Drone Range Detection Using Extracted Mel Frequency Cepstral Coefficients with Logistic Regression and Support Vector Machines

Small UAS (sUAS) pose hazards to local, and federal agencies when controlled by negligent, reckless, criminal, or military operators. The government must protect and defend personnel, facilities, and assets in an environment where an increasing number of sUAS's share the skies with aircraft, operate in the air over protected air space, and are employed as weapons [3]. This research proposes a unique approach to predicting the range of a drone based on the strength of the auditory signal it emits using machine learning techniques. Two experiments are conducted during this research. In the first, the range boundary is from zero to 100 meters, and the second is from zero to 140 meters respectively. Finally, the models are compared with each other given a binary look at the outcome, including data about drone detection versus noise detection, and omitting range information. In both experiments, pre-processed audio signals are passed to logistic regression and support vector machine models, with labels separating ranges by twenty meters each. These models accurately predicted the range of sUAS for experiment 1 80.1% and 77.7% of the time for logistic regression and SVM respectively, and 70.1% and 68.2% of the time for experiment 2 respectively.