Research On Traffic Sign Detection Combining Feature Fusion Attention Optimization And Image Maskin

Traffic sign detection is one of the important technologies in the neighborhood of automatic driving and assisted driving,which is of great significance for the development of intelligent transportation and the guarantee of traffic safety.In the natural environment,cars travel faster,so traffic sign detection related methods need to have the ability to detect quickly.In addition,because the observed traffic sign targets are usually small and the environmental background is diverse,in order to obtain traffic sign information and make judgments early,the relevant detection methods need to have excellent small target positioning and background discrimination capabilities.This paper aims to study the traffic sign detection method based on deep learning,and improve it on the basis of the classical object detection model to make it more suitable for traffic sign detection tasks,including the following two aspects:(1)Traffic sign object detection combined with multi-channel attention mechanism: In this paper,a multi-channel attention mechanism method is proposed for the object detection model to obtain better fusion results in the process of feature fusion,so as to improve the model’s ability to distinguish complex backgrounds.The multi-channel attention mechanism is formed by the parallel connection of the chunked-channel attention and the multi-attention spatial pyramid pooling.The feature map is divided into regions through the chunked-channel attention mechanism,and the adaptive channel weights are carried out to obtain the channel weights of different regions of the feature map.In the process of obtaining a variety of different receptive field feature maps,the multi-attention spatial pyramid pooling is reconstructed from part to whole,and the splicing and fusion is carried out,and finally the feature map with multiple receptive fields is combined with the weight obtained by the chunked-channel attention,so as to improve the multiple attention to the required detection targets and effectively optimize the detection performance of the model.(2)Traffic sign detection based on image mask and feature fusion: In this paper,a circular feature pyramid network is proposed to improve the small target detection ability of the model,and an image mask-assisted training method is introduced to improve the discrimination ability of the model from non-target signs.In this paper,the feature maps of different scales are fused twice,the results are analyzed after the first fusion,the small target area is located,the attention mechanism is used to enhance the small target information in the positioning area,and then the second fusion is carried out,and the experimental results show that the accuracy of the small target is significantly improved.In addition,this paper reclassifies the importance of negative samples in training by image color mask and image texture mask,and optimizes the loss function weighting method to enable the model to better distinguish traffic signs from interference signs.In addition,through data balance and model simplification,the model achieves better detection effect,while reducing the amount of calculation and parameters.Experimental data show that the optimized model is more suitable for traffic sign detection tasks.