Research On Deep Learning Method Of Aircraft Target Extraction In Remote Sensing Images

The target detection of remote sensing images is an important part of the intelligent application of remote sensing data.For the target detection of remote sensing images,there are several difficulties: the first remote sensing image is easily deformed by stretching or compression,and the second spatial background contains Rich in information,the third detection of small targets and stacking together have overlapping targets,the fourth motion taget.These problems have brought great challenges to the automatic detection of remote sensing image targets.This topic starts with the preprocessing of remote sensing images,the processing of small targets,and the expression of features,and conducts detailed research on the above issues.Finally,a multi-scale detection model based on deep convolutional neural network is constructed.This method can automatically learn the characteristics of the object.The structure of the algorithm is clear and the end-to-end learning is realized.Experiments were conducted on the datasets and published excellent datasets to verify their validity.The main research content of this paper is divided into the following three aspects:First,the characteristics of remote sensing data sets were studied and data sets were constructed.These remote-sensing data include sunny days,haze and other climatic conditions,which can more effectively illustrate the robustness of detection algorithms and label samples.Secondly,a multi-scale detection convolutional neural network(CNN)detection model is constructed in context.This method uses the full convolutional layer of residual network(ResNet)to extract more in-depth convolution features,and then constructs the image pyramid and the template pyramid,and detects the target end-toend by associating context features.Finally,on the Caffe in-depth learning and development platform,train the set network.By constructing a cost function,back propagation is implemented and a gradient descent optimization network model is used.In the data set produced in this paper,the precision of aircraft target detection in remote sensing images reaches 89.5%,and the accuracy in NWPU VHR-10 dataset reaches 80%.This illustrates the multiscale feature detection model based on residual network in this paper.The effectiveness of the target detection.The experimental results in this paper demonstrate the superiority of the deep learning algorithm,effectively solve the problem of small and medium targets in remote sensing image plane detection,and the difficulty of detection of deformation targets,promote the application of deep learning in remote sensing images,and further promote the target of aircraft in remote sensing images automatic detection research process.