Big Data Storage And Query Technology For Ship Recognition Application

In recent years,in face of various applications for ship target recognition,ocean video sensing technology and systems are often applied to supplement other regulatory systems such as VTS,AIS and VHF,due to its intuition,timeliness and effectiveness.At the same time,a lot of high-definition surveillance cameras are deployed,however,massive video data taken by them is still mainly monitored manually.The big data of video surveillance failed to be united for management and sharing where collaboration is difficult,intelligence level is low,information cannot be processed in time,and application mode is inefficient.This paper,based on the big data of the multi-source video surveillance of ocean,studies the storage and query technology of the big data of video surveillance centered on the ship target recognition application.The main research contents are as follows:(1)Based on Hadoop and HBase,a distributed storage and management scheme of the big data on video and image is proposed.The experimental results show that the scheme bears such advantages as high reliability,fast reading and writing speed,and supporting TB/PB data storage and management.(2)A semantic annotation method for ship target images is proposed.The RDF resource description framework is used to convert unstructured image data into structured data to achieve annotation information sharing of the big data of images,which can not only support the big data exploiting of subsequent image,and can also provide an annotated data set for a ship target recognition algorithm based on deep learning.(3)Based on deep learning,a distributed storage and management solution supporting the application of ship target recognition algorithm is proposed.The experimental results show that the solution can not only directly convert the ship target semantic annotation data into various common set format annotation data,and can also store and manage the data,such as features and performance indexes obtained during deep learning training and testing.(4)A ship target image search and visualization method is proposed.The depth Hash network model is used to extract the ship target feature.The similarity calculation is made between the ship target feature and the features in database.And then the track of ship target is extracted.At last,the search results is presented by visualized method.The experimental results show that the method can obtain the retrieval results well and quickly,and it can provide support for further analysis of key ship target behaviors.(5)Hadoop distributed platform is used,HBase is considered as the underlying data storage.Java,JavaScript and other software of developing languages and middle wares is integrated used.We design and implement a set of big data storage and query prototype system platform for ship target recognition application.