Improving Localization in Wireless Sensor Networks for the Internet of Things Using Data Replication-Based Deep...

Area of Science:

• Computer Science
• Electrical Engineering
• Network Engineering

Background:

• Localization in wireless sensor networks (WSNs) is crucial for Internet of Things (IoT) applications.
• Machine learning (ML) algorithms struggle with limited training data, causing overfitting and reduced accuracy in WSN localization.
• Low sensor node counts exacerbate accuracy issues in WSN localization systems.

Purpose of the Study:

• To enhance the localization accuracy of WSNs using a novel data augmentation strategy (DAS).
• To address the challenges of limited training data and overfitting in deep neural network (DNN) based WSN localization.
• To improve the performance of DNNs in WSN localization, particularly in scenarios with a low number of sensor nodes.

Main Methods:

• Proposed an intelligent data augmentation strategy (DAS) integrated with a deep neural network (DNN).
• DAS replicates estimated node positions from the Dv-hop algorithm and adds Gaussian noise to create diverse datasets.
• Combined augmented datasets with original training data to increase dataset size and diversity.

Main Results:

• The data augmentation technique significantly reduced the normalized root mean square error (NRMSE).
• Improved DNN performance substantially compared to the traditional Dv-hop algorithm with a low number of nodes.
• Demonstrated efficient computational cost for the data augmentation process.

Conclusions:

• The proposed DAS-based DNN method offers a scalable and effective solution for WSN localization.
• Data augmentation is a viable strategy to overcome limitations of small datasets in WSN localization.
• The method enhances localization accuracy in WSNs without significant computational overhead.