Research On Remote Sensing Monitoring In The Typical Salt-affected Zone Of The Yellow River Delta Based On BP Neural Net

The Yellow River Delta has broad development prospects with rich land resource, but serious soil salinization affects local agricultural production and poses a threat to the stability of the ecological environment. Improvement and utilization of saline soil in the achievement of the regional agricultural and ecological sustainable use of resources is of great significance, while improving saline land is on the premise that the extent of land salinization has been known. Remote sensing technology is used in investigation and mapping about soil salinization with the measured spectrum data and multi-spectral image data. So it is possible to monitor salinization fast, accurately and comprehensively.This thesis mainly researches on salinity inversion and identification of halophytic vegetation from remote sensing with the measured spectral data and multi-spectral image data, using remote sensing technology and BP neural network technology. First, a field survey has been taken. Saline soil samples were collected where different vegetation grew. The coordinate data and spectrum data were obtained in the place of corresponding sample points, by the handheld GPS detector and AvaField spectroradiometer. Soil gravimetric method was used for determination of soil salinity. Then, the spectral characteristics of saline soil and halophytic vegetation are analyzed. For spectral data of the saline soil, correlation analysis and multiple linear regression analysis are used to determine the better band combination which represents the relationship between saline soil spectral data and salt content. BP neural network model is established and accuracy test is done. The result shows that BP neural network method is feasible to simulate and predict soil salinity by means of remote sensing. For Vegetation spectral data, principal component analysis (PCA) is introduced on spectral data to extract and sorting variables, distinguishing effectively a variety of halophyte. Textural features are extracted by the method of gray level co-occurrence matrix, using multi-spectral ALOS image. Classification and recognition is carried out by BP neural network method, on the synthetic image which is with spectral characteristics and spatial characteristics. Classification accuracy is improved. Finally, the distribution map of regional salinity vegetation types is achieved after category image is edited. The experiment confirms that BP neural network can simulate the relationship between soil salinity and spectral data, which it is feasible to use this method in salinity inversion from remote sensing. The distribution map of regional salinity vegetation types which is obtained based on BP neural network with texture feature extraction provides the scientific basis for reasonable development and utilization of local saline soil resources. This study will promote quantitative development of the research on regional salinization by means of remote sensing.