Research On Key Technology Of Underwater Image Processing And Targets Classification

In recent years,with the global population growth,inland resources consumption is increasingly serious,the marine environment research and exploration has become a new space for human survival.Due to the particularity of the marine environment,the autonomous underwater vehicle(AUV)technology with independent environmental perception and detection performance is widely used.AUV utilizes the visual system to perceive the external environment information for identification,planning and decision-making.Based on the Ministry of Industry and Information Technology(MIIT)project named "ocean exploration research project with autonomous underwater vehicle for engineering",processing methods are studied in this paper for underwater image noise reduction and enhancement,the image object contour extraction and the image target contour recognition classification method.On this basis,the underwater light visual target recognition and classification system is constructed.The main contents of this paper are:(1)Briefly reviews the development of image object recognition methods in and abroad,and points out the research difficulties and research directions of current underwater image object recognition technology.(2)According to the characteristics of underwater images,morphological component analysis method is proposed to decompose the image into low frequency image and high frequency image.The low-frequency image with sparse expression has the main structure information from the original image and the high-frequency image has the characteristics of the main texture,edge and a lot of noise information.A sparse dictionary training method is proposed to denoise the high-frequency image,and single-scale Retinex algorithm is used to enhance the low-frequency texture image.The two-part image processing results are added to restore the final underwater image.The results show that the underwater image preprocessing method proposed in this paper can improve the image contrast and reduce the noise in the image by comparing with the multi-scale wavelet enhancement algorithm with noise suppression.(3)Study the method of object contour extraction for underwater images,an adaptive distance preserving level set evolution method is proposed to extract the contours of the underwater images object.Combine the noise characteristics of underwater images and the weak edge properties arounding the image objects,the thresholds are introduced into the variable energy coefficients of the level set energy functional and Stop function,which can enhance the turbulence interference caused by noise and grayscale nonuniformity in the process of level set evolution,so as to obtain the image object contour under weak edge condition.Through the underwater image segmentation experiments,the results show that the algorithm proposed in this paper can obtain the more accurate underwater object contour,and improve the robustness of the level set method to the noise and uneven gray in the underwater image.(4)The method of image object contour classification was studied,analysis the types and application of image feature description.Combined with the contour characteristics which is extracted by level set method,a Dense-SIFT descriptor with local invariance is proposed as an image feature.Based on Fisher's vector as a semantic representation of image features,according to the large amount of computational complexity,this paper introduces the sparse fisher vector coding with K-nearest neighbor's thought,and utilize the linear support vector machine is used as the classifier.The classification experiment of the typical image set shows that the thinning Fisher coding method proposed in this paper has high classification accuracy and classification speed.(5)This paper constructs the hardware and software architecture of the underwater image target contour recognition system,designs the model,material and panoramic camera matrix design of the underwater target by using the 3D digital simulation method,and establishes the underwater target contour image training library.The effectiveness of the underwater image processing and classification method is verified by the image acquisition and processing of the underwater artificial target.