A Neural Network-Based Weighted Voting Algorithm for Multi-Target Classification in WSN.

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Wireless sensor networks (WSN) are crucial for monitoring applications.
• Accurate classification of mobile targets within WSNs is a significant challenge.
• Existing methods may lack the precision required for complex monitoring scenarios.

Purpose of the Study:

• To propose a novel neural network (NN)-based weighted voting classification algorithm for WSNs.
• To enhance the accuracy of mobile target classification in WSNs.
• To evaluate the performance of the proposed algorithm using real-world data.

Main Methods:

• Development of a NN-based weighted voting classification algorithm.
• Implementation using an "upper training, lower transplantation" approach on WSN nodes.
• Utilizing deep neural networks (DNN) and deep belief networks (DBN) as base classifiers.
• Verification with real-world experimental data.

Main Results:

• The proposed algorithm achieves an average classification accuracy of about 85% using DNN and DBN.
• It enhances target classification accuracy by approximately 5% compared to single NN-based classifiers.
• Achieved an 8.8% improvement over the Feedforward Neural Network (FFNN) classifier.
• Increased memory and computation time were observed as trade-offs.

Conclusions:

• The NN-based weighted voting algorithm significantly improves mobile target classification accuracy in WSNs.
• The method offers a viable solution for enhancing WSN monitoring capabilities.
• Further research may focus on optimizing computational overhead for practical deployment.