Research On Method Of Remote Sensing Image Agricultural Information Extraction Based On Visual Information Mining

In recent years,with the development of domestic high resolution remote sensing satellite technology,e.g.,Gaofen series and Ziyuan series satellites,remote sensing images are more widely used in various large-scale production fields.During the third national agricultural census,using high-resolution remote sensing images in the survey area for extracting information,surveying and evaluating agricultural resources has become an important part of the census work.A main research content of remote sensing application is to use computer to extract information from remote sensing image according to the demand,and improve classification accuracy and utilization rate of remote sensing image.In this context,this thesis summarizes the relevant studies of theory and application of remote sensing image information extraction,and uses the high-resolution remote sensing image of regions in Hubei Province in 2016 to study the agricultural land information extraction from the remote sensing image based on visual information mining.The main research contents are as follows:(1)Through the experiment and analysis of multi-dimensional feature extraction method,the features of remote sensing images in target areas are mined and corresponding feature datasets are the constructed,to express the target feature specification.(2)In view of the limited number of suitable remote sensing object data sets for neural network training,a high-dimensional feature learning method is proposed for remote sensing images with a small number of labeled samples by using Feature Mining and Neural Network,according to the principle of learning and constructing image features in the training process of neural network.(3)Based on the feature dataset,the CNN and the FCN are used for the scene analysis and semantic segmentation of high-resolution remote sensing images,and the accuracy of information extraction are analyzed.The experimental results show that using feature mining to build a reasonable depth learning training sample set can achieve better results in image information extraction tasks in specific areas,which provides a new perspective for the application of remote sensing image in the field of national condition monitoring.