Optical Cognitive Technology Of Autonomous Underwater Vehicle

The vast ocean contains extremely abundant energy,biological and mineral resources.With the rapid development of the human society,the demand for various kinds of energy and resources and the consumption rate are increasing all the time,therefore,the exploitation of ocean energy and resources receives more and more attention from countries all over the world,and a large amount of human,material and financial resources have been devoted in the development of marine technical equipment.Autonomous underwater vehicle(AUV)is one of the most important technical means for human to implement observation and operation in the ocean,especially in the deep sea.The cognitive capability of an AUV,which decides the intelligence level of the AUV to some degree,is an important guarantee for its safe navigation and accurate,effecient mission accomplishment.Optical cognition,which has significant advantages such as high resolution,similarity to human vision system and solid technical foundation,becomes a promising cognitive technology of AUVs,and it has important theoretical and engineering meanings for rapid development and wide application of AUVs.The cognitive technology of AUVs is faced with many challenges and difficulties,and there are some key problems to be solved,therefore,supported by a project “Underwater Information Analysis and Fusion Technology”,research on optical cognitive technology of AUVs is conducted.An underwater laser imaging system used as the main optical sensor,the research work on image restoration,image enhancement,target detection and target identification is carried out to improve the accuracy and robustness of underwater target cognition,and the optical cognitive system and relative methods are verified and evaluated on offline datasets and in test tank environment,respectively.The main contents and contributions of the paper are as follows:(1)Basic mathematical models are constructed to describe the transmission and imaging process of light underwater,and through neron model optimization a novel pulse coupled neural network filtering method is proposed,which effectively eliminates the noise and preserves such the details as textures and edges.The proposed method is verified and evaluated through experiments.(2)The primary reasons for underwater optical image degradation are analyzed.Based on visual attention theory the global saliency of image gray level is analyzed,the gray level frequency is suppressed adaptively,and a histogram equalization method weighted by saliency is proposed that the target regions are selectively enhanced,while background regions are suppressed.The proposed method is verified and evaluated through experiments.(3)The target detection methods based on image segmentation and model learning are analyzed and compared,and A target detection method based on online Hough forests is proposed.Stable,robust simple binary test features are designed for tree models,based on target appearance correlation an online learning method is proposed to boosting the Hough forests,and target location prediction is produced by probability votes of leaf nodes.The proposed method is robust to target appearance variations,interference factors and information loss,and the performance is verified and evaluated through experiments.(4)To solve the problems of basic Adaboost algorithms,an more effective training sample weight updating method is proposed,and a kind of salient,stable,robust compressed Haar-like features is designed according to typical properties of underwater optical images.With consideration of real time requirement,the simple Bayesian classifiers are applied to be basic weak classifiers,and classifier parameters are learned in real time with maximum likelihood estimation.The construced strong classifier is stable to scale variations of training datasets,and the performance is verified and evaluated through experiments.(5)Based on the essential requirements for optical cognition of AUVs,the hardware and software structure of optical cognitive system are constructed,the main information procedure is designed,and after the optical cognition system configuration,the system and relative methods are verified and evaluated in test tank environment.Experimental results demonstrate that the proposed optical cognitive system could achieve accurate,real time,reliable,robust underwater target cognition effectively.