| Eutrophication is a common water environment problem of lakes,harmful algal blooms outbreak seriously affect agricultural production and people’s life.Remote sensing technology is an efficient way to monitor the occurrence and development of algal blooms quickly and widely,and so provide a strong support for the research and control of algal blooms.In this paper,the identification model about blue-green algal blooms distribution in lake based on ETM+ imagery has been built and validated.34 ETM+ images covering the Taihu Lake,Chaohu Lake and Dianchi Lake of China from the year 2000 to 2012 were selected and normalized by the flat field method.Based on 14 case images,the features sensitive to blue-green algal blooms were extracted and refined,and then used as the key variables.Results and conclusions are as follows:(1)The image features on the blue-green algal blooms and extraction of case featuresBased on the previous research result,total of 1491 initial features related with blue-green algal blooms were selected and extracted,including spectral,GLCM texture and local spatial statistics index.The 14 images with the blue-green algal blooms distribution from previous research were used as case images,and case feature of typical water body,blue-green algal blooms and aquatic plants were extracted and selected.(2)Sensitive case feature selection and determination combining T test with hierarchical clusteringThe T-test process was used firstly to get the significant feature from the initial features,and then hierarchical clustering was used to group the features with the most correlation.According to the separate ability among water,algal blooms and aquatic plants,11 sensitive features in these groups were determined,including five spectral features,three GLCM texture features and three local spatial statistics features.(3)The algal blooms identification model based on case features and model validationBased on the sensitive features,the blue-green algal blooms discriminant model was built by naive Bayes method,with the 97%model accuracy.Random verification and sequential verification of case images showed that the average accuracy is 91.9%.Compared to FAI index and NDVI index,the model has higher precision and lower error rate,and avoided the threshold selection shortages in FAI and NDVI index.Using this model,blue-green algal blooms were identified in the other 20 typical images covering Taihu Lake,Chaohu Lake and Dianchi Lake,in which the three typical images of visual interpretation was selected to evaluate the precision.The result shows that blue-green algal blooms identification accuracy is above 95%,indicating that the discriminant model built in paper has better applicability.On the whole,as for ETM+ images,the blue-green algal blooms discriminant model built in paper is robust and has a good application precision,it can be used to identify blue-green algal blooms effectively from ETM+ images in Taihu Lake,Chaohu Lake and Dianchi Lake.The proposed model is helpful to the rapid remote sensing monitoring of blue-green algal blooms in lakes. |
