Detection of Drone and Non-Drone Objects with Machine Learning Methods Using Visual Data
Yagidar Deniz Gencer, Sema Servı, Yavuz Selim Taşpınar · 2025
Nowadays, with the development of technology, the use of drones has increased in different areas. It is used effectively in many areas, especially defense, shopping, exploration, photography, entertainment, education and health. This technology is used in processes that are beneficial to humanity, as well as in undesirable situations that may harm humanity. Image classification is done successfully using deep learning methods. It is aimed to detect drone and non-drone objects from the image datasets given in this article using image processing methods. For this purpose, the train folder was used in the data folder in the data set, and among the 901 images in this folder, 502 images with drones and 399 images without drones were used. Images were classified with Convolutional Neural Network (CNN) learning methods using SqueezeNet and InceptionV3 architectures. Each image was classified with Artificial Neural Network (ANN), K Nearest Neighbor (KNN), Random Forest (RF), AdaBoost (AB), Decision Tree (DT) machine learning models found in SqueezeNet and InceptionV3 models. Cross validation method was used to evaluate the performance of the models. Accuracy, AUC, precision, F1 Score, recall and specificity metrics were used to measure the performance of the models. In the SqueezeNet model, classification success was achieved with ANN $91 \%$, AB $79 \%$, KNN 88.8%, DT 81.4% and RF 84.2% rates. In the classification made with InceptionV3, classification success was achieved with ANN 94.9%, AB 82.8%, KNN 92.6%, DT 84.4% and RF 88.5%. The study contributes to its use in real life, especially in detecting enemy targets in defense technologies, separating foreign plants from normal plants in agriculture, and many other areas.