<![CDATA[The use of YOLO (You Only Look Once)-based object detection algorithms has become one of the main approaches in visual object recognition and training. This study aims to compare the performance of three versions of YOLO, namely YOLOv5, YOLOv8, and YOLOv11, in training models to detect objects in Chinese chess images. The dataset used consists of images of Chinese chess boards and pieces in various positions and lighting variations. The training process was carried out using uniform parameters to ensure fair evaluation, including batch size, number of epochs, and image resolution. The performance of each model was evaluated based on detection accuracy, inference speed, and computational efficiency metrics. The results of the study show that each version of YOLO has specific advantages in certain aspects, such as training speed or detection precision. From the 7224 images used as the dataset, several results were obtained that were necessary in helping to compile this journal. These included Precision (YOLOv5: 0.94, YOLOv8: 0.96, YOLOv11: 0.98), Recall (YOLOv5: 0.93, YOLOv8: 0.98, YOLOv11: 0.96), and mAP (YOLOv5: 0.96, YOLOv8: 0.98, YOLOv11: 0.99). This study provides important insights into the advantages and disadvantages of each version of YOLO in the specific application of Chinese chess object recognition, as well as providing guidance for developers in choosing the model that suits their project needs. This study provides insights into the strengths and limitations of each YOLO version, offering guidance for selecting appropriate models in real-time Chinese chess object detection applications.]]>
