Research on vehicle detection based on improved YOLOv7-Tiny
In order to improve the accuracy and speed of computer recognition and vehicle detection,an improved YOLOv7-Tiny vehicle detection algorithm is proposed.Among numerous object detection models,YOLOv7 has a very fast detection speed and high detection accuracy,making it very suitable for real-time detection tasks.This article improves on the original YOLOv7-Tiny model by incorporating the shallowest ELAN-T module into the feature pyramid,and cross layer fusion of shallow and deep features is achieved through skip connections,resulting in richer output feature information.At the same time,the SE attention mechanism is introduced to allocate computing resources to information that is more critical to the current task.And the nonlinear activation function HardSwish was replaced to improve the model's expression ability.Experiments were conducted on the 2D autonomous driving dataset SODA10M released by Huawei,and the results showed that the improved model showed improved detection accuracy for all four types of targets,with an average accuracy of mAP@0.5 reached 66.1%,an increase of 5.1%compared to the original YOLOv7-Tiny model of 61.0%.
万方数据
维普
