Information Extraction Methods Of Subsidence Cultivated Land In High-groundwater-level Coal Mines Based On Unmanned Aerial Vehicle Remote Sensing

The coal resources and cultivated land areas largely overlap in high-groundwater-level coal mine,and water accumulation easily forms on the ground surface which damages and destroys a large amount of cultivated land resources.It brings huge losses to farmers in the coal mining area.Accurate extraction of subsidence cultivated land distribution,area,and damage degree can provide objective reference data for the assessment of cultivated land damage degree,area calculation,and formulation of compensation standards.Low-cost unmanned aerial vehicle(UAV)visible-band remote sensing technology is one of the important ways to obtain data for surveying and mapping production units.In this study,easy-to-obtain and high-resolution UAV visible-band remote sensing images were used as data sources.From the actual situation of the subsidence of the mining area(subsidence degree,distribution of water accumulation),the characteristics of the ground features on the image(spectrum,texture,geometry,etc.),the growth of cultivated land crops(chlorophyll content,plant height)and other perspectives,research has been conducted on the extraction method of subsidence cultivated land.The purpose is to design and build a direct or indirect extraction methods suitable for subsidence cultivated land.At the same time,the depth and breadth of the application of UAV low-altitude remote sensing technology is also further expanded.The main findings are as follows:(1)Extraction method of subsidence cultivated land based on the optimal scale.The image was segmented based on the edge detection algorithm and the scale equals 44.Among the visible-band vegetation indexes,the extraction accuracy of the combination index 2(COM2)was the highest.The sample-based,object-oriented method,the COM2 visible-band vegetation index method and based on the color and texture features extraction method were used to extract the cultivated land information.Compared with the extraction results,it is found that the extraction method based on the color and texture features has the highest correct rate(86%),both the commission(28%)and the omission(14%)are the lowest.To verify the applicability and reliability of the extracted results by using the color and texture features,another area was selected for verifcation in this study.The results show that the correct rate(92.5%)based on color and texture feature extraction is also the highest,both the commission(30%)and the omission(7.5%)are the lowest.(2)Extraction method of subsidence cultivated land based on the hierarchical classification.Regular cultivated land was extracted based on the contrast of 14.83(5.66~25.00)and the length/width of 29.13(4.55~36.74).Scattered cultivated land was extracted based on the entropy of 5.07(3.13~6.83)and the contrast of 6.01(4.58~8.70).Compared with the results of sample-based,object-oriented method and feature combination-based hierarchical classification method,it is concluded that the correct rate of the hierarchical classification extraction is the highest(88%),both the commission(24%)and the omission(12%)are the lowest.The same method was applied to the verification area image.The results show that the correct rate(95%)based on color and texture feature extraction is also the highest,both the commission(20%)and the omission(5%)are the lowest.(3)Extraction method of subsidence cultivated land based on winter wheat growth.The three indexes with the highest correlation to chlorophyll content were selected,including combination index 2(COM2),visible-band diference vegetation index(VDVI),red green blue vegetation index(RGBVI),and the linear,exponential,logarithmic,quadratic,and power functions were used to model the chlorophyll content univariately.The hightest coefficient of determination(R~2)of the model was only 0.624.The results show the accuracy of univariate inversion extraction model is low.A new multi-variable index with the lowest colinearity-the combined visible-band vegetation index(CVVI)was constructed,which composed of these three indexes including COM2,RGBVI and VDVI,and its coefficient of determination reached 0.75,which is higher than that of the univariate model.Based on the verification samples,the coefficient of determination of CVVI index is 0.639,the root mean square error(RMSE)is 1.750,and the normalized root mean square error(NRMSE)is 0.04.The accuracy and stability of the extraction result by using CVVI inversion model is higher than univariate inversion extraction model.The feasibility of extracting crop plant heights in subsidence land from UAV visible image was discussed.The highest height extracted from the established winter wheat plant height model is 47.6 cm,and the lowest height is 8.7 cm.These three indicators including the coefficient of determination,the root mean square error,the normalized root mean square error is 0.537,2.781 and 0.284,respectively.The consistency between the estimated value and the measured value is moderate.(4)The advantages and disadvantages of three methods based on based on the optimal scale,hierarchical classification,and winter wheat growth inversion were summarized.The results of different methods for extracting cultivated land were compared.It is recommended to use the hierarchical classification method as the preferred method for subsidence cultivated land or other information.Finally,the damage degree of the coal mining subsidence cultivated land in the experimental area was divided.