| In the 21st century,mankind has entered an era of large-scale development and utilization of the ocean.Ships are the main tools for human cognition and exploration of the ocean.The detection and supervision of ships are related to the development of marine resources,the maintenance of marine rights and interests,the innovation of marine technology and the construction of marine ecological civilization.The detection of ships can be used to support maritime search and rescue and combat illegal activities such as fishing,and the wake of the ship can be used to estimate the speed of the ship,which plays an auxiliary role in marine traffic and ship supervision.With the continuous advancement of optical image technology,ship detection and wake extraction based on optical images have become hot spots in ship research.Compared with SAR images and remote sensing images,optical images have the advantages of being easy to obtain,strong intuition,and high resolution.They play an important role in ship supervision.With the increase in the number of satellites,the shortening of the access cycle arid the development of machine learning,ship detection and wake extraction are gradually developing in the direction of rapid detection and machine learning.However,the current research on ship detection and wake extraction is carried out separately,which makes it impossible to use the correlation between them,so that the detection efficiency and accuracy are not high;and in terms of algorithms,very few researchers choose LightGBM Fast and efficient algorithm with high accuracy.In view of the above problems and research status,this paper has done the following research on ship recognition and wake extraction based on optical images.First,use bilateral filtering to reduce noise and smooth the optical image to reduce the interference of sea wave and light extraction to the ship candidate area.Use the K-Means++algorithm to complete the image segmentation,and then perform small area removal and image morphology processing on the segmented image And the hole filling operation to complete the extraction of the ship candidate area;extract the shape and texture features of the ship candidate area,including 5 shape features,Hu invariant moments,LBP features and HOG features,and then use Pearson correlation coefficients and principal components Analysis,reduce the feature dimension to 15 dimensions,and reduce the time complexity of model training while preserving data information;after that,the LightGBM algorithm is used to complete the model establishment of ship detection and ship detection with wakes,in which the ship is on the verification set The detection accuracy rate is 98.3%,and the ship detection accuracy is 95.8%.Finally,the K-Means++algorithm is used to aggregate the ship images with wakes into 3 categories,which are the ocean background,ship targets and ship wakes,and use support vectors.Machine training ship wake extraction model,the final calculated ship wake extraction The accuracy of the model was 96.7%.This paper introduces LightGBM algorithm to ship detection and wake extraction,and organically combines ship detection and wake extraction to improve the accuracy of wake extraction and provides a complete set of technologies for ship detection and wake extraction of optical images The scheme has high accuracy and easy programming.It has strong practical application value and provides new methods and techniques for China’s ship supervision. |
