Research On Object Detection Method Based On Context Information Fusion And Attention Awareness

With the rapid development of society,electronic devices are becoming more and more popular.Every day,massive amounts of image data are generated,so it is increasingly important to intelligently identify the objects that exist in these images.Therefore,object detection has broad application prospects.Object detection is a basic task in computer vision.This task aims to locate all the objects in the image and classify them correctly.In recent years,with the rapid development of deep learning,the object detection method based on convolutional neural network has made many breakthroughs,but there are still some shortcomings.On the one hand,since there are various interference factors in the image,the convolutional neural network may extract less feature information and using the limited feature information will cause a higher classification error.On the other hand,the key areas in the image may have the problem of low discrimination,which makes it difficult to find out the essential feature information of objects.Therefore,how to extract more feature information and improve the discernibility of key areas in the image is the key problem to be solved in this paper.This paper studies the above two problems and proposes corresponding solutions.The main contents include:1)Based on the Faster R-CNN algorithm,this paper introduces a novel feature optimization method and further proposes a new network,namely CA-NET(Context Aware Net).CA-NET can effectively extract context information of different scales in the image.The context information can provide additional clues for the classification of the objects,and thus effectively supplement the feature information of the objects.2)In the Faster R-CNN algorithm,the use of RoI(Region of Interest)pooling will reduce the discriminability and lose many details in the feature map.To solve this problem,this paper introduces a new RoI attention-awareness method,and further proposes another new network,namely CAA-NET(Context-Aware Attention Net).CAA-NET can learn the importance of different regions in the feature map after RoI pooling,supplement the pixel information and improve the discriminability of key regions,thus minimizing the loss caused by RoI pooling.This paper tested the proposed two networks on the public data set.Through the comparison experiments,CA-NET and CAA-NET is outperformed with the methods presented in recent years.