Research On Object Detection Algorithm Based On Multi-Scale Semantic Information Fusion

Image object detection is a hot topic in the field of computer vision.Its main task is to locate the target of interest from the input image,and then accurately determine the category of each target of interest.In recent years,with the rapid development of deep learning technology,object detection technology has been widely used in daily life safety,robot navigation,intelligent video surveillance,traffic scene detection,aerospace and other fields.In particular,the successful application of convolutional neural networks has greatly improved the performance of object detection,and a large number of different object detection algorithms and network models have emerged.The object detection technology has been rapidly developed.Based on a thorough investigation of the object detection technology,this paper analyzes the deficiencies of the classic SSD detection method and makes improvements.The main research content of this article includes the following aspects:(1)This article first briefly introduces the research background,significance and difficulties of object detection,Then the object detection algorithms based on deep learning were reviewed according to two categories: candidate region-based and regression-based.For the candidate region-based algorithms,we first introduced the R-CNN(Region with Convolutional Neural Network)based series of algorithms,and then the R-CNN based methods were overviewed from four dimensions: the research of feature extraction networks,and the Region of Interesting Pooling researches,improved works based on Region Proposal Networks,and some improved approaches of Non Maximum Suppression algorithms.Next the regression-based algorithms were surveyed in terms of YOLO(You Only Look Once)series and SSD(Single Shot Multibox Detector)series.Finally,according to the current trend of object detection algorithms that are developing more efficient and reasonable detection frameworks,the future research focuses of unsupervised and unknown category object detection directions were prospected.(2)The SSD algorithm not only has real-time detection speed,but also has good detection accuracy.However,due to the lack of connection between the features of various scales extracted by the SSD algorithm,the shallow features lack semantic information,and the high-level features lack spatial details.The detection effect of small targets and dense targets is very unsatisfactory.In response to these problems,this paper proposes a shallow enhancement network based on the SSD algorithm,and designs a small feature pyramid structure to perform multi-scale semantic information fusion on the extracted shallow layers.Obtain enhanced shallow features with detailed spatial details and rich semantic information,and then use this as a basis to extract multiple different scale features for detection.The experimental results on the PASCAL VOC2007 dataset and the MS COCO2014 dataset effectively prove that the model in this paper has a good detection effect for small targets and dense targets.(3)In order to further study the impact of multi-scale semantic information on feature representation capabilities,this paper improves on the basis of shallow enhancement networks,and proposes a multi-scale feature attention enhancement network,by using the first two large-scale feature detection branches Introduce the receptive field module to enhance the representation ability of large-scale features for large and medium-sized targets,and then design four attention guidance modules to construct a similar feature pyramid network feature fusion structure for multiple different scale features,not only enhance the attention of multiple scale features,and deepen the connection between different scales.The experimental results on the PASCL VOC2007 dataset effectively prove that the improved method in this paper can effectively improve the detection accuracy of the model.