Water Extraction By Using Improved XGBoost Algorithm Combined With HSI Space Transform From Remote Sensing Images

In recent years,as the global water resources environment has been severely disrupted resulted from the frequent occurrence of natural disasters such as droughts and floods on the earth,land surface water resources have become one of the key protection targets in China.The investigation of surface water resources is a topic worth to be discussed in various fields,such as the management of rivers and lakes reservoirs,the protection of freshwater wetlands,the regulation and storage of surface flow and the assessment of water resources quality.With the continuous development of remote sensing technology,surface water information extracted with remote sensing technology has become more and more widely used.Remote sensing technology has the characteristics of wide simultaneous observation range and short information acquisition period,with which surface water information can be obtained quickly and accurately,and be monitored and investigated.It plays a vital role in the rational planning and protection of water resources that surface water information can be gotten in a timely manner and appropriate measures are taken.At present,the main methods for water body extraction from remote sensing images are the maximum inter-class method,the water body index method,and the inter-spectrum relationship method.Due to the fact that the shadows of mountains and the information of water bodies in the “karst” landforms of Guilin,Guangxi often appear in remote sensing images,it is difficult to guarantee the accuracy of the extraction of water information.When water bodies are extracted by using traditional methods from remote sensing images,mountain shadows are often mistakenly recognized as water body information.Therefore,the accuracy of water body information extraction is difficult to reach the desired effect.Based on the deficiencies of the above methods,the applicable algorithm for extracting the water body information in the "karst" terrain was studied with the two Landsat-8 remote sensing images covering Guilin City,Guangxi and Lijiang River Basin in this paper.In order to reduce the interference of the "foreign body homo-spectrum" phenomenon to the subsequent extraction of water bodies,first,different classifiers were built to identify the mountain shadows and water body information in the remote sensing images.The main work of the paper is:(1)By preprocessing the original remote sensing image,fusing the best band combination to synthesize the false color image.After multi-features were fused with the color and geometric features from the training sample data set,the mountain shadow and water information were classified with decision tree,random forest and support vector machine model from the remote sensing images.The experimental results from different algorithms were compared one another.(2)Through the fusion of multiple features,the improved XGBoost algorithm was used to classify the water bodies and mountain shadows from the remote sensing images.The combination of cross-validation and grid search was applied to the XGBoost algorithm,and the average error of the parameters was minimized as the final goal,which avoided the influence of random sampling of training samples on algorithm performance,and also improved the accuracy of parameter optimization.After the classification results of decision tree algorithm,random forest algorithm and support vector machine were compared with one another,the shadow informations and water bodies can be fully extracted from the remote sensing images by the CVGS-XGBoost classification algorithm.The overall classification accuracy was up to 93.9%,It laid the foundation for the subsequent extraction of water bodies.(3)In order to extract the water body information further from the remote sensing images,a new water body extraction algorithm was constructed based on the CVGS-XGBoost classification algorithm and the HSI space transformation.Firstly,the water bodies and mountain shadows were classified from the remote sensing images by using the CVGS-XGBoost algorithm,then the mountain shadows were removed from the images,after that,the result was converted to HSI space to make the image be separated,the water body information was morphologically manipulated in the I space to have the final water body extraction results be obtained.At the same time,the water body extraction results of remote sensing images based on decision tree algorithm,random forest algorithm and support vector machine combined with HSI space transformation were compared.The experimental results show that the water body extraction algorithm combined with CVGS-XGBoost and HSI space transformation has high accuracy in water body recognition,and the water body recognition accuracy rate is as high as 98.4%,which is higher than the current research results of existing papers.In summary,the improved algorithm in this paper can provide scientific data for remote sensing monitoring of water resources protection because of its advantages of high accuracy,applicability,and robustness in the process of water body extraction.