Detection and Classification UAVs Radio Frequency Signals Using Artificial Intelligence (AI)

Abstract

The use of drones for surveillance purposes is quite common these days. The drones are widely used by military services or secret services to monitor their targets. But the presence of such vehicles or drones can hinder the security and can sometimes cause a great loss of lives for the country. So, detecting drones is quite a major problem that needs to be addressed. Security agencies are constantly looking for the latest technologies to address this issue of breached security. Artificial Intelligence has been a boon and bane to humans. But, using AI carefully, it can cause wonders to human lives. In this report, we have tried to incorporate AI to solve three tasks which are drone detection, type of drone identification, drone mode identification. We have used a publicly available Radio Frequency Database (RF) for drone detection. Deep learning has been on the top for a decade due to the high advancement in computers and processors. Extensive experimentation has been done using Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN) and Long Short-Term Memory Networks (LSTM). The raw RF data is preprocessed using MATLAB and the preprocessed data is used for training the neural networks. For each task, we have performed 4 individual experiments which are ANN, CNN, ANN with PCA and LSTM with Principle Components Analysis (PCA). Since the processed data is high dimensional, we have also used PCA to reduce data dimension. CNN based approach works best for all the three tasks giving an accuracy of 99.97 % for drone detection, 86 % for drone identification and 50 % for drone mode identification. ANN based approach performs little less than the CNN based approach with accuracies of 99.78 % for drone detection, 85 % for drone identification and 51 % for drone mode identification. While the use of PCA does not perform well as it might have caused data loss. Hence, PCA with ANN and PCA with LSTM performed poorly. All the experiments have been carried out on Google Colab platform with Keras framework and python language.