Development of an Acoustic System for UAV Detection †
Abstract
:1. Introduction
- (1)
- Is it possible to build an audio detection, recognition, and classification system able to detect the presence of several drones in the environment, with relatively cheap commercial equipment (COTS)?
- (2)
- Assuming that it can function as a prototype, what challenges could be raised when scaling the prototype for practical use?
- (3)
- Are the techniques used in the development of the prototype drone detection, recognition, and classification superior in performance to existing commercial systems?
2. Related Work
3. Proposed Solution
3.1. Proposed Framework and System Architecture
3.2. Extracting the Features of the UAV-Generated Acoustic Signal
- (a)
- Filtering: The detector’s input sound needs filtering to get rid of unwanted frequencies. On the other hand, the filter must not affect the reflection coefficients. In the experiments an IIR notch adaptive filter has been used.
- (b)
- Segmentation: the acoustic signal is non-stationary for a long-time observation, but quasi-stationary for short time periods, i.e., 10–30 ms, therefore the acoustic signal is divided into fixed-length segments, called frames. For this particular case, the size of a frame is 20 ms, with a generation period of 10 ms, so that a 15 ms overlap occurs from one window to the next one.
- (c)
- Attenuation: Each frame is multiplied by a window function, usually Hamming, to mitigate the effect of finishing windows segmentation.
- (d)
- Mel Frequency Cepstrum Coefficients (MFCC) parameters: To recognize an acoustic pattern generated by the UAV, it is important to extract specific features from each frame. Many such features have been investigated, such as linear prediction coefficients (LPCs), which are derived directly from the speech production process, as well as the perceptual linear prediction (PLP) coefficients that are based on the auditory system. However, in the last two decades, spectrum-based characteristics have become popular especially because they come directly from the Fourier transform. The Spectrum-Based Mel Frequency Cepstrum coefficients are employed in this research and their success is due to a filter bank which make use of wavelet transforms for processing the Fourier Transform, with a perceptual scale similar to the human auditory system. Also, these coefficients are robust to noise and flexible, due to the cepstrum processing. With the help of the UAV sonic generated specific MFCC coefficients, recognition dictionaries for the training of neural networks are then shaped.
- (e)
- Feature Extraction for MFCC. The extraction algorithms of the MFCC parameters are shown in Figure 4. The calculation steps are the following:
- Performing FFT for each frame of the utterance and removing half of it.
- The spectrum of each frame is warped onto the Mel scale and thus Mel spectral coefficients are obtained.
- Discrete cosine transform is performed on Mel spectral coefficients of each frame, hence obtaining MFCC.
- The first two coefficients of the obtained MFCC are removed as they varied significantly between different utterances of the same word.
- Liftering is done by replacing all MFCC except the first 14 by zero.
- The first coefficient of MFCC of each frame is replaced by the log energy of the correspondent frame.
- Delta and acceleration coefficients are found from the MFCC to increase the dimension of the feature vector of the frames, thereby increasing the accuracy.
- Delta cepstral coefficients add dynamic information to the static cepstral features. For a short-time sequence C[n], the delta-cepstral features are typically defined as:
- Coefficients describing acceleration are found by replacing the MFCC in the above equation by delta coefficients.
- Feature vector is normalized by subtracting their mean from each element.
- Thus, each MFCC acoustic frame is transformed into a characteristic vector with size 35 and used to make learning dictionaries for feature training of concurrent neural networks (feature matching).
A set of 30 MFCC coefficient matrices was created for each drone, corresponding to drone flying the distances (0 to 25 m), (25 to 50 m), (50 to 100 m), (100 to 200 m) and (200 to 500 m). - (f)
- The Adaptive Filters. The role of the adaptive filter is to best approximate the value of a signal at a given moment, based on a finite number of previous values. The linear prediction method allows very good estimates of signal parameters, as well as the possibility to obtain relatively high computing speeds. Predictor analysis is since a sample that can be approximated as a linear combination of the previous samples. By minimizing the sum of square differences on a finite interval, between real signal samples and those obtained by linear prediction, a single set of coefficients called prediction coefficients can be determined. The estimation of model parameters according to this principle leads to a set of linear equations, which can be solved efficiently for obtaining the prediction coefficients.Equations (2) and (3) are considered:This means that the input signal is proportional to the error signal. Practically, it is assumed that the error signal energy is equal to that of the input signal:It should be noted, however, that for the UAV-specific audio signal if , it is necessary for the p-order of the predictor to be enough large so as to consider all the effects, eventually the occurrence of the transient waves. In the case of sounds without a specific UAV source, the signal s(n) is assumed to be white Gaussian noise with unitary variation and zero mean.
- (g)
- Time—Frequency Analysis. The analysis of the acoustic signals can be performed by one-dimensional or two-dimensional methods. One-dimensional methods involve that the analysis is made only in the time domain or only in the frequency domain and generally have low degree of complexity.
- (a)
- at any moment t, multiply the signal with the conjugate “mirror image”, relative to the moment of evaluation:
- (b)
- calculate the Fourier transform for the result of this multiplication, in relation to the offset variable τ.
3.3. Analysis of UAVs Specific Acoustic Signals Employing Cohen (Wigner-Ville) Energy Distributions
- The energy structure of the analyzed signals can be identified and located with a good accuracy in the time-frequency plane.
- When the type, duration, frequency, and temporal arrangement of the signals are not a priori known, they can be estimated using time-frequency distributions.
- The possibility of implementing these analysis algorithms in systems for analyzing the transient acoustic signals generated by the drones becomes thus available.
4. Employing Concurrent Neural Networks in UAVs Classification Process
4.1. Artificial Neural Networks and Wigner-Ville Spectrograms, MFCC, Mean Instantaneous Frequency (MIF) Classes
- Step 1.
- Create the database containing the training vectors obtained from the preprocessing of the acoustic signal.
- Step 2.
- The sets of vectors specific to each neural network are extracted from the database. If necessary, the desired outputs are set.
- Step 3.
- Apply the training algorithm to each neural network using the vector sets created in Step 2.
- Step 1.
- The test vector is created by preprocessing the acoustic signal.
- Step 2.
- The test vector is transmitted in parallel to all the previously trained neural networks.
- Step 3.
- The selection block sets the network index with the best answer. This will be the index of the class in which the vector is framed.
4.2. Multiple Drone Detection and Confusion Matrix
- (1)
- Concurrent Neural Networks (CoNN) with Wigner-Ville spectrogram class.
- (2)
- Concurrent Neural Networks (CoNN) with MFCC dictionary class.
- (3)
- Concurrent Neural Networks (CoNN) with MIF class.
4.2.1. Confusion Matrix
4.2.2. CoNN with Wigner-Ville Spectrograms
4.2.3. CoNN with MFCC
4.2.4. CoNN with MIF
5. Experimental Results
- ⮚
- Model DJI PHANTOM 4, type of classification—small (5)
- ⮚
- Model Homemade multirotor, type of classification—medium (2)
- ⮚
- Model DJI Matrice 600, type of classification—medium (1)
- ⮚
- Model Homemade Octocopter, type of classification—medium (3)
- ⮚
- Model Homemade Octocopter, type of classification—large (4)
- ⮚
- Parrot AR drone 2 (mini), type of classification—small (6)Figure 17. Results obtained for stationary distance 130 m, altitude 2 m.Figure 18. Results obtained for stationary distance 80 m, moving altitude 2–10 m.
- (1)
- Takes raw audio data and creates functions for each of the three mentioned networks
- (2)
- Run the data through each network and get an “answer” (a distribution of the probability of class predictions)
- (3)
- Select the “correct” output of the network with the highest response (highest class confidence)
- (4)
- This architecture being explained in Figure 7.
5.1. Computational Time
5.2. Comparison with Other Similar Methods
6. Discussion and Conclusions
- -
- Development of a spiral microphone array, combining microphones in the audible and ultrasonic fields, set in an interchangeable configuration with multichannel adaptive weights.
- -
- introduction of the possibility of detecting low intensity acoustic signals specific to multirotor mini drones, at a distance of ~120 m.
- -
- Dhe development of the training base with the Wigner-Ville spectrogram, MFCC dictionaries and MIF coefficients.
- -
- The use of multilayer perceptron modules, time delay neural networks and self-organizing maps.
- -
- The use of a set of networks equal to the number of classes in which the vectors are grouped, with a supervised preparation.
- -
- The recognition scheme consists of a collection of models trained on a subproblem and a module that selects the best answer.
- -
- Tests have shown that large multirotor (diameter 1.5 m) can be detected at a distance of ~500 m, and medium multirotor (diameter less than 1 m) can be detected at a distance of at least 380 m.
- -
- The possibility of integrating the microphone area in a network structure (scalability), which can be controlled by a single cRIO system by integrating several acquisition boards. The placement of the acoustic sensors within the network can be done linearly and in depth, so that a safety zone can be created around the perimeter restricted for the flight of drones.
7. Patents
Author Contributions
Funding
Conflicts of Interest
References
- Gartner. Available online: https://www.gartner.com/ro/documents/3984390 (accessed on 20 May 2020).
- Leszek, B.; Jerzy, S.; Zofia, W. Advances in Intelligent Systems and Computing. In Proceedings of the 40th Anniversary International Conference on Information Systems Architecture and Technology (ISAT 2019), Wroclaw, Poland, 15–17 September 2019. [Google Scholar]
- Nguyen, P.; Ravindranathan, M.; Nguyen, A.; Han, R.; Vu, T. Investigating Cost-effective RF-based Detection of Drones. In Proceedings of the DroNet’16, Singapore, 26 June 2016. [Google Scholar] [CrossRef]
- Trundle, S.S.; Slavin, A.J. Drone Detection Systems. U.S. Patent No. US 2017/0092.138A1; Application No. US 15/282,216, Publication of US 20170092138A1, 30 March 2017. [Google Scholar]
- Hearing, B.; Franklin, J. Drone Detection and Classification Methods and Apparatus. U.S. Patent No. US 9.275,645 B2; Application No: 14/258,304, Publication US 9275645B2, 1 March 2016. [Google Scholar]
- Byeong, H.K.; Danish, K.; Ciril, B.; Wonju, C.; Hyun, J.L.; Min, Y.K. V-RBNN Based Small Drone Detection in Augmented Datasets for 3D LADAR. Syst. Sens. 2018, 18, 3825. [Google Scholar] [CrossRef] [Green Version]
- Guvenc, I.; Koohifar, F.; Singh, S.; Sichitiu, M.L.; Matolak, D. Detection, Tracking, and Interdiction for Amateur Drones. IEEE Commun. Mag. 2018, 56, 75–81. [Google Scholar] [CrossRef]
- Laurenzis, M.; Hengy, S.; Hommes, A.; Kloeppel, F.; Shoykhetbrod, A.; Geibig, T.; Johannes, W.; Naz, P.; Christnacher, F. Multi-sensor field trials for detection and tracking of multiple small unmanned aerial vehicles flying at low altitude. In Proceedings of the Signal Processing, Sensor/Information Fusion, and Target Recognition XXVI, Anaheim, CA, USA, 2 May 2017. [Google Scholar] [CrossRef]
- Drozdowicz, J.; Wielgo, M.; Samczynski, P.; Kulpa, K.; Krzonkalla, J.; Mordzonek, M.; Bryl, M.; Jakielaszek, Z. 35 GHz FMCW drone detection system. In Proceedings of the 17th International Radar Symposium (IRS), Krakow, Poland, 10–12 May 2016; pp. 1–4. [Google Scholar]
- Busset, J.; Perrodin, F.; Wellig, P.; Ott, B.; Heutschi, K.; Ruhl, T.; Nussbaumer, T. Detection and tracking of drones using advanced acoustic cameras. In Proceedings of the Unmanned/Unattended Sensors and Sensor Networks XI; and Advanced Free-Space Optical Communication Techniques and Applications, Toulouse, France, 13 October 2015. [Google Scholar] [CrossRef]
- Liu, Y.; Wan, X.; Tang, H.; Yi, J.; Cheng, Y.; Zhang, X. Digital television based passive bistatic radar system for drone detection. In Proceedings of the IEEE Radar Conference, Seattle, WA, USA, 8–12 May 2017. [Google Scholar]
- Aker, C.; Kalkan, S. Using Deep Networks for Drone Detection. In Proceedings of the 14th Conference on Advanced Video and Signal Based Surveillance (AVSS), Lecce, Italy, 29 August–1 September 2017. [Google Scholar]
- Redmon, J.; Farhadi, A. YOLO9000: Better, faster, stronger. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 27–30 June 2016. [Google Scholar]
- Nielsen, R.O. Acoustic detection of low flying aircraft. In Proceedings of the IEEE Conference on Technologies for Homeland Security, Boston, MA, USA, 11–12 May 2009. [Google Scholar]
- Popovic, V.; Ott, B.; Wellig, P.; Leblebici, Y. Near-infrared high-resolution real-time omnidirectional imaging platform for drone detection. In Proceedings of the Target and Background Signatures II, Edinburgh, UK, 24 October 2016. [Google Scholar] [CrossRef]
- Anwar, M.Z.; Kaleem, Z.; Jamalipour, A. Machine Learning Inspired Sound-Based Amateur Drone Detection for Public Safety Applications. IEEE Trans. Veh. Technol. 2019. [Google Scholar] [CrossRef]
- Martins, E.; Fatih, E.; Chethan, K.A.; Ozgur, O.; Ismail, G. Micro-UAV detection and classification from RF fingerprints using machine learning techniques. In Proceedings of the 2019 IEEE Aerospace Conference, Big Sky, MT, USA, 2–9 March 2019. [Google Scholar]
- Nuss, B.; Sit, L.; Fennel, M.; Mayer, J.; Mahler, T.; Zwick, T. MIMO OFDM radar system for drone detection. In Proceedings of the 2017 18th International Radar Symposium (IRS), Prague, Czech Republic, 28 July 2017. [Google Scholar] [CrossRef]
- Fu, H.; Abeywickrama, S.; Zhang, L.; Yuen, C. Low-Complexity Portable Passive Drone Surveillance via SDR-Based Signal Processing. IEEE Commun. Mag. 2018, 56, 112–118. [Google Scholar] [CrossRef]
- Richardson, M. Drones: Detect, Identify, Intercept and Hijack. Available online: https://www.nccgroup.trust/uk/aboutus/newsroom-and-events/blogs/2015/december/drones-detect-identify-intercept-and-hijack/ (accessed on 7 November 2016).
- Zhang, X.D.; Shi, Y.; Bao, Z. A New Feature Vector Using Selected Bi-spectra for Signal Classification with Application in Radar Target Recognition. IEEE Trans. Signal Process. 2001, 49, 1875–1885. [Google Scholar] [CrossRef]
- Bernadini, A.; Mangiatordi, F.; Pallotti, E.; Capodiferro, L. IS&T International Symposium on Electronic Imaging 2017, Imaging and Multimedia Analytics in a Web and Mobile World. Electron. Imaging 2017. [Google Scholar] [CrossRef]
- Yang, B.; Matson, E.T.; Smith, A.H.; Dietz, J.E.; Gallagher, J.C. UAV Detection System with Multiple Acoustic Nodes Using Machine Learning Models. In Proceedings of the IEEE International Conference on Robotic Computing (IRC), Naples, Italy, 25–27 February 2019. [Google Scholar] [CrossRef]
- Dumitrescu, C.I.; Moise, I.M.; Soare, B.; Dumitru, N. Adaptive Noise Cancellation using LabVIEW. In Proceedings of the 2013 IEEE 19th International Symposium for Design and Technology in Electronic Packaging (SIITME), Galati, Romania, 24–27 October 2013. [Google Scholar]
- Jeon, S.; Shin, J.W.; Lee, Y.J.; Kim, W.H.; Kwon, Y.; Yang, H.Y. Empirical Study of Drone Sound Detection in Real-Life Environment with Deep Neural Networks. In Proceedings of the 25th European Signal Processing Conference (EUSIPCO), Kos, Greek, 28 August–2 September 2017. [Google Scholar]
- Li, S.; Kim, H.; Lee, S.; Gallagher, J.C.; Kim, D.; Park, S.; Matson, E.T. Convolutional Neural Networks for Analyzing Unmanned Aerial Vehicles Sound. In Proceedings of the 2018 18th International Conference on Control, Automation and Systems (ICCAS), Daegwallyeong, Korea, 17 October 2018; pp. 862–866. [Google Scholar]
- Anand, R.; Mehrotra, K.; Mohan, C.K.; Ranka, S. Efficient classification for multiclass problems using modular neural networks. IEEE Trans. Neural Netw. 1995, 6, 117–124. [Google Scholar] [CrossRef] [PubMed]
- Bishop, C.M. Pattern Recognition and Machine Learning; Springer: Berlin/Heidelberg, Germany, 2006. [Google Scholar]
- Galar, M.; Fernández, A.; Barrenechea, E.; Bustince, H.; Herrera, F. An overview of ensemble methods for binary classifiers in multi-class problems: Experimental study on one-vs-one and one-vs-all schemes. Pattern Recognit. 2011, 44, 1761–1776. [Google Scholar] [CrossRef]
- Dumitrescu, C.; Costea, I.M.; Nemtanu, F.; Stan, V.A.; Gheorghiu, A.R. Application of the Wavelet Transform in Machine-Learning. U.P.B. Sci. Bull. Series A 2014, 76, 167–178. [Google Scholar]
- Sonain, J.; Fawad; Rahman, M.; Ullah, A.; Badnava, S.; Forsat, M.; Mirjavadi, S.S. Malicious UAV Detection Using Integrated Audio and Visual Features for Public Safety Applications. Sensors 2020, 20, 3923. [Google Scholar]
- Shi, L.; Ahmad, I.; He, Y.; Chang, K. Hidden Markov model-based drone sound recognition using MFCC technique in practical noisy environments. J. Commun. Netw. 2018, 20, 509–518. [Google Scholar] [CrossRef]
- He, Y.; Ahmad, I.; Shi, L.; Chang, K. SVM-based drone sound recognition using the combination of HLA and WPT techniques in practical noisy environment. KSII Trans. Internet Inf. Syst. 2019, 13, 5078–5094. [Google Scholar]
- Mezei, J.; Molnar, A. Drone sound detection by correlation. In Proceedings of the 2016 IEEE 11th International Symposium on Applied Computational Intelligence and Informatics (SACI), Timisoara, Romania, 12–14 May 2016; pp. 509–518. [Google Scholar]
- Liu, H.; Wei, Z.; Chen, Y.; Pan, J.; Lin, L.; Ren, Y. Drone detection based on an audio-assisted camera array. In Proceedings of the 2017 IEEE Third International Conference on Multimedia Big Data (BigMM), Laguna Hills, CA, USA, 19–21 April 2017; pp. 402–406. [Google Scholar]
- Dumitrescu, C.; Minea, M.; Ciotarnae, P. UAV Detection Employing Sensor Data Fusion and Artificial Intelligence, Information Systems; Springer: Cham, Switzerland, 2019. [Google Scholar] [CrossRef]
Property | Spectrogram | Wigner-Ville | Born-Jordan | Choi-Williams |
---|---|---|---|---|
Energy | √ | √ | √ | √ |
Time limitation | - | √ | √ | √ |
Frequency limitation | - | - | - | |
Reality | √ | √ | √ | √ |
Positive Value | √ | - | - | - |
Causality | - | - | - | - |
Reversibility | - | √ | - | √ |
Expansion | - | √ | √ | √ |
Filtering | - | √ | - | - |
Modularity | √ | √ | ||
Temporal support | - | √ | √ | - |
Frequency support | - | √ | √ | - |
Unit value | - | √ | - | - |
Instantaneous frequency | - | √ | √ | √ |
Group delay | - | √ | √ | √ |
Distribution Type | ||
---|---|---|
1 | Wigner-Ville | |
Born-Jordan | ||
Choi-Williams | ||
Spectogram |
Predict Label | ||||
---|---|---|---|---|
Background Noise | Single Drone | Two Drone | ||
True Label | Background Noise | 1.00 | 0.00 | 0.00 |
Single Drone | 0.00 | 0.94 | 0.06 | |
Two Drone | 0.00 | 0.23 | 0.77 |
Classes | Precision | Recall | F1—Score |
---|---|---|---|
Backgroud Noise | 1 | 1 | 1 |
Single Drone | 0.80 | 0.94 | 0.87 |
Two Drone | 0.93 | 0.77 | 0.84 |
Avg/total | 0.91 | 0.90 | 0.90 |
Predict Label | ||||
---|---|---|---|---|
Background Noise | Single Drone | Two Drone | ||
True Label | Background Noise | 0.89 | 0.08 | 0.03 |
Single Drone | 0.03 | 0.84 | 0.13 | |
Two Drone | 0.03 | 0.13 | 0.85 |
Classes | Precision | Recall | F1—Score |
---|---|---|---|
Backgroud Noise | 0.93 | 0.89 | 0.91 |
Single Drone | 0.80 | 0.85 | 0.82 |
Two Drone | 0.85 | 0.84 | 0.84 |
Avg/total | 0.86 | 0.86 | 0.86 |
Predict Label | ||||
---|---|---|---|---|
Background Noise | Single Drone | Two Drone | ||
True Label | Background Noise | 0.98 | 0.01 | 0.01 |
Single Drone | 0.09 | 0.86 | 0.06 | |
Two Drone | 0.07 | 0.16 | 0.77 |
Classes | Precision | Recall | F1—Score |
---|---|---|---|
Backgroud Noise | 0.86 | 0.98 | 0.91 |
Single Drone | 0.83 | 0.84 | 0.83 |
Two Drone | 0.92 | 0.78 | 0.84 |
Avg/total | 0.88 | 0.88 | 0.88 |
UAV No. | 1 | 2 | 3 | 4 | 5 | 6 |
---|---|---|---|---|---|---|
Training | 472 | 616 | 439 | 625 | 553 | 54 |
SOM | 45 | 38 | 14 | 25 | 25 | 4 |
CoNN | 27 | 27 | 27 | 27 | 27 | 27 |
Drone Model | SOM Accuracy | CoNN Accuracy | Recognition Distance [m] | Drone Class |
---|---|---|---|---|
1 | 73.65% | 85,21% | ~380 | medium |
2 | 60.47% | 94.95% | ~380 | medium |
3 | 50% | 97.34% | ~380 | medium |
4 | 5.89% | 90.43% | ~500 | large |
5 | 15.03% | 82.37% | ~150 | small |
6 | 45% | 85.6% | ~150 | small |
© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Share and Cite
Dumitrescu, C.; Minea, M.; Costea, I.M.; Cosmin Chiva, I.; Semenescu, A. Development of an Acoustic System for UAV Detection. Sensors 2020, 20, 4870. https://doi.org/10.3390/s20174870
Dumitrescu C, Minea M, Costea IM, Cosmin Chiva I, Semenescu A. Development of an Acoustic System for UAV Detection. Sensors. 2020; 20(17):4870. https://doi.org/10.3390/s20174870
Chicago/Turabian StyleDumitrescu, Cătălin, Marius Minea, Ilona Mădălina Costea, Ionut Cosmin Chiva, and Augustin Semenescu. 2020. "Development of an Acoustic System for UAV Detection" Sensors 20, no. 17: 4870. https://doi.org/10.3390/s20174870
APA StyleDumitrescu, C., Minea, M., Costea, I. M., Cosmin Chiva, I., & Semenescu, A. (2020). Development of an Acoustic System for UAV Detection. Sensors, 20(17), 4870. https://doi.org/10.3390/s20174870