Border Trespasser Classification Using Artificial Intelligence

Mohsen Othmani,Tahar Ezzedine,Qing-Guo Wang,Mohamed Hechmi Jeridi

doi:10.1109/access.2021.3079702

Abstract

Monitoring the border is a very important task for national security. Wireless sensor networks (WSN) appear well suited in this application. This work aims to monitor a large-scale geographical framework that represents the borders of countries. Researchers take the Tunisian Algerian border as an example. This border is labeled by the illegal passage of intruders between the two countries. The task is to identify the intruders and study their kinematics based on speed, acceleration, and bearing. The appropriate types of sensors are determined according to the nature of intruders. Six classification techniques are compared which are: Naïve Bayes, Support Vector Machine (SVM), Multilayer Perceptron, Best First Decision Tree (BF-Tree), Logistic Alternating Decision Tree (LAD-Tree), and J48. The comparison of the performance of the classification techniques is provided in terms of correct differentiation rates, confusion matrices, and the time taken to build each model. Four different levels of cross-validation are used to validate the classifiers. The results indicate that J48 has achieved the highest correct classification rate with a relatively low model-building time.

Highlights

Monitoring of border areas can be realized with Wireless sensor networks (WSN)
The standard decision tree classification process is triggered by placing a root node we divide the learning instances into as many subsets as branches extending from the root node
The analysis of the confusion matrices relating to the different classification approaches (Naïve Bayes, Support Vector Machine (SVM), Multilayer Perceptron, BF Tree, LAD Tree, and J48), shows that the best overall accuracy with a rate equal to 88.022% is on the side of the J48 algorithm

Summary

INTRODUCTION

Monitoring of border areas can be realized with WSN. due to the low cost, no maintenance convenience, absence of complex equipment requirements, etc. . . , the WSN is very suitable for monitoring in poor conditions or large area. Execute the classification procedure to obtain a prediction model This model is used to estimate for each given sensor node S and class Ci the membership of S in class Ci. In this study, the region of interest in the general framework is the border zone between neighboring countries, researchers focus on the border area between Tunisia. B. KINEMATIC MODEL Our border monitoring focus on three targets classes: unarmed person, a person wearing metal (Soldier) and an ACP that transport a mass of ferrous. In the design phase of WSNs. Based on the study [15] on the design of the detection device, researchers can classify a sensor according to several points of view: Power supply, output type, and measurement properties. As for the unarmed person who is not detected, we choose the radar sensors

CLASSIFICATION

NAÏVE BAYES

MULTILAYER PERCEPTRON

METRICS

Findings

CONCLUSION