Adaptive Deep Cascade Broad Learning System and Its Application in Image Denoising.

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Broad Learning Systems (BLS) offer efficient learning but can be improved for complex tasks.
• Image denoising remains a critical challenge in image processing, impacting visual quality and subsequent analysis.
• Existing methods may struggle with extracting sufficient information from raw image data.

Purpose of the Study:

• To introduce a novel regularization deep cascade broad learning system (DCBLS) architecture.
• To enhance feature representation extraction for improved image denoising performance.
• To develop a parallelizable framework compatible with large-scale datasets.

Main Methods:

• Proposed a DCBLS architecture with cascaded feature mapping and enhancement node layers.
• Integrated enhancement and feature mapping nodes for transformation feature representation.
• Employed convex optimization for the final output layer construction.
• Designed a parallelization framework for scalability and incorporated adaptive regularization parameters.

Main Results:

• The DCBLS effectively extracts more information from raw data compared to standard BLS.
• Achieved significant success in image denoising tasks, including natural and hyperspectral images.
• Experimental results demonstrate superior performance against state-of-the-art denoising methods.

Conclusions:

• The proposed DCBLS architecture offers a powerful and effective solution for image denoising.
• The method's parallelizability and enhanced feature extraction contribute to its superiority.
• Validated effectiveness and advantages through rigorous experiments on benchmark datasets.