Algal Bloom Discrimination Method Based On Spaceborne SAR Images

In recent years,due to the frequent human activities and the rapid development of industry,the eutrophication of lakes in China is becoming increasingly serious,which leads to the frequent outbreak of algal bloom,and seriously affects the daily life of people in the surrounding areas.Algal bloom refers to the phenomenon that algae grow in large numbers and gather on the surface of water body,which is a typical manifestation of eutrophication.With the development of science and technology,satellite remote sensing technology plays an important role in the field of environmental monitoring.Among them,optical remote sensing technology is widely used,but it is susceptible to cloud,rain,fog and light.The spaceborne Synthetic Aperture Radar(SAR)is a microwave remote sensing technology,which has the ability of observing the earth in any kind of weather all day long,and can effectively make up for the shortage of optical remote sensing.Therefore,it is of great significance to monitor algal bloom based on spaceborne SAR image.Algal bloom area in SAR image presents a more obvious "dark spot".However,the dark spots in SAR image are not all caused by algal bloom.Lake surface under low wind also presents a "dark spot" state in SAR image.There are many technical challenges to distinguish different types of "dark spots" in SAR image and identify the algal bloom area.At present,the research of algal bloom discrimination based on SAR image is rare.The few existing researches are mainly carried out in the traditional way of artificial feature extraction,which does not make optimal selection of dark spots features and can not avoid the impact of negative features on the accuracy of algal bloom discrimination.At the same time,the method of algal bloom discrimination based on artificial feature extraction has the problems of low efficiency of feature extraction,easy to be affected by human factors and difficult to obtain deep features of image data.In order to solve the above problems,the methods of SAR image algal bloom discrimination by means of feature optimization and automatic feature extraction are studied in this dissertation.Based on the detailed analysis of the multiple features of two kinds of dark spots,the feature sample set and image sample set of SAR image dark spots are constructed.Based on the two technical routes of feature selection and automatic feature extraction,the improved algal bloom discrimination method is proposed,and its effectiveness is verified by using feature sample set and image sample set.The research work of this dissertation mainly includes the following three parts:(1)The image sample set and feature sample set of dark spots in SAR image are constructedThe time series sentinel-1A SAR images are preprocessed by radiometric correction and geometric correction.Based on the preprocessed SAR image,the dark spot image set is obtained by using cutting,fine segmentation.Then data enhancement technology is applied to build the dark spot image sample set.At the same time,the feature extraction work is carried out for the dark spot image set to build the dark spot feature sample set.(2)A method of algal bloom discrimination based on artificial feature selection is proposedAiming at the problem of negative features in artificial feature extraction,a method based on artificial feature selection for algal bloom discrimination is proposed.In order to improve the effectiveness of the feature set in the process of algal bloom discrimination,this method carried out the optimal selection processing for the extracted feature set.The feature set of dark spots in SAR image contains 22 features,including angular second moment,contrast,entropy,reciprocal difference moment,correlation,and so on.According to the correlation between features and tags,negative features are removed by using the Relief F algorithm.The finally the optimal feature set containing only 10 kinds of dark spot features is obtained.Back propagation(BP)neural network is used as discrimination classifier to verify the proposed method.Compared with the complete feature set,the results show that the discrimination accuracy of the optimal feature set is improved by 19.38%,which reach of 81.39%.(3)Two methods of algal bloom discrimination based on automatic feature extraction are proposedIn view of the problems of low efficiency of feature extraction,easy to be affected by human factors and difficult to obtain deep features of image data in artificial feature selection method,based on automatic feature extraction,two methods of algal bloom discrimination methods are proposed.One is Algal bloom Discrimination method based on Improved Inception V3 structure using SAR image(ADII-SAR),and the other is Algal bloom Discrimination based on Improved Dense Net structure using SAR image(ADID-SAR).In order to extract texture features more accurately,the pooling operation is modified in these two methods,replacing the mean pooling with the max pooling.These two methods can extract the deep features in the image more efficiently.Among them ADID-SAR makes full use of the feature information of each layers,giving a smoother decision-making boundary,and has higher discrimination accuracy in small data sets.The discrimination accuracy of ADII-SAR and ADID-SAR is 85.08% and 87.30% respectively,which is 4.60% and 6.64% higher than that of mean pooling,and 3.69% and 5.91% higher than that of proposed artificial feature selection method.