Monitoring Of Soil Organic Matter In Coal Mining Tensile Fracture Zone Based On Machine Learning And Spectral Transformation

Coal mining not only brings economic growth,but also causes surface subsidence and the formation of tensile fissures,resulting in the destruction of soil structure and the loss of water,soil and nutrients.It seriously affects the living environment of residents around the mining area and affects agricultural production.As organic matter is one of the most important components of soil nutrients,it is particularly important to monitor the change of organic matter content under the special environmental impact of the fracture area.The traditional way of monitoring soil organic matter is often costly and inefficient,and the rapid development and wide application of hyperspectral technology provide great convenience for soil organic matter content monitoring.In this study,the surface tensile fracture area of Zhuzhuang mining area of Huaibei City was used as the experimental research area,and the sensitivity of different bands to soil organic matter was compared and analyzed based on different forms of spectral transformation methods,and the characteristic bands that were sensitive to soil organic matter were screened by using soil samples collected in the field and the spectral reflectance measured by ASD high-resolution spectrometer indoors.Hyperspectral inversion models for estimating soil organic matter content were established using partial least squares(PLSR),BP neural network(BPNN),support vector machine regression(SVR),extreme learning machine(ELM)and random forest(RF).(1)Based on spectral transformation,partial least squares and BP neural network,a hyperspectral inversion model of soil organic matter in the tensile fracture zone was constructed.The FD-PLSR model has the highest accuracy in the built PLSR model,while the SD-BP model has the highest accuracy in the BP model.Among them,the R~2and RMSE of FD-PLSR model modeling set and verification set were 0.8761,0.4972,0.8459 and 0.6806,respectively.The R~2 and RMSE of the SD-BP model modeling set and verification set were 0.7842,0.6955,0.8111 and 0.8137,respectively.(2)Based on the spectral transformation of different fractional orders and three different machine learning methods,compared with the original spectrum(0 order)combined with machine learning method,the modeling and verification effect is obviously better,indicating that the preprocessing method using fractional differential can improve the modeling accuracy.(3)The study found that the correlation coefficient between the spectral reflectance values of many bands and the soil organic matter content was high.The original spectral reflectance was subjected to fractional differential pretreatment,and support vector machine(SVR),extreme learning machine(ELM)and random forest(RF)algorithms were selected to estimate soil organic matter content.Among them,the 1.4-order SVR model has the best modeling effect,R~2=0.9449,RMSE=0.6323,RPD=3.367.Figure[15]Table[7]reference[81]...