Water Body Extraction And Application Based On Random Forest Algorithm

With the continuous development of social economy,the demand for water resources is increasing,and the problem of water shortage and water environment deterioration is becoming more and more serious.In order to make further rational use of water resources and protect water and their surrounding ecological environment,more and more attention has been paid to the monitoring of water.As the first step of remote sensing data in water resources monitoring and analysis,how to extract water quickly and accurately is a key problem in remote sensing image application,which has very important practical application value.In order to realize the automatic processing of massive remote sensing images,the development of artificial intelligence algorithm has brought great help to remote sensing image processing.Among them,random forest algorithm,as an important integrated learning algorithm in machine learning,has been widely used in remote sensing image classification research due to its advantages of strong stability,fast speed and being applicable to small samples.In this paper,Gf-1 WFV image is mainly used as the data source.The four bands of the image and the operational feature of the band are used as the input features of the random forest algorithm,and the random forest algorithm is used to mine the feature combination information,build the sample database of intelligent algorithm for water extraction,and improve the extraction effect of the remote sensing image of water range.The research work of this paper mainly includes:(1)The current waterbody extraction technology is summarized and analyzed,the advantages of random forests algorithm and its application in Remote sensing field are introduced.(2)Using GF-1 WFV image,a water body extraction model based on random forest method was constructed.According to the features of GF-1 WFV image,the classification features of random forest algorithm are selected and analyzed,and the features that have a great influence on the classification results are selected,and finally the steady input feature structure is obtained.Secondly,according to the finally input feature structure,the parameters of the random forest algorithm are optimized to determine the number of decision-making trees(ntree)of the random forest model and the number of randomly selected features of the internal nodes of the decision tree(mtry).(3)Building an algorithm sample base.When model files obtained from different training samples and applied to classification,there is a large difference in classification results.Therefore,when selecting the corresponding samples in the remote sensing image,using the spectral features and texture features,the differences between the water and the background information are fully considered,and the types of features contained in the images and the possible forms of various features are analyzed.A sample base of water body extraction intelligent algorithms will be construct.(4)Based on the analysis of the previous chapters,using the selected feature samples,setting the number of random forest algorithm parameter ntree=70,and the internal nodes of the decision-making tree randomly select the number of features mtry=6 to build the model and obtain the model file.Using the obtained classification model file,the water samples of 125 important drinking water sources in the northern region were extracted by random forest method to obtain high-quality data sets.