جعفر نجفي كسلاني

As the use of drones continues to expand in commercial and military sectors, concerns regarding their potential misuse have grown significantly. Addressing and mitigating drone-related threats has become a pressing necessity, emphasizing the demand for accurate and cost-efficient systems for their detection and classification. This thesis presents an innovative approach for drone detection and classification based on the analysis of environmental audio signals. The proposed method begins with extracting novel audio features using the Prony algorithm, which is specifically designed to capture distinctive frequency characteristics of drone sounds. The effectiveness of these extracted features is initially eva‎luated using conventional classifiers, such as k-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Shallow Neural Networks (SNN). Subsequently, the same features are employed in an optimized deep neural network, which significantly reduces computational complexity by minimizing the number of learnable parameters, thereby enhancing overall system performance. Experimental results validate the high efficiency of the proposed method, achieving detection accuracy above 97% and classification accuracy exceeding 94%. Furthermore, a comprehensive and diverse audio database containing drone sounds recorded under various flight conditions has been developed, offering a valuable resource for future research. Compared to conventional methods that primarily utilize Convolutional Neural Networks (CNNs) and Short-Time Fourier Transform (STFT) spectrograms, the proposed approach leverages numerical audio features and an optimized deep neural network. This design not only ensures superior accuracy but also minimizes the reliance on complex hardware, making it highly suitable for practical deployment in resource-constrained systems.

آشكارسازي ريزپرنده‌ها، دسته‌بندي ريزپرنده‌ها، الگوريتم پروني، ويژگي‌هاي صوتي، شبكه‌هاي عصبي عميق، پايگاه داده صوتي

Drone Detection, Drone Classification, Prony Algorithm, Audio Features, Deep Neural Networks, Audio Database

Jafar Najafi Kasalani

Dr. Sattar Mirzakuchaki