Object Detection Algorithm Research Based On Multi-feature Multi-scale Convolutional Neural Networks

The purpose of computer vision research is to use computers to realize human perception,recognition and understanding of the objective world.As one of the core research topics in the field of computer vision,object detection technology has attracted extensive attention in the field of computer vision theory research and has broad application prospects.It extracts object features through analysis,and then obtains object category and location information.Obj ect detection is an important task of computer vision.It has been widely used in intelligent transportation systems,intelligent monitoring systems,human-computer interaction,automatic driving,image retrieval,and intelligent robots.In recent years,with the improvement of hardware computing capabilities,the birth of large data sets,and the development of deep learning technologies,object detection performance has been greatly improved.Among them,the deep convolutional neural network has powerful feature extraction ability and learnability for its structural advantages,and it can extract image feature information effectively.This paper proposes an object detection algorithm based on multi-feature and multi-scale convolutional neural networks.It fully considers the inherent hierarchical structure of convolutional neural networks.It uses the super feature fusion algorithm to fuse multi-layer features with different feature information,and builds multi-scale feature pyramid based on fused feature.We further use multi-path dense feature fusion algorithm to enrich feature information and use the multi-level feature pyramid generation and fusion model to construct multi-scale fused features.Based on the multi-scale fused features,we construct region proposals.We design a convolutional module to extract multiple features and improve the model training process.Finally we use the multi-task loss to optimize the detection network to achieve object classification and localization jointly.We train and test the detection framework on the standard data set and the results show that the proposed algorithm can effectively improve the feature presentation ability and improve the object detection accuracy.