Evaluating the Effectiveness of YOLO Models in Different Sized Object Detection and Feature-Based Classification of Small Objects

Abstract

The YOLO tool has been increasingly developed to assist in object classification. However, the problem of object classification has many difficulties, including small objects, background effects, or noise loss of information. Therefore, to evaluate the objective of classifying small, information-losing objects, the research team installed and evaluated new YOLO models such as YOLOv5, YOLOv6, and YOLOv7. In addition, the study conducted a feature-based object classification test for comparison and evaluation. The study was conducted on the self-collected data set, divided into 2 datasets: a dataset used to evaluate object classification and a dataset used to classify by features. The evaluation results show certain advantages of the YOLOv7 model on parameters such as Precision, Recall, and a mAP threshold of 50%. The evaluation results show certain advantages of the YOLOv7 model on parameters such as Precision, Recall, and a mAP threshold of 50%. The study results show that YOLOv7 achieves specific effects when the accuracy of object recognition is over 90%, in which the feature-based classification also achieves an accuracy of over 70%. This issue may need different future studies in object recognition and object feature recognition.