Yolov7-tiny road target detection algorithm based on attention mechanism
Year: 2024
Type: conference-paper
Abstract: Aiming at the problems of low multi-scale detection accuracy, missed detection and false detection of road scene targets, a lightweight YOLOv7-Tiny algorithm that integrates polarization self-attention mechanism is proposed. First, the bounding box position loss function in the algorithm is replaced by It is the α-CIoU loss function to improve the convergence accuracy; secondly, the polarization self-attention mechanism is introduced into the path aggregation network to enhance the information expression of the region of interest; finally, the training set is preprocessed, and the grayscale world is used for data enhancement. The algorithm reduces the influence of different lighting conditions on the real color performance of the object and improves the detection accuracy. Compared with the original algorithm, the performance of the improved algorithm in the road target recognition task of the KITTI dataset is significantly improved compared with the original algorithm; mAP, Precision and Recall increased by 3.2%, 0.25% and 3.64% respectively, and the robustness and generalization ability are better.
Source: Procedia Computer Science
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4404844331