Experimental design of sea ship tracking based on SiamFC
Synthetic aperture radar(SAR)is a high-resolution imaging radar that works under any weather conditions without environmental constraints.It has wide applications in military and civilian fields.Maritime ship tracking based on SAR images mainly involves identifying a ship in a SAR image and monitoring its trajectory in subsequent SAR images,thus achieving the monitoring of the marine field.It is currently an important branch of remote sensing image information processing and has extremely important research value.This study aimed at the characteristics of coherent speckle noise in SAR images and the difficulty in capturing ships due to their large moving range.To improve the accuracy of ship tracking on the sea,an experimental scheme for ship tracking based on SAR images was designed.First,three pairs of continuous satellite SAR images were used to construct a dataset.Then,according to the characteristics of smallships on the sea and the layered structure of the SiamFC algorithm network,a SiamFC algorithm was proposed fortracking ships on the sea.Because of the relatively simple feature extraction network structure of SiamFC,which was more suitable for tracking small targets such as satellites and ships,this study adjusted the parameters of the SiamFC algorithm and trained and tested a model using a constructed dataset.The model was trained in a completely offline mode,and the tracking process did not update the model parameters.Only the trained parameters were used for tracking.The test results showed that the model is feasible.To achieve a better tracking effect,a gamma operator was introduced to enhance the SAR images.The coherent speckle noise in SAR images was reduced,enhancing the details of ship targets.This was mainly reflected in the fact that the grayscale value of the ocean was lower compared with ships and land after SAR image enhancement,resulting in a more pronounced shape and size of ships.The Gamma Map filtering algorithm was used to improve the model,which reduced a large amount of useless information in the image,the pressure of feature extraction,and the error of correlation convolution operation so that the model could effectively track the target according to the characteristics of the target.This effectively improved the performance of the model and solved the technical difficulties of SAR image tracking.Finally,the improved model was tested,and the experimental results showed that the average accuracy of the improved model on test set A increased from 12.7％to 19.6％.When the coincidence rate met at least 40％,the accuracy of the improved model reached 30％,which is an increase of more than 10％.When the test set was B,the average accuracy of the model increased from 39.0％to 41.7％.When the coincidence rate met at least 40％,the accuracy of the improved model increased from 55％to 61％,increasing by 6％.The improved method proposed in this study achieved the expected goal.The results showed that the ship tracking accuracy of this scheme was improved and that the new scheme had good tracking performance.
deep learningsynthetic aperture radartarget trackingSiamFC algorithmgamma operator
万方数据
维普
