Multimodel Sensor Fusion For Target Classification

Abstract

Unattended ground sensors (UGS) have proven to be valuable in various military missions. Specifically, UGS adds significantly to the capability and security of reconnaissance and surveillance units during military operations by monitoring the area. The classification of the vehicles as tracked or wheeled also plays a vital role in border monitoring. Apart from classification accuracy, the other essential parameters for UGS are timely classification and low power. Generally, acoustic and seismic sensors are used for the classification of vehicles. Analysis from a single sensor is insufficient to meet these requirements. newlineThe main objective of this research study is to develop feature extraction and fusion algorithms to early classify the two types of vehicles accurately with reduced space and time complexities. In this research work, an attempt is made to develop algorithms to address these challenges using various sensor fusion algorithms, machine learning, and deep learning models. The design and development of the proposed frameworks are carried out on numerical computing tools (i.e., MATLAB and Python), and classification accuracy is considered as the critical metric for the performance evaluation. The overall contribution of the proposed research study is carried out in multi-fold implementation. newlineThe first contribution is to evaluate vehicle classification problems considering different domains like Time, Frequency, and Time-Frequency for signal analysis. This study is carried out with single sensor features extracted from these domains, and classification accuracy is evaluated using K-Nearest Neighbor (KNN) and Symbolic Dynamic Filtering (SDF). The research study in this phase is carried out with a single sensor, either acoustic or seismic. However, it is required to improve the accuracy for real-time implementation of vehicle classification. The next phase of the study includes the classification evaluation using the fusion approach. newline