Content-based Remote Sensing Image Retrieval Methods And Applications

With the rapid development of space technology,high-precision remote sensing image data is also increasing.Remote sensing image data is rich in information and large in data volume.It has become a major source of spatial information.However,how to quickly and efficiently find the right remote sensing image from the image data,it becomes a problem that needs to be resolved.Traditional image retrieval methods of searching based on keywords,increasing a lot of subjective uncertainties.Now studying content-based remote sensing image retrieval methods avoids subjective description of artificial text.According to the features of color,texture and shape of the image,similar images are looked up.This method is applied in many fields,which has very important theoretical and practical significance.Content-based remote sensing image retrieval method written in this paper.After summarizing the theoretical knowledge related to image retrieval at home and abroad,this paper proposes an overall technical approach for this study,and the corresponding problems encountered in the search and the optimized research plan.The main research contents include the following three parts: extraction of remote sensing image features,similarity measurement between images,and analysis verification of experimental results.we extracted dozens of features using feature extraction methods such as grayscale histogram,spectrum,and shape.Based on the gray histogram of the image,six types of features are extracted,and based on the frequency image,a total of 25 types of features are extracted,such as ring features,wedge features,and length-width features,and so on.Shape features mainly use surf algorithm to extract key points.Relief F algorithm is used to calculate the weights of multiple types of features,the time complexity of image matching is greatly reduced.The organization of images is stored in the form of classification,which speeds up the efficiency of image retrieval.The similarity measure used in this paper improves on the basis of the cosine of theangle,combined with the weight of the feature,the cosine of the angle is superior to the multidimensional vector,and at the same time,it makes up for its insensitivity to the weight of the feature.In experiments,several typical objects are used for image retrieval,sorting the similarity of images,The larger the similarity measure is,the more similar the image is.The accuracy of the retrieval method is ensured by using the precision rate and the ranking evaluation method to test the output of different numbers of images.The image is retrieved using a combination of global features and local features.Adding grayscale histograms based on only spectral features,then add key point features,to compare and analyze the experimental results of different features.The accuracy rate of the image and the result of the sorting evaluation have increased by about 5% on the original basis,which achieved a good retrieval effect.