Aircraft Detection Method Research Of Remote Sensing Image Based On Deep Learning

With the development of high-resolution remote sensing image acquisition technology,a large number of high-resolution remote sensing images containing rich feature information have yet to be interpreted.The traditional object detection method can not meet the current detection efficiency.In recent years,deep learning object detection methods have made breakthroughs in the field of images.Applying deep learning object detection methods to high-resolution remote sensing image detection tasks has become a hot research topic.The existing deep learning object detection methods are mostly based on the general data set in the field of computer vision for training and testing.The general data is based on real life photos,and there are many kinds of object objects in the image,and the object object is large in volume.The high-resolution remote sensing image of the airport area studied in this paper has the characteristics of small volume,large number,variety and shape of the object,and the detection accuracy of the object detection model is higher.Therefore,the existing deep learning object detection methods cannot meet the detection requirements.According to the characteristics of high-resolution remote sensing image,the airport is used as the detection scene,and the aircraft is the detection object.A deep learning aircraft object detection network framework is designed for the object detection task of high-resolution remote sensing image of the airport.The deep learning aircraft object detection network framework proposed in this paper is divided into four main parts: extracting feature map,generating candidate frame,pooling candidate region,object discrimination and detection frame regression.For different detection scenarios and training data sets,the aircraft detection model and the airport detection model are trained by designing different network structures and hyperparameters for the deep learning aircraft object detection framework.The network structure of the aircraft detection model is shallow.The image features are extracted by the DGG network and the UGG network proposed in this paper.The aircraft detection model is used to verify the detection advantage of the existing deep learning object detection model in high-resolution remote sensing images.The model extracts image features from the Inception-Resnet-v2 network and the Resnet-101 network with deeper network structure,which is used to verify the object detection effect of the model on large-scale airport areas,and evaluates objectively with evaluation indicators.The experimental results show that the deep learning aircraft detection framework proposed in this paper can be successfully applied to the object detection task of high resolution remote sensing image.The aircraft detection model has greatly improved the detection range and detection accuracy compared with the existing deep learning object detection model.The airport detection model can accurately and comprehensively detect aircraft objects in a wide range of airport areas with an optimal F1 score of 0.9763.The model has strong robustness and can be applied to full-color remote sensing images and multi-spectral color remote sensing images with resolutions of 0.5m and 2m.In addition,the model has a good detection effect on aircraft objects with incomplete features in the image range.