Comprehensive Evaluation of UAV Object Detection Models Using the Analytic Hierarchy Process and Entropy Weight Method

A single accuracy metric cannot adequately describe how object detection models perform in unmanned aerial vehicle (UAV) applications. This study develops a multidimensional evaluation method by combining the analytic hierarchy process (AHP) with the entropy weight method. The index system is built around three dimensions, usability, efficiency, and robustness. AHP assigns subjective weights according to task requirements, while the entropy weight method calculates objective weights from variation in the experimental data. These two sets of weights are then combined to evaluate each model as a whole. Experiments were carried out on the VisDrone2019 dataset using five representative object detection models: YOLOv8s, YOLOv7-tiny, DINO-DETR, FCOS, and RetinaNet. The results indicate that YOLOv8s obtains the highest overall score of 86.48. YOLOv7-tiny and DINO-DETR follow with scores of 80.43 and 79.09.This method distinguishes differences among models across detection performance, computational efficiency, and robustness under complex imaging conditions. It therefore offers a quantitative basis for comparing and selecting UAV object detection models..