Analysis of typical remote sensing tasks

remote sensing technique It is an important means for human beings to carry out scientific investigation on their own living environment , According to the remote sensing satellite load ( sensor ) Different , It can be roughly divided into optical remote sensing and microwave remote sensing . The purpose of remote sensing image interpretation is to make human beings understand their own living environment more comprehensively , By interpreting high-resolution remote sensing images , It can more profoundly reflect the characteristics of the target features and guide human practical activities . in recent years , With the continuous improvement of China's satellite R & D capability , especially “ High marks ” A series of satellites are constantly launched into space , It provides high-resolution remote sensing images with more abundant feature information in the time and space domain , This brings opportunities for the task of intelligent interpretation of remote sensing images . in recent years , Deep convolution neural network (DCNN) Technology has developed rapidly , A large number of based on DCNN The emergence of image processing network , bring DCNN It has become a very important technology in the field of remote sensing image intelligent interpretation . Remote sensing image interpretation can realize the extraction of feature categories and change information .
The classification of remote sensing images is Classify the ground objects represented by remote sensing images according to their attributes , There are two ways to present ： First of all , Census based on scene classification technology . It is different from the traditional natural image classification task , Scene classification is a method to classify different scenes in remote sensing images ( Such as ： residence community 、 road 、 Buildings, etc ) Multi category tasks for . Divide the image rule network , Form slices that contain different scenes (patch), And according to each patch Properties of . second , Detailed survey based on semantic segmentation technology . Semantic segmentation can classify every pixel in remote sensing image , Get the semantic annotation information corresponding to each feature category , So as to outline , The spatial extent of each feature . The higher the resolution of the remote sensing image , The more obvious the role of semantic segmentation . In other words ,“ High marks ” The remote sensing images obtained by the task can be used for more fine-grained classification tasks . therefore , in the light of “ High marks ” data , The results of intelligent interpretation of remote sensing images can be widely used in the field of land resources investigation , Including land resource planning 、 Land cover type statistics 、 Road extraction 、 Water extraction, etc .
Remote sensing image change detection It is a remote sensing image for time series , The process of extracting target change information . be based on DCNN The change detection result can reflect the change of target attributes in the same geographical region 、 Range change, etc , It is of great significance for multi temporal remote sensing image analysis .“ High marks ” The combination of data and this technology , It can be used for urban expansion statistics 、 Disaster prediction 、 Provide technical support in the field of resource management and dynamic monitoring .
Remote sensing image target recognition is the process of extracting the position information and category information of the target to be detected from the remote sensing image . This technology can be used to locate the target 、 Location data statistics and analysis 、 Statistics of target quantity and proportion of different categories, etc . therefore , utilize “ High marks ” data , be based on DCNN The technology of remote sensing image target recognition can be widely used in traffic management 、 Port monitoring and other fields .
notes ： The article is excerpted from 《 High resolution remote sensing image processing and application based on deep neural network 》 Zhang Qiang waits