A Research And Application Of Faster RCNN Target Detection Algorithm Based On Attention Mechanism

With the continuous development of science and technology,the target detection technology in the field of computer vision has been applied to many aspects of human life.For example,security detection,smart cities,unmanned driving,intelligent robots and other fields have all applied target detection technology.Due to the complexity of application scenarios,there are still many challenges in target detection technology.In recent years,Faster R-CNN(Towards Real-Time Object Detection with Region Proposal Networks)algorithm has been widely used in target detection.However,there are the following two problems in this algorithm: 1)A large number of smaller convolution kernels are used in the convolutional neural network,which leads to the problem that the long-distance correlation of the feature map is weak.2)There is a problem of insufficient multi-scale feature fusion within the feature pyramid network.This article conducts research on the above two issues:1.In view of the problem that the Faster R-CNN target detection algorithm has a weak long-distance correlation within the feature map inside the convolutional neural network,this paper proposes the AT-FCNN(Attention-Faster Convolutional Neural Networks)target detection algorithm in conjunction with the visual attention mechanism.This algorithm embeds the attention module into the convolutional neural network.During the feature extraction process of the convolutional neural network,the attention module calculates the correlation coefficient between the internal features of the feature map to achieve the effect of enhancing the representation ability of the feature map.Finally,this paper verifies the effectiveness of the algorithm in the public data set.The experimental results show that the algorithm in this chapter has a 5.8% improvement in the average accuracy value.2.In the feature golden tower network within the framework of Faster R-CNN,the original feature fusion strategy adopts the method of fusion of deep feature maps to shallow feature maps,which makes the shallow layer Feature maps fully contain multiple layers of feature information,while deep feature maps are not fused to shallow feature map feature information,which ultimately leads to the problem of inadequate fusion of multiple scale features.Aiming at the above problems and combining the visual attention mechanism,this paper proposes ATF-FCNN(Attention Feature Pyramid-Faster Convolutional Neural Networks)feature fusion algorithm.The algorithm first fixes the multi-scale feature information to the same size,performs feature fusion,and outputs a fusion feature map of that size.Secondly,the fusion feature map is sent to the attention module for processing.Finally,the processed feature map is up-sampled or down-sampled,re-transformed into a feature pyramid multi-scale feature form,and input to the next stage of the network.This paper verifies the effectiveness of the algorithm in the public data set,and the experimental results show that the performance of the algorithm in this chapter has been further improved.In summary,through the research of the Faster R-CNN target detection algorithm based on the attention mechanism,the experimental results show that the proposed problem is effectively solved,and the overall performance of the algorithm has been significantly improved to achieve The purpose of learning and researching the target detection algorithm is expected.In the aspect of target detection,a new theoretical solution is given,which has strong practical significance.