Research On Classification Technology Of HJ-1 Image Sea Ice Based On Fuzzy Decision Tree

Sea ice disaster is one of the major marine natural disasters in China, which has a great impact on the coastal aquaculture, fishery production, transportation, oil and gas production and the production and life of the residents of the island. The type of sea ice can indicate different phases of the ice’s growth. It can be used to invert the ice’s thickness, and give better support for the sea ice prevention by making the sea ice forecasting and monitoring more accurate. The satellite “HJ-1” designed by China is very suitable for the sea ice disaster forecasting, because it can accept the spectral range that the sea ice reflect and generate the image of sea ice type for further analysis.However, the analysis of the images now is mainly based on manual calculation and classification, which is inefficient. There is a need for a new technology that can classify the sea ice type automatically, make the analysis efficient, and improve the sea ice forecasting and monitoring.According to the actual business needs of the North Branch Prediction Center,this thesis discusses the automatic classification technology for sea ice types in the“HJ-1” satellite image using Gray-level Co-occurrence Matrix(GLCM), Cluster Analysis, Fuzzy Decision Tree, etc. Then a HJ-1 satellite sea ice automatic classification algorithm has been provided. The main research contents of this thesis include the following aspects:(1) This thesis researches the characteristics of sea-ice image and the theoretical knowledge about how to extract the image characteristics, focuses on the use of Gray-level Co-occurrence Matrix method for the image characteristics extraction. This thesis builds a sea ice type characteristics data set by extracting the representative characteristic of sea ice image.(2) This thesis discretizes the continuous attribute value of the sea-ice type classification data set by using clustering analysis algorithm, and determines the triangular membership function parameters by the cluster center, so as to complete the fuzzy processing of the sea-ice type classification data set. It provides basic data for the construction of the sea-ice classification models.(3) The sea-ice data set is divided into two parts, training set and test set. This thesis constructs the decision tree of the sea-ice type classification by using training set and fuzzy ID3 algorithm, tests the accuracy of decision tree by using the test set, and extracts classification rules of the sea-ice type based on the decision tree.(4) According to the business needs of the North Sea Forecasting Center, this thesis develops a HJ-1 satellite sea ice automatic classification system, which can automatically classify the HJ-1 satellite sea ice type according to the sea ice classification decision tree proposed by this thesis.