Research On Remote Sensing Inversion Of Salt Content In Saline-alkali Soil Based On BP Neural Network In Songliao Plain

The western part of Songliao Plain is one of the three main saline-sodic soil areas in the world. Owing to the co-effect of natural and man-made factor, salinization in soil is more and more serious, and the environment gets worse and worse, which affect sustainable development severely in this region. Acting as an effective technology for salinity detection in soil, remote sensing has been widely applied in reversing and mapping of salinity in soil. The information of salinity in soil is the key for studying saline-alkali soil scientifically, which plays an instructional role in eco-environment recovery. Studying deeply the information of salinity in saline-alkali soil in remote sensing image, has an important meaning in harnessing saline-alkali soil effectively and avoiding farther soil degeneration, exploiting the resource of saline-alkali soil reasonably and keeping the sustainable development of environment.In this thesis, using eight hydronium contents, the author estimates the salinity in saline-alkali soil, combining with the result of remote sensing image interpretation, analyzes the dynamic change of salinization in Songliao Plain. Through analyzing the spectral characteristic of saline-alkali soil in Aster remote sensing image, the author establishes the statistical model of the soil spectrum and the salinity in saline-alkali soil, and tries to design the salinity model of remote sensing reversion based on BP neural network, which can be applied in preventing and reducing and controlling for salinization in Songliao Plain.On the basis of the chemical analysis of soil sample in Songliao Plain, the optimal model of forecasting the contents of hydronium and salinity with the actual soil electric conductivity is established, besides, the grade of salinization is classified through studying the status of salinity in saline-alkali soil. In this paper, Da'qing and Sui'hua City in Heilongjiang Province, Song'yuan and Bai'cheng City in Jilin Province are chosen as the research areas, based on the analysis results of distribution characteristic of saline-alkali soil and dynamic change of salinization in Songliao Plain.Through abstracting the spectral reflectivity of sample point in Aster remote sensing image, the author obtains the spectral curve of the saline-alkali soil, and studies the spectral curve character of saline-alkali soil. Through analyzing the correlation coefficient, standard deviation and diagnosis index of the former nine bands in Aster remote sensing image, finds that the soil reflectivity of b1, b2, b3 in Aster remote sensing image is more sensitive to salt content in the soil, therefore, this spectrum is more conformable to reverse the salinity in saline- alkali soil in study area.Stepwise regression analysis is made by using the b1, b2, b3 in Aster remote sensing image as independent variable and salinity in soil as dependent variable, and the statistical model of the soil reflectivity data and the soil salt content data is established, which examines the precision of model by the actual sample data. The surface object is classified as non-saline-alkali soil, low-grade saline-alkali soil, secondary saline-alkali soil, heavy saline-alkali soil and saline soil by using decision tree sort, afterwards, the author reverses the salinity saline-alkali soil in study area through using regression equation in order to obtain the spatial distribution map of salinity in saline-alkali soil. The research indicates that the reversion result of salinity in saline-alkali soil and the visual interpretation result in Aster remote sensing image are coincident, which shows that the effect of remote sensing reversion based on statistical model is better and the result corresponds the actual situation. Thus, the research result has a certain value of reference and application.A Back-Propagation neural network model is established which has three layers consists of input layer, hidden layer and output layer, by using the software MATLAB7.0 to design the framework of neural network and using the superior ability for solving the non-linear problem of BP neural network. After that, the remote sensing reversion of salinity in saline-alkali soil is designed based on BP neural network, which uses the b1(520-600nm), b2(630-690nm), b3(780-860nm) in Aster remote sensing image as input, salinity as output, and the nerve cell number of hidden layer is ten, for the sake of realizing the precise reversion of salinity in saline-alkali soil. Through the accuracy evaluation of the neural network reversion model, the author finds that the reversion accuracy of salinity in soil is greatly improved in contrast to that with the statistical model.On the basis of the previous research, four primary conclusions are drawn as follows:1. The relation model of the electric conductivity and the hydronium and the salinity in saline-alkali soil which is suitable for the study area in Songliao Plain is established.2. The diagnosis spectrum which is more sensitive to the salinity in saline-alkali soil is obtained, through analyzing spectral characteristic in Aster remote sensing image.3. The spatial distribution map of the salinity in saline-alkali soil in the study area is acquired, by using multi-linear regression equation to reverse the salinity in saline-alkali soil.4. The RS reversion model of the salinity in saline-alkali soil based on BP neural network in the study area in Songliao Plain offers technique support in RS reversing the salinity in saline-alkali soil in other large-scale area.In this article, the author tries to do remote sensing reversion research of the salinity in saline-alkali soil in Songliao Plain using two mathematic models, and hopes the result can make profitable explore in estimating salinity in soil quantitatively. The research has an important role in lightening the loss of agricultural yield owing to salinization and developing agricultural yield steadily. Besides, several useful improved suggestions for salinazation in Songliao Plain are put forward. The research result of this paper can be acted as reference material of land investigation and agricultural production, which offers the related official department helpful decision-making support in reducing and controlling salinization in Songliao Plain.