Research Of Ship Detection And Classification Based On Feature Fusion

The research of ship detection and classification methods has always been a research hotspot in the conmercial and ruilitary fields,which is of great significance for improving sea traffic order and maintaining marine safety.However,the influence of wave,weather and the diversity of ship’s shape and structure make this problem a difficulty in the research field.A ship detection and classification method based on feature fusion is proposed utilizing the characteristics of the complementary features of Histogram of Oriented Gradient(HOG)and Completed Local Binary Pattern(CLBP)in this thesis.In order to study in-depth,the ship detection and classification problem is refined to two parts:ship detection and ship classification.Then,the relevant analysis is carried out according to different research purposes of the two questions.The research on the two parts is as follows:(1)Ships in an image may vary in size depending on their type or location.The feature fusion will easily make the feature dimension higher,whieh will slow down the detection speed.To solve the above problems,the image pyramid sliding detection method is used to realize multi-scale detection of ship targets,and then reduces the dimensions of features of two groups by increasing the size of HOG cell and block and using uniform pattern CLPB calculation method.(2)For the reason that single-scale features does not have spatial feature information of local objects in images,a multi-scale HOG feature is proposed,which can describe the spatial structure change of the ship and effectively enhance the ability of feature representation.Multi-scale CLPB features are achieved by changing the resolution of the image.The above two sets of features use the same scale change values,and then feature fusion is used for ship classification.Support Vector Machine(SVM)is used for classification,and experiments are done on open datasets to verify the effectiveness of the proposed method.The experimental results show that the HOG-CLBP ship detection method is obviously better than the HOG method,and the fusion CLBP feature dimension is low,which has little impact on the detection time,indicating that the method has high practical value.Multi-scale features have higher classification accuracy than single-scale features,indicating that multi-scale features are effective for improving ship classification accuracy.Feature fusion can fully combine the effective information of the two sets of features,further improve the classification accuracy and the classification effect is superior to many existing algorithms.