CBN-VAE: A Data Compression Model with Efficient Convolutional Structure for Wireless Sensor Networks.

Area of Science:

• Computer Science
• Electrical Engineering
• Machine Learning

Background:

• Wireless Sensor Networks (WSNs) face high communication energy consumption.
• Existing neural network compression methods often overlook computational demands and WSN applicability.
• There is a need for efficient compression models tailored for WSN constraints.

Purpose of the Study:

• To propose a novel, computationally efficient neural network data compression model for WSNs.
• To reduce both communication energy and computational load on WSN nodes.
• To enhance the applicability of neural networks in resource-constrained WSN environments.

Main Methods:

• Developed the CBN-VAE model, integrating Convolutional Neural Network (CNN) feature extraction with Variational Autoencoder (VAE) and Restricted Boltzmann Machine (RBM) data generation.
• Introduced a Downsampling-Convolutional RBM (D-CRBM) to replace standard convolutions, minimizing parameters and computation.
• Utilized a VAE composed of D-CRBM layers for learning data features, enabling compression and reconstruction.

Main Results:

• CBN-VAE achieved a 73.88% reduction in parameters and a 96.43% reduction in Floating-Point Operations (FLOPs) compared to CNNs of similar size, with negligible accuracy loss.
• Demonstrated significant reduction in node communication energy consumption by 95.83% on real-world WSN datasets.
• Achieved high Signal-to-Noise Ratio (SNR) of 32.51 dB and low reconstruction error (0.0678 °C) for Intel Lab temperature data.
• Exhibited robust fault detection and anti-noise capabilities, effectively avoiding faulty and noisy data during reconstruction.

Conclusions:

• The proposed CBN-VAE model is highly suitable for WSN applications due to its efficiency and low computational requirements.
• CBN-VAE offers superior compression and reconstruction accuracy compared to traditional methods.
• The model demonstrates practical utility in WSNs, enhancing data processing and reducing energy expenditure.