| Rocky desertification is one of the three natural ecological disasters in China.Due to the regional deterioration caused by the unreasonable economic production and living activities of nature and human beings,the original and available land resources are greatly reduced,water resources are polluted and the living conditions of biological resources are lost.In recent years,the country attaches great importance to ecological civilization construction,and rocky desertification is also the most serious ecological problem restricting the social and economic development in southwest China.In order to curb the expansion of rocky desertification,the country has carried out a seriesof comprehensive control projects on rocky desertification,and conducted monitoring and evaluation on the effectiveness of the control.Traditional rocky desertification monitoring methods include visual interpretation,supervised classification,knowledge-based model construction and extraction based on feature information etc.Remote sensing technology is characterized by fast information acquisition,wide detection range and large amount of information etc.As an important means of timely and accurate geographic information acquisition,it has become the main method of dynamic monitoring and investigation of rocky desertification.Traditional rocky desertification survey in small areas mainly extracts related rocky desertification information through remote sensing images,but the precision of the extraction results of this method is greatly uncertain due to the limited quality,resolution and information extraction methods of remote sensing images.As a newly developed machine learning method,deep learning method has been widely used in image recognition.This method converts the traditional feature classification method to deep learning,so as to complete the feature classification of high-resolution images.The bare rock rate is an important index for the classification of rocky desertification,thus calculating the bare rock rate scientifically and quickly is the basis for the classification of rocky desertification and the information extraction of rocky desertification areas.Rock size varies in rocky desertification areas and its color seems close to that of road and bare soil,therefore the more clear aerial images can be obtained by setting the unmanned aerial vehicle(uav).And unmanned aerial vehicle(uav)low altitude remote sensing system can conform to the actual demand of the southwest of rocky desertification,thus it is helpful to acquire remote sensingimage data of rocky desertification and cloudy areas in southwest China.Moreover,uav is characteristized by all-weather,high resolution,lowcost of aerial photography,fast data acquisition,small size and flexible operation etc.Therefore,it can conduct investigation and research on rocky desertification areas in a small range;GF-2 satellite image is the characteristized by high precision and high radiation,as well as spectral information and abundant texture feature information.Therefore,it is feasible to classify natural forest and plantation based on GF-2 satellite image.In this paper,aerial aerial data of uav and gaofeng no.2 image were taken as the main data sources.It is the first time that a deep learning model was used to classify and extract bare rock rates,natural forests and artificial forests in karst rocky desertification areas.The classification results indicate that the total classification accuracy of bare rock fraction extraction is 89.33% and kappa coefficient is 0.82.The total accuracy of classification of natural forest and plantation was 89.22% and kappa coefficient was 0.83.Compared with the traditional classification and extraction methods of bare rock rate,natural forest and artificial forest,the deep learning method combined with uav and high-resolution remote sensing data has high extraction accuracy and strong timeliness.Thus this method can effectively solve the problem of information extraction of karst rocky desertification. |
PDF Full Text Request