Research On The Methods Of Underwater Image Restoration And Object Classification Based On Optical Vision

Underwater vehicles are the important equipment that can work in deep sea.They play an irreplaceable role in the development of ocean.In the underwater operation,the autonomous underwater vehicles obtain clear underwater images and precise object information through the optical visual system.At the same time,the autonomous underwater vehicle can help the manipulator to grasp the different parts of the object with different shape and improve the operational efficiency.Therefore,it is of great theoretical significance and practical application value to study the underwater image restoration and object classification methods based on optical vision and to improve the autonomous ability of underwater vehicle.This paper studies the methods of underwater image restoration and object classification based on the problem of classify the underwater object.The object classification method mainly studies the creation of the visual dictionary and the spatial position information in the bag-of-words model,and provides the accurate object information for the underwater vehicle.In the research of underwater image restoration methods,this paper mainly studies the restoration of blurred images caused by the scattering of light under two conditions of natural light and illumination.This paper analyzes the scattering layered transmission model and restores the blurred images caused by the scattering of light.It cannot obtain the optimal restoration effect directly when use the parameter value obtained from the power spectrum to restore the degraded images under the illumination condition.According to the problem,this paper improves the underwater image restoration method based on the scattering model and introduces the clarity-evaluation function.We find the optimal parameter value in the vicinity of the initial value,so as to realize the restoration of blurred images caused by the scattering of light and the identification of the optimal parameters under the two kinds of light conditions.The experiments verify the effectiveness of this algorithm.In the study of visual dictionary creation method based on the bag-of-word model,the classification accuracy of the traditional bag-of-word model decreases severely when the edge of the object is blurred and the noise is strong.According to the problem,this paper proposes a method based on the contour fragments.In this method,the shapes are decomposed to contour fragments with different resolutions.It will improve the classification accuracy when these fragments are as visual words.The maximum correlation-minimum redundancy criterion is used to remove the redundant words in the visual dictionary.Because of the different use sequence of the criterion's two properties lead to the classification results are not stable,this paper proposes a weighted maximum correlation-minimum redundancy criterion to balance the two properties in order to obtain stable and efficient classification results.The experiments verify the classification accuracy of the improved algorithm is higher than that of the traditional bag-of-word model.In the study of the bag-of-word model with spatial position information,when the model is used to deal with the object with simple shape,the classification accuracy is low due to the lack of relative spatial location information between the contour fragments.In this paper,we study two methods of adding spatial information.Firstly,considering the sensitivity of the spatial pyramid model to the change of the position and size of the object,an improved algorithm is proposed to describe the spatial position information by combining the locality constrained linear coding and pooling processing.Secondly,the classification accuracy rate is reduced when we use the description method based on relative spatial position information because of the random combination of couple words.According to the problem,this paper proposes a method that the adjacent contour fragments are expressed as visual phrases to improve the classification accuracy.According to the problem that the contribution of contour information and spatial location information to image classification is different when the complexity of different types of image shapes is different,this paper defines a weight value to measure the importance of both,and improve the classification accuracy.The experiments verify the classification accuracy of the two improved algorithms is higher.