Research On Sea Clutter Suppression And Ship Target Detection Based On Infrared Polarization Information

Detection and recognition of ship targets in complex sea background is a difficult problem faced by infrared detection technology.When the temperature difference between ship and water surface is small,the target is easy to submerge in background clutter.Traditional detection methods based on infrared intensity show great limitations.Infrared polarization imaging is a new technology developed rapidly in recent years,which can make full use of polarization information to complete detection of target.Different from infrared intensity-based imaging technology,infrared polarization-based technology can provide some inherent information of object,such as types of material,roughness of surface etc.In this paper,polarization simulation modeling,polarization characteristic analysis and polarization feature extraction,ship target enhancement based on polarization information,ship detection and recognition based on depth learning are studied.The research results can be divided into the following four parts:(1)Firstly based on the Stokes vector,the degree of polarization and the angle of polarization are obtained.Secondly according to Fresnel formula,energy conservation law and Kirchhoff's law,a new model combining infrared reflected effect and emitted effect is developed to calculate the degree of polarization.With this model,the difference of polarization degree between sea surface and ship target in long-wave infrared is simulated.Then to validate correctness of the simulation some experiments were carried out to measure the polarization degree of sea scenarios.(2)Next according to the analysis on the states of polarization in different scenarios,a method of orthogonality difference is proposed in this paper,which is used to extract the feature of scenarios.The infrared polarization results from infrared emitted radiation effect and reflected radiation effect.The polarization generated by infrared reflection is perpendicularly polarized while polarization generated by infrared emission is parallelly polarized.Using polarization feature in different directions can enhance the detection and discrimination of target.(3)And then based on the polarization difference between sea background and ship target,a method of image fusion based on transform is used to fused polarization image and intensity image.Utilizing the image evaluation indexes of contrast in target to background(C)and local signal-to-noise ratio(LSCR)to evaluate fused image and original intensity image,the results demonstrate that every index of fused image with polarization feature is improved significantly,which validates the effectiveness to enhance dim target with the polarization features extracted in sea scenarios.(4)Finally,the depth convolution neural network and polarization enhancement information are combined to detect and recognize ship targets in the sea background.Aiming at the ship recognition in the sea background,the structure of depth convolution neural network is improved by using ResNet-50 as the front-end of the network to extract features and adopting the optimal prediction frame length-width ratio.Then the infrared sample data sets is enhanced by using polarization information.The detection accuracy of neural network trained and tested using enhanced polarization data sets is improved significantly.