Research On Maritime Small Target Tracking Of Staring Satellite Sequence Images

In recent years,maritime security has gradually become an important focus and development direction of China’s national defense and security.Once local conflicts occur at sea,the real-time acquisition of the marine situation is an important factor for mastering the strategic initiative and making correct decisions.As China’s first staring satellite with high temporal resolution and high spatial resolution,the GF-4 satellite shows great application potential in maritime target monitoring.This paper is mainly about tracking ship targets in GF-4 satellite sequence images.The specific contents are as follows:The first part of this paper is about the background suppression of GF-4 satellite sequence images.Firstly,the water area is extracted by comprehensively using the reflection difference of ground objects and the initial information of ship targets,so as to reduce the impact of redundant information,such as land and thick clouds,on the subsequent target tracking processing.Secondly,the dark channel prior method is optimized to model the background of the sequence images,and the additive noise in the images is removed.Finally,the optimized Retinex method is used to remove the multiplicative noise in the sequence images and realize ship targets enhancement.In the second part of this paper,a target tracking algorithm based on the sequence images’ spatio-temporal information is proposed.Firstly,analyze the target tracking model based on image correlation,and take the imaging characteristics of GF-4 satellite sequence images into account when calculating the sizes of ship targets’ neighborhood in different bands of the sequence images.Secondly,extend the gray gradient distribution feature to extract the multi-scale features of ship targets’ neighborhood in different bands of the sequence image,so as to enhance the correlation of ship target features in the adjacent images.Finally,ship targets’ high-dimensional and multi-scale features are mapped into the kernel space through kernel method,which makes them suitable for the target tracking model based on image correlation,reduces the computational complexity of the algorithm,and improves the target tracking effect.The third part is to optimize the tracking results with the characteristics of ship targets.Firstly,the actual situations of ship motion are taken into considered to constrain the results of ship target tracking,and then find the abnormal nodes of ships’ tracks.Secondly,according to ship targets’ historical data,the unscented Kalman filter model corrects ship targets’ incorrect locations.Finally,ship detection is completed by calculating the significance of the sequence images to improve the accuracy of ship targets tracking.