Based on diverse landforms, the correlation between soil organic matter content and multi-spectral band of remote sensing image was analyzed in this pa- per. In addition, the inversion models were built for the soil o...Based on diverse landforms, the correlation between soil organic matter content and multi-spectral band of remote sensing image was analyzed in this pa- per. In addition, the inversion models were built for the soil organic matter content in different landforms. The results showed that the spectral reflectance was nega- tively related to soil organic matter content; linear regression analysis of remove was performed throughout the bands using SPSS. When the inversion models were built based on all the bands, better fitting effect was obtained. The precision of in- version models built based on different landforms was higher than those built re- gardless landforms. Compared with the actual value, the identification level of soil organic matter content was 91 65% under the allowable error was 7%. It indicated that the extraction of soil organic matter with inversion model that was built based on different landforrrs was feasible with higher precision.展开更多
The LIBS （Laser induced-breakdown spectroscopy） combined with BPNN （Back propagation neural network） was applied in rock sorting and distinguishing for 26 rock samples of 6 types. According to contents of major el...The LIBS （Laser induced-breakdown spectroscopy） combined with BPNN （Back propagation neural network） was applied in rock sorting and distinguishing for 26 rock samples of 6 types. According to contents of major elements in samples, we selected lines of Si, Al, Fe, K, Ca, Mg, Na, Ti and Mn. These lines of 9 elements composed three characteristic spectral models which were the WSLM （Wide spectral line model）, the PM （Peak model） and the PRM （Peak ratio model）. The first and the second characteristic spectral model were divided into 9 kinds, as follows： the characteristic spectrum with 1 element, the characteristic spectrum with 2 elements, we can deduce the rest from this and the last one has 9 elements. The third model was divided into 8 kinds which were using AI as reference element. We analysed spectrums of the three models by BPNN. Experimental results shown that whether sorting or distinguishing these samples, identification accuracies of the PM were more than that of the PRM overall, the same as the WSLM did to the PM. While the selected number of elements was 5, 6 or 7, the identification accuracy of the WSLM could reach more than 90%. Continuing to add the number of elements to improve identification accuracy was not very obvious.展开更多
文摘Based on diverse landforms, the correlation between soil organic matter content and multi-spectral band of remote sensing image was analyzed in this pa- per. In addition, the inversion models were built for the soil organic matter content in different landforms. The results showed that the spectral reflectance was nega- tively related to soil organic matter content; linear regression analysis of remove was performed throughout the bands using SPSS. When the inversion models were built based on all the bands, better fitting effect was obtained. The precision of in- version models built based on different landforms was higher than those built re- gardless landforms. Compared with the actual value, the identification level of soil organic matter content was 91 65% under the allowable error was 7%. It indicated that the extraction of soil organic matter with inversion model that was built based on different landforrrs was feasible with higher precision.
文摘The LIBS （Laser induced-breakdown spectroscopy） combined with BPNN （Back propagation neural network） was applied in rock sorting and distinguishing for 26 rock samples of 6 types. According to contents of major elements in samples, we selected lines of Si, Al, Fe, K, Ca, Mg, Na, Ti and Mn. These lines of 9 elements composed three characteristic spectral models which were the WSLM （Wide spectral line model）, the PM （Peak model） and the PRM （Peak ratio model）. The first and the second characteristic spectral model were divided into 9 kinds, as follows： the characteristic spectrum with 1 element, the characteristic spectrum with 2 elements, we can deduce the rest from this and the last one has 9 elements. The third model was divided into 8 kinds which were using AI as reference element. We analysed spectrums of the three models by BPNN. Experimental results shown that whether sorting or distinguishing these samples, identification accuracies of the PM were more than that of the PRM overall, the same as the WSLM did to the PM. While the selected number of elements was 5, 6 or 7, the identification accuracy of the WSLM could reach more than 90%. Continuing to add the number of elements to improve identification accuracy was not very obvious.
