Research On Remote Sensing Image Object Detection Algorithm Based On Feature Pyramid Attention

With the rapid rise of artificial intelligence technology,object detection has made important progress with the help of a good situation,and it has become a topic for research scholars.Object detection obtains the characteristics of a specific target in the image,and then determines the type of the target and determines the location of the target.In recent years,deep learning technology has become more and more influential.In particular,a series of results and advancements made by convolutional neural networks have made it possible to apply deep learning in more research fields.At the same time,it has promoted the development of computer vision and brought new opportunities to the development of target detection.At the same time,methods based on deep learning have also begun to be applied in the research of remote sensing images.Because remote sensing images are more complex than natural images,detection is more difficult.Therefore,the target detection of remote sensing images still faces many challenges.This article reviews the existing target detection algorithms and finds that the existing algorithms mainly exist in remote sensing images.The following problems are as follows:(1)The missed or wrong detection of small target objects is serious during the detection process.(2)When the similarity between the target object and the background is high,the detection difficulty increases and the detection accuracy is low.In response to the above problems,this paper first proposes a feature pyramid attention module based on the YOLOv3 algorithm,which solves the problem of missed and wrong detection of small targets in the detection process,and then uses the K-means++ algorithm to obtain more matching candidate frames during the network prediction process.The CIo U loss is used to locate the coordinate frame to further improve the detection accuracy;finally,the algorithm in this paper is verified in the remote sensing data set.The main tasks are as follows:(1)To meet the needs of the algorithm in this article,make a data set suitable for the needs of this article on the basis of the public data set,and mark the data set and convert it into a format suitable for the algorithm.In order to increase the number of samples,data enhancement operations are performed to improve the training effect.(2)In response to the problem of high similarity between the target and the background,the SENet network is added to the backbone network to improve the feature extraction ability of the network.By strengthening the network’s attention to the feature information of key areas,the network’s ability to distinguish between the target and the background is improved.(2)Aiming at the problem of low detection accuracy of small targets,a feature pyramid attention module is proposed.By improving the feature information of important factors in the object feature extraction process,the model’s feature recognition ability is strengthened,thereby improving the detection effect of the dense distribution of small targets.(3)In the design of the prediction network,the design and matching of the Anchor box are improved,and the K-means algorithm is improved to obtain a candidate box with better clustering effect;when evaluating the performance of the detection model,position the coordinate box in the error evaluation,the Io U loss function is improved.In addition to considering the change in the area of the overlap area,it is combined with many other factors to improve the positioning accuracy of the prediction frame.Improve the detection accuracy from the above two aspects.(4)Verify the algorithm in this paper,and compare the accuracy of the detection results obtained with other algorithms to further test the performance of the algorithm in this paper.Experiments show that the algorithm in this paper can achieve the effect of improving the detection accuracy and reducing the false detection rate.