Research On Foggy Road Target Detection Algorithm Based On YOLOv4-Tiny

In recent years,with the wide application of target detection technology in industry,life and other fields,many scholars have carried out research on the target detection algorithm on foggy roads.However,it is mainly realized based on the idea of independent operation of defogging algorithm and detection algorithm,without considering the problem that the use of defogging algorithm may lead to the loss of image details.Based on this,the algorithm of fog removal and detection was improved,and the idea of joint optimization was introduced to establish the joint optimization algorithm of road target detection on foggy days,PD-YOLO,to ensure the detection speed and improve the detection accuracy.A lightweight dehazing algorithm PD-Net based on convolutional neural network is proposed.Drawing on the idea of AOD-Net dehazing algorithm to build a 5-layer convolutional neural network model;An improved spatial pyramid pooling structure is introduced to improve the learning ability of the model.The loss function is improved to improve the model optimization effect and combining the atmospheric scattering model to restore no fog image.To verify the defogging effect of the algorithm,experiments were carried out on the foggy data set RESIDE.The results show that the PSNR and SSIM values are 21.7and 0.85,respectively,and the processing time of a single image is 0.29 s,with high running speed.Based on YOLOv4-Tiny,an improved target detection algorithm is proposed.Based on YOLOv4-Tiny,the channel attention mechanism ECANet is combined to improve the model’s ability to focus on key information;the path aggregation network PANet is introduced to improve the model’s utilization of deep feature information;Soft-NMS and Focal Loss functions are used to improve the model’s ability to analyze feature information.Experiments are carried out on the road target dataset KITTI,and the results show that the m AP value of the detection algorithm is 78%,which improved the accuracy by 3.7% compared with the original algorithm and met the practical requirements.The combined optimization idea is introduced to build the combined optimization network PD-YOLO for fog removal detection.The problems of foggy detection algorithm were analyzed,and the proposed de-fogging algorithm PD-NET was embedded into the improved YOLOv4-Tiny network structure.Combined with the de-fogging detection,the loss function was optimized to improve the target detection performance in foggy environments.The practicability of the algorithm is verified by the real-world Task-driven Testing Set,and the self-made foggy road target detection data set is used for comparative experiment verification.The results show that the m AP value of the joint optimization algorithm for fog removal detection is 72.5%,which can meet the requirements of foggy weather conditions.road object detection requirements.