A Comparative Study of Machine Learning and Deep Learning Models for Drone Object Detection

As drone usage in commercial and prohibited airspaces has increased markedly [1], the need for strong and automated drone detection systems has become increasingly imperative. This research uses the UAV [12], [18] Drone Dataset and assesses various ML and DL models for precise drone identification using aerial images. Five popular detection and classification models were tested and compared: YoLOv5, Convolutional Neural Network (CNN), Artificial Neural Network (ANN), Support Vector Machine (SVM) and Random Forest. This process comprises model training, annotation, image dataset preprocessing, and performance comparison. The models are compared with respect to crucial performance indicators such as precision, precision, recall, F1score, and AUC-ROC. Comparative analysis determines the potential and constraints of each model, with YOLOv5 [10] showing high applicability in real-time detection scenarios. A comparative evaluation identifies the capability and limitations of each model, and YOLOv5 demonstrates strong suitability for real-time detection applications.Furthermore, this research postulates the viability of using lightweight models for edge-based drone monitoring and enabling real-world integration into monitoring and security systems.