Vehicle Intrusion Classification using Deep Learning and Simulated Sensor Networks

This paper presents a development of classification Deep Learning (DL)-based model for simulated vehicles intruding a sensor network. The study proposes a DL architecture that is capable of learning and classifying a set of six different vehicle classes including motorcycles, military SUV, trucks, tank, etc. The proposed DL architecture consists of number of layers including: input, convolution, activation, pooling, flatten, fully-connected, and soft-max layers. To train and validate the proposed model, a simulated sensor network was developed to detect intruding vehicles to a guarded area. Simulated sensors were deployed on large scale, detection patters are collected and fed to the DL model for training and validation. The experimental results showed that the proposed model was effective and reliable with an average accuracy of 95.41% and a highest accuracy of 98.31%. The primary outcome of this study is that simple large-scale deployment of simple sensors is effective in detecting and tracking objects within a sensor network. We believe the proposed model can be extended to different domains such as wildlife surveys and forest protection.