereby enhancing the model＇s discriminative ability; 3) An iterative optimization algorithm based on the alternating direction method of multipliers is proposed to achieve efficient model optimization. Extensive experiments on facial image datasets, object image datasets, scene image datasets and handwritten character image datasets demonstrate that the proposed model outperforms other advanced recognition algorithms.