Adapted Model: Drone vs. Bird Based on Pretrained Knowledge

Drones have become popular among citizens due to safety problems that need to be solved and controlled. In recent years, there have been tensions about the safety of people and the protection of buildings due to the sharp rise in the use of drones in industrial contexts. The development of drone classification is one field that has arisen in response to these tensions. Several current techniques have drawbacks for the inaccurate detection of other objects such as the presence of birds. In this research, we have proposed a drone classification model using a fine‐tuning learning approach to enhance performance results with the best metrics, instead of the previous research that used different approaches. Four models are applied, which are Xception, Inception, ResNet50, and MobileNet V2, for drone classification. Deep transfer learning is applied to accelerate execution time by a training model on ImageNet. A dataset containing drone and bird images is essential to deal with this problem and train accurate classification models. The performance of these models is evaluated by comparing them with earlier studies that employed various methodologies. The Xception model led to the highest accuracy, precision, and loss function and produced the best decision performance compared to the other models, with an accuracy of 99.18% and a loss function of 11.53%.