Small Scale Vehicle Target Detection Algorithm Based on Improved YOLOv3
Aiming at the problems of false detection,missed detection,and low accuracy in the detection results of small target vehicles relatively far from the camera during traffic checkpoint vehicle target detection,an improved method was proposed to enhance the feature extraction network layer of the original YOLOv3 object detection model,and optimized the YOLOv3 original loss function to dynamically adjust the proportion of detection boxes at different scales.A model selection comparison experiment was conducted on 6 different pixel datasets,and the experimental results showed that the YOLOv3-5L model with a feature extraction layer set of 5 layers achieved the best convergence speed and small object detection accuracy as the image resolution improved.At 2 400辽×2 400 pixels,the average accuracy of the YOLOv3-5L model reached 96.5％,which was 2.0％higher than the original YOLOv3 network.
万方数据
维普
