Road Object Detection Algorithm Based on Attention Mechanism and Feature Fusion
Aiming at the problems of small detection targets and insufficient model feature fusion,this paper proposes a road target detection algorithm MFFDM based on attention mechanism and multi-scale feature fusion.In this algorithm,Resnext50 network and attention module are integrated to form a new backbone feature extraction network.Secondly,a new bottom detection layer with spatial location information is added to match the detection of small objects.In addition,deconvolution module and feature texture extraction module are used to design the multi-scale feature fusion network DEFTFN.Experiments show that,compared with FCOS algorithm,the average accuracy of this algorithm on KITTI data set is improved by 9.3％,and the detection accuracy of pedestrian targets on the road is improved significantly by 14.6％.
万方数据
维普
