METHODS AND ALGORITHMS FOR IDENTIFYING DANGEROUS OBJECTS FOR ROBOTIC DEVICES BASED ON ARTIFICIAL INTELLIGENCE
A. A. Khakimov, F. M. Nazarov, I. Musurmonova · Zenodo (CERN European Organization for Nuclear Research) · 2026
This study explores the development of methods and algorithms for detecting hazardous objects in robotic systems powered by artificial intelligence. A specialized dataset named DENGROUS was created using the Roboflow platform, consisting of 10 classes, 15,668 images, and 21,944 annotations. Their performance was then comparatively evaluated on a validation dataset. Experimental results showed that YOLO26s achieved the highest accuracy with [email protected] = 0.876 and [email protected]:0.95 = 0.701. The YOLO11s model ranked second with a Precision score of 0.897, while YOLOv8n demonstrated the fastest inference speed at 4.5 ms per image.