Research On Multi-Scale Object Detection Algorithm Based On Anchor Free

In the era of continuous development of artificial intelligence,computer vision has become an indispensable part of it,especially multi-scale target detection,which plays an important role in intelligent cities,intelligent monitoring,and other fields.However,in the process of multi-scale target detection in the airport environment,there are still some problems in the existing object detection algorithms,such as missed detection and false detection of small targets and dense targets.In order to achieve efficient management of the airport and improve the intelligence level,two object detection algorithms without anchor are designed and constructed for aircraft targets and X-ray dangerous goods targets under the bird ’s-eye view,so as to improve the detection performance of the improved algorithm on small targets,dense targets,and overlapping targets under the bird ’s-eye view in the airport environment,and accelerating the construction of intelligent airports.The main contributions of this paper are as follows:(1)Aiming at the problems of missed detection and false detection of dense multiscale aircraft targets in the airport,a Center Net algorithm based on receptive field enhancement and memory feature fusion is proposed.Firstly,the multi-scale receptive field module and residual attention module are introduced to improve Res Net50,and it is used as a new feature extraction network.The receptive field enhancement module broadens the network width,enhances the multi-scale feature extraction ability,and the residual attention model strengthens the localization ability of the end of the network.Secondly,the memory feature fusion module is used instead of cubic deconvolution to eliminate the checkerboard effect caused by deconvolution.At the same time,shallow detail information and deep semantic information can be fully integrated through the fusion network,and the key information in the output feature map is further enhanced.The experimental results show that the accuracy rate of the improved algorithm to detect the aircraft parked at the airport is 88.46%,which is 18.21% higher than that of the Center Net algorithm,and the detection speed is also guaranteed.(2)Aiming at the problem of insufficient detection performance of multi-scale overlapping dangerous goods caused by the similar background and target of X-ray dangerous goods,an object detection algorithm of feature recalculation and channel shuffle is proposed.Firstly,the improved decoupling detection head is used as the detection head of the algorithm,in which the feature weights of different channels are recalculated to reduce the loss of important features and enhance the ability to locate multi-scale targets.Secondly,an atrous space convolution pyramid model guided by ghost module is introduced to expand the perception range of multi-scale targets and suppress the interference of background information.Finally,the fusion method of channel shuffling is introduced to improve the fusion method,which rearranges the features of different channels to enrich context information and enhance semantic information.The experimental results on the multi-scale X-ray dangerous goods dataset show that the average accuracy of the improved algorithm in detecting dangerous goods reaches 92.35%,which is 5.48% higher than the basic algorithm,and the real-time performance is guaranteed.(3)The Pascal VOC dataset is introduced as the cross-validation dataset of receptive field enhancement and memory feature fusion algorithm,and the CVC-09 night pedestrian dataset is introduced as the cross-validation dataset of feature recalculation and channel shuffle algorithm,the robustness and universality of the improved algorithm are further illustrated by different datasets and different kinds of targets.