SBCS-Net: Sparse Bayesian and Deep Learning Framework for Compressed Sensing in Sensor Networks.

Area of Science:

• Signal Processing
• Machine Learning
• Sensor Networks

Background:

• Compressed Sensing (CS) is vital for resource-constrained sensor networks.
• Traditional CS methods struggle with low sampling rates and noise.
• Deep learning CS models show promise but often fail with complex noise.

Purpose of the Study:

• To propose SBCS-Net, a novel framework for robust signal reconstruction in sensor networks.
• To address limitations of existing CS methods in handling noise and low sampling rates.
• To improve signal reconstruction accuracy and stability in challenging sensor network environments.

Main Methods:

• Developed SBCS-Net, integrating sparse Bayesian compressed sensing (SBL) with CNNs and Transformers.
• Optimized key SBL parameters via end-to-end learning for adaptive sparsity and noise processing.
• Leveraged deep learning for feature extraction and global context modeling.

Main Results:

• SBCS-Net demonstrated superior reconstruction accuracy and visual quality compared to mainstream methods.
• The framework exhibited excellent robustness under extremely low sampling rates and strong noise.
• Experimental validation included benchmark datasets, noise tests, ablation studies, and statistical significance tests.

Conclusions:

• SBCS-Net offers an effective solution for high-fidelity, robust signal recovery in sensor networks.
• The proposed method significantly advances the capabilities of deep learning in compressed sensing.
• SBCS-Net provides a stable and accurate approach for challenging real-world sensor network applications.