Acoustic Based Drone Detection via Deep Learning

Project Title

Acoustic Based Drone Detection via Deep Learning

Project Area of Specialization Artificial IntelligenceProject Summary

In the last few years, Aerospace Industry has seen an increased influx of technology known as ‘Unmanned Aerial Systems (UAS)’. These include all types of Aerial Vehicles that do not have an onboard physical (living) presence controlling them, and are either controlled remotely or are self-controlled e.g., military drones and hobby-drones, UAVs, space shuttles, agro-drones, etc. This UAS technology has garnered huge attention lately, primarily because it has become a threat to general peace. Apart from its use in agriculture for pest control; meteorology and weather forecast, entertainment, and fashion industry – its use for military purposes is frightening and lethal. It is the recent use of ‘Loitering Munition’ and ‘Kamikaze Drones’ round the world, a sub-category of UAS, that makes a ‘Counter-Unmanned Aerial System (C-UAS)’ a dire need as of today.This proposal focuses on only the drone and UAV sub-category of UAS. It concerns itself with developing an Acoustic-based C-UAS that can detect, localize and identify a drone/ UAV in midair using acoustic arrays, machine learning and data processing techniques.

Project Objectives

To localize the drone in midair and gather info on its angular position with respect to time using the conventional beamforming technique.  Form a database of drone signatures and noise recordings to feed our machine learning algorithm, Support Vector Machine. ? To form a filter that could differentiate the drone signals from acoustic signatures. ? To obtain a reasonable accuracy in a closed environment.

Project Implementation Method

Selection of the best MEMS Microphones to form an acoustic array. 2. Understanding the use of Conventional Beam-forming Technique and Overlap-Add Technique. 3. Understanding of Machine Learning Algorithm, Support Vector Machine. 4. Using this gained insight of these techniques, plot an angular position as a function of time for the incoming drone acoustic signature. Then convert the 3-Dimensional signal to a 1-Dimesional Signal. 5. Provide this signal to the Machine Learning Algorithm for drone signal differentiation from noise.

Benefits of the Project

To Pakistan Airforce: In locating and identifying unwanted drones and then developing a mitigation system (to possibly intercept/ hijack/ destroy target) once the target’s coordinates are known through this system. ? Cost effective sensing system compared to radar, radio-frequency, vision based and data fusion techniques.

This system can endlessly evolve into better sensing system because of the inherent capability of machine learning. The more cast the training set we provide, the better the results we get. ? Once sufficient data is achieved, the machine learning algorithms will be superseded by deep learning algorithms like Convolutional Neural Networks, which will then have potential to become Artificial Intelligence, an AI Sensing Technique. ? This technology may be modified to detect any aerial and/ or ground vehicle that makes noise and find use in every field, not just in military.

Technical Details of Final Deliverable

Arrange an acoustic array, record different drone signals, process them for use as a data base. 2. Then start by forming a training set that will be used to train our machine learning algorithm. 3. This algorithm will then identify any signal as drone or noise.

Final Deliverable of the Project Hardware SystemCore Industry EducationOther Industries Others Core Technology Artificial Intelligence(AI)Other Technologies Internet of Things (IoT)Sustainable Development Goals Quality EducationRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	80000
Respeaker Mic Array v2.0	Equipment	2	20000	40000
Raspberry Pi 3 B+	Equipment	1	30000	30000
Cables	Miscellaneous	7	1000	7000
SD Card 512Gb	Miscellaneous	1	3000	3000