The Study Of Underwater Object Recognition Techonlogy Based On Polarization Imaging Technology

Object recognition technology is a very active research direction in the field of computer vision,pattern recognition and machine learning,which uses computer to simulate the cognitive learning process of human.Specifically,it extracts the characteristic parameter based on the existing object image so as to a certain regularity can be formed and the object in the new image can be classified.The technology has been widely used in many fields such as security,transportation and the internet.However,there are many limiting factors of optical imaging in underwater environment,which makes it difficult to extract characteristic parameters,so there are few relevant data of underwater object recognition.Polarization imaging technology can eliminate the influence of scattered light,improve the imaging quality and characterize the object,it is suitable for underwater imaging environment.This paper combines the object recognition technology with the polarization imaging technology as the research goal,a new underwater object recognition technique based on polarization imaging is proposed.The identification of material and structure of the object can be realized at the same time.It mainly includes the acquisition of polarizing image of underwater object,digital image processing,and the recognition of material and object based on depth learning technology.Firstly,the polarizing image of the underwater target should be obtained.A polarizing imaging platform is set up,specifically,a phase retarder is introduced as the polarization device,a vertical imaging structure is designed to reduce the effect of mechanical operation and side wall of water tank on imaging quality.Stokes imaging technology is used to obtain multi-dimensional information such as light intensity,space,polarization degree,etc.Each parametric image has strong complementarity and redundancy,which can significantly improve image quality.Secondly,digital image processing is performed to obtain to acquire the characteristic parameters of the object and realize human eye recognition preliminary.The degree of polarization and polarization angle image of the object can be obtained after the polarization image fusion using the computer,extracting polarization information that can not be expressed by ordinary images effectively.The pixel intensity of the polarization image can directly reflect the polarization reflection characteristics of objects.Mapping the degree of polarization and polarization angle image image to other color spaces can convert the grayscale image to a color image,representing the characteristics of the object in different colors instead of grayscale,and is more suitable for the human eye to distinguish the material difference;By drawing the scatter plot of the color of the object,the material recognition can be realized preliminarily according to the position and distribution of the color.Finally,on the basis of color space recognition,the Deep Learning technology is used to intelligently identify the materials and objects.After configuring the deep learning environment,we build the neural network and config learning parameters,and we can get the deep learning process of material recognition start after the format of different material's Pseudo color images are transformed and inputed.The highest accuracy of material identification in the test library is up to 99.7%.Then the intelligent recognition of objects was performed in combination with other features.The accuracy rate of common object recognition in the self-built test library was 79.2%.The recognition method is more objective and accurate,and the result is satisfactory.