Research On Object Detection Algorithm Based On Feature Fusion And Information Enhancement

Object detection is to analyze and locate the object in the input image with the help of computer,and obtain the category and position coordinates of the object.It is an important research direction in the field of computer vision.With the application of deep learning in object detection technology,the performance of object detection has been significantly improved.It has been successfully applied in intelligent security,medical diagnosis,defect detection,intelligent agriculture and other fields,which has greatly improved the production efficiency.However,due to the large change of object scale,the similarity between object and background,the dense and overlapping number of objects in complex scenes,it is difficult for the object detection model to adaptively extract the discriminative object features with different scales,resulting in the problems of low detection accuracy and weak generalization ability,which brings great challenges to object detection.To solve the above problems,this paper uses feature fusion,attention mechanism,expansion convolution and other methods to enhance the semantic information and receptive field area of features.These measures increase the detection precision,especially the ability to detect small objects.The principal contributions of this paper are as lists:(1)Firstly,this paper briefly expounds the research background,research significance and research status of object detection,then summarizes the classical algorithms of object detection,summarizes the improvement research of corresponding algorithms from the perspectives of multi-scale object detection and small object detection,and analyzes and compares the detection performance of each algorithm on MS COCO data set.Finally,the commonly used data sets and evaluation indicators in object detection tasks are summarized.(2)In order to reduce the missed or false detection of objects in the detection process,and detect objects with different sizes in the image preferably.An algorithm based on shallow feature fusion and semantic information enhancement is proposed in this paper.At first,we add high-level semantic information into low-level features to enhance the detail information.Then,the relationship between local and global context information is established to highlight the location of the object and reduce information confusion.Finally,extended convolution is used to enhance the receptive field of shallow features to adapt to different scales of object detection.The method has carried out a large number of experiments on PASCAL VOC and MS COCO data sets.The results show that it has good detection performance,especially for small objects,which outperforms the state-of-the-art methods.(3)Due to the attributes of small objects and influence of complex background information,we propose a small object detection method based on feature information enhancement.Firstly,the global information,local information and multi-scale information of input features are fused to establish and strengthen the communication and connection between information.Then,the receptive field features of different scales are densely connected to obtain discriminative features that integrate different levels of abstract information.The detection accuracy of this model on PASCAL VOC,MS COCO and UCAS-AOD data sets has reached 81.3%,34.8% and87.0% respectively.In addition,the detection results of small objects are also slightly higher than the current advanced detection algorithms YOLOv4 and DETR by 0.2% and 0.1%respectively,which further proves the effectiveness of this algorithm for detecting small objects.