Research On Anomaly Detection Of Urban Riverine Water Based On Multispectral Remote Sensing Image

In recent years,the environment issues of urban rivers have received increasing emphasis.It is of great significance to carry out long-term management,effective monitoring,and comprehensive management of water quality in urban rivers.In addition to monitoring the river water quality by using monitoring stations,the application of satellite remote sensing image data to monitor,analyze and alarm the spatial and temporal distribution of water quality in urban rivers is a useful supplement for manual detection and automatic station monitoring.Based on the advantages of multispectral remote sensing with large monitoring range and high ground resolution,this paper studies the urban riverine water anomaly detection method based on multispectral remote sensing images.The purpose is to detect and identify the changes of river banks and the water quality anomalies affecting multispectral reflectance from the spatial and temporal scales using satellite imagery information,and to provide information support for the management department of urban rivers.This thesis focuses on the water body(water area)extraction of urban river:s,the detection of riverside line changes in two-phase images,the detection of spatial anomalies in water quality and the detection of abnormal water quality changes in two-phase images.The anomalies including the variation of river banks and integrated water quality anomalies affecting multispectral reflectance,such as sediment pollution and wastewater pollution with black-odor,are detected and analyzed.The main work and innovations of the thesis are as follows:(1)The methods of urban river water body(water area)extraction and riverside line change detection in two-phase images suitable for multispectral remote sensing images are researched.After preprocessing operations including calibration,correction and image fusion of the remote sensing images,using the prior knowledge that the geographical location of the target river is relatively invariable,the methods of river water body extraction and riverside line registration are studied.The water index modeling method is applied to calculate the normalized difference water index(NDWI)and distinguish the water area from background objects in remote sensing images.Image segmentation and similarity matching of the river regions in remote sensing images are conducted using region growing method.The pixels corresponding to the approximate position vector of the target river are screened and taken as seed points in the method.The regional growth criterion is established according to the NDWI,and the water body extraction is realized by the region growing method.The riverside line registration and change detection are performed using the two-phase images.(2)The anomaly detection method for urban riverine water quality on a spatial scale is studied.Based on the local spatial characteristic of the anomalous water quality in urban rivers(characteristics of displaying different performances with surrounding normal waters on multispectral images),a method of detecting spatial anomalies of water quality by combining the dual-threshold anomaly detection based on multivariate normal distribution with two-dimensional sliding window is proposed.After determining the initial training set.the multi-band multispectral data of normal water quality is used to construct a multivariate normal distribution model to achieve water quality anomaly detection in adjacent areas.The two-dimensional sliding window is used to detect different positions in the space and dynamically update the training set.Experiments on anomaly detection are conducted using GF-2 multispectral remote sensing images to verify the effectiveness of using multi-band spectral information to achieve anomaly detection,such as wastewater discharge and sediment pollution.(3)The method for detecting abnormal water quality changes in urban rivers using two-phase remote sensing images is researched to detect new changes in target water quality.Considering that the water quality anomaly detection method based on single-phase remote sensing image may falsely report the inherent spatial differentiation of river water quality as anomaly and is difficult to detect large-scale regional anomalies of water quality in rivers,a water quality anomaly detection method of urban rivers based on two-phase remote sensing images is proposed.The method focuses on the abnormal changes in water quality at different times,and complements mutually with the anomaly detection method on the spatial scale.First of all,using the characteristics that PCA can concentrate information and ICA can realize signal separation with eliminating the high-order correlation between the data,a combination of PC A and ICA is used to extract invariant background regions in two-phase images.Then,based on the linear correlation between the same bands of the two-phase images,linear regression models are established using the spectral data of the invariant background region,and the image to be detected is corrected.Finally,the statistical characteristics of the residual image after correction are used combined with the threshold method to detect the multispectral characteristic changes of the target river and the abnormal changes of water quality,and the experimental verification was carried out.In summary,focusing on the method of urban riverine water anomaly detection based on multispectral remote sensing images,the techniques of multispectral image data preprocessing,target water body extraction,water quality spatial anomaly detection,and change detection in different phase waters are mainly studied in this thesis.The new idea of using high-resolution multispectral remote sensing images for water anomaly detection to assist long-term monitoring and management of urban rivers is discussed.The remote sensing data used in this paper are obtained from the GF-2 satellite of the China Centre for Resources Satellite Data and Application.Relevant work has received substantial support from the related institute of remote sensing technology.