Study On Infrared Split-Aperture Polarization Imaging Technology

Compared with traditional infrared detection,infrared polarization detection is highly concerned in recent years,because it obtains more information such as the degree of polarization and the polarization angle.Polarization detection plays an important role in target recognition,image enhancement,atmospheric environmental monitoring,biomedical,military and many other fields.This paper studies two cases in details based on the split-aperture infrared polarization detection system,including the acquisition of a polarized image and the target recognition based on polarization information.This paper proposes a phase correlation algorithm based on sub-image to give accurate registration to ensure the accuracy of the polarization image in a split-aperture infrared polarization detection system.The algorithm solves two problems that have negative effects on image registration: the first comes from different polarization angles that give different image characteristics,and the second is that infrared polarization images have high noise and unobvious details.The algorithm includes three steps.In step one,it uses an image preprocessing method to reduce noise and enhance details.In step two,it uses phase correlation algorithm to register the images coarsely,which performs well in solving the first problem.In the last step,the algorithm divides the coarsely-registered images into multiple sub-images and eliminates the high noise subimages to reduce the negative effects of background noise.This paper also provides experiments to verify the registration effect of the algorithm in different scenarios,and their results show that using the algorithm in this paper,in the three scenarios of short distance,medium distance and long distance,the edge of the polarization image is clearer and the normalized mutual information index is increased by 0.067%.Recently,using deep learning method to recognize targets in detection has become a new research direction.The deep learning method uses neural networks with nonlinear relationships to extract and identify target features,which is different from traditional methods using deterministic mathematical models.The deep learning method can deal with multiple features and classify linearly indivisible targets more accurately.This paper uses polarization images rather than intensity images to increase edge information and uses polarization data to migrate the neural network to make it have some recognition ability for infrared polarization images.The experimental results show that the combination of polarization information and intensity information can effectively improve the recognition rate of the target and reduce false recognition.