Identification of Military Vehicles Using Vision Systems and AI Methods

Abstract. This paper presents a method for identifying graphical equivalents of military vehicles. A YOLO11L neural network model, an enlarged version of the YOLO11 network, was used. YOLO models are algorithms that use the RGB color palette for object detection. LabelStudio software was used to train the neural network for image labeling. To properly train the artificial neural network for vehicle detection, approximately 3,000 images were used [1-3], each containing over 12,000 silhouettes of allied Polish/American and enemy Russian military vehicle models. The model was trained at various standard resolutions: 640x640 and 1024x1024. Because the computing power available on personal computers was insufficient for model training and time was a key factor, Google servers on the Google Colab platform were utilized, leveraging A100 GPUs. Achieved around 7-15 fps for object detection using laptops with RTX 3060 graphics cards. Further research is planned on modifying the network models and training the final network model [3-5].