Research On Target Recognition Method Of UAV Remote Sensing Image Based On Deep Learning

Beijing is one of the cities with many geological safety problems.The types of geological hazards that have great impact on traffic network are debris flow,landslide,rock collapse and other sudden disasters.Beijing's "one ring six radiation" transportation network almost passes through the geological hazard prone areas,which require timely emergency response and rescue.With the development of Surveying and Mapping Science and technology,environmental protection,disaster relief and other needs,the advantages of UAVs are increasingly apparent,they have been more and more widely used in the military field and the national economy.Unmanned aerial vehicles(UAVs)have the advantages of all-weather,high resolution,long distance,real-time,confidentiality,miniaturization and generalization,which make their applications in geological disasters more and more mature.The combination technology of UAV and remote sensing has been widely applied in geological hazard monitoring.Classification and recognition of remote sensing image is the earliest way to distinguish the classification of objects by artificial visual interpretation method.This method requires high professional knowledge and takes a long time.With the development of computer vision technology,image features are gradually applied to remote sensing image classification and recognition,but this method requires a large number of training samples and experts.Knowledge can not be satisfied in practice.With the development of remote sensing means,more and more remote sensing data,if relying on traditional methods to interpret and analyze massive remote sensing data,it can not meet the needs of people,and the accuracy is difficult to guarantee.In recent years,with the rapid development of depth learning in the field of computer vision,it provides a new technical means for remote sensing image scene classification,target recognition,image segmentation and other fields.This paper mainly studies the application of UAV remote sensing image processing and convolution neural network in UAV remote sensing image classification and recognition,in which the classification and recognition objects are different types of vehicles.In UAV remote sensing image processing,the feature points are extracted and matched firstly,and then the target classification of UAV remote sensing image is obtained by global optimization and iterative mosaic.Firstly,the database is constructed by manual labeling,and the typical vehicle is trained by Faster R-CNN and improved YOLOv3 convolution neural network.The UAV remote sensing observation sample generates the recognition model of different types of vehicles,and carries out vehicle detection and recognition,which can better identify different types of vehicles such as domestic cars,buses,trucks and engineering vehicles.Experiments show that the average detection accuracy of Faster R-CNN in UAV remote sensing image is 90.59%,which can meet the detection requirements.The network is improved on the basis of YOLOv3,and multi-scale detection is added to YOLOv3 to achieve the purpose of applying to UAV remote sensing images.Through experimental comparison and analysis,relu function can effectively reduce the training time,increase dropout optimization and regularization constraints can effectively improve the training accuracy.The average accuracy of the improved YOLOv3 model is about 2% higher than that of the YOLOv3 model,and the average detection accuracy reaches 92.25%,which can meet the detection requirements.UAV remote sensing technology can be well applied to traffic network geological safety monitoring and early warning,and subsequently through the establishment of a sound UAV image knowledge base,can improve the automation and intelligence of image understanding,thereby improving the accuracy of UAV image automatic classification and recognition.