| Affected by biological characteristics,complex environment and many other factors,shallow water aquaculture has been difficult to achieve accurate monitoring,detection and utilization for a long time.Although China’s shallow water aquaculture has developed for a long time,the degree of aquaculture informatization and intelligence is low,and the technology of big data mining and analysis is weak.Due to historical and practical reasons,aquaculture data has the characteristics of diverse resource approaches,complex structure,uneven quality,wide application scope and low overall quality of data.The existing shallow water aquaculture mostly depends on labor,which also makes it impossible to make effective use of data resources.In recent years,with the diversified development of aquaculture big data technology and ecological aquaculture,Through the interdisciplinary way,a large number of data are analyzed and mined,and finally the results are presented to the production decisionmakers in the form of intuitive interface,which has gradually become an idea to solve the above problems.Therefore,it is particularly important to systematically analyze the big data of shallow water ecological aquaculture,summarize the sources and acquisition methods,integrate the data of the aquaculture industry and the whole process industrial chain,and then realize the automatic decision-making of the whole industry,so as to provide theoretical and technical support for the reform and upgrading of China’s aquaculture industry.Facing the key problem of big data analysis of shallow water ecological aquaculture,due to the complexity and diversity of the data itself,this paper puts forward higher requirements for data processing.This paper takes this as the starting point to study the low efficiency of large-scale data and model training in the scene of aquaculture biological image classification from two aspects: multi computer data communication and GPU parallel method,Firstly,the optimization of data transmission process is realized through the principle of interval pairing of nodes,and a ring all reduce data communication algorithm with improved time complexity is obtained,which is used to improve the transmission efficiency between data parallel multiple devices and alleviate the bandwidth loss of traditional parameter server parallel structure;Secondly,taking advantage of the characteristics that the weight parameters of the traditional deep learning backbone network are smaller than those of the full connection layer,the synchronization overhead is small,and the weight of the full connection layer is huge,and the gradient transmission overhead is too high,the data of the backbone network is processed in parallel,and the full connection layer adopts model parallel processing,and a hybrid parallel algorithm is proposed,It solves the problem that the data parallel mode is difficult to support large-scale network parameters and accelerate and delay.Experiments also show that compared with data parallel,this method has greater acceleration advantages on finsh4knowledge,and is more suitable for parallel training of massive breeding image classification data.In addition,based on the analysis of aquaculture data,this paper further built a shallow water intelligent ecological aquaculture platform,and completed the overall technical framework design of the intelligent information platform for the whole process of shallow water ecological aquaculture and the business process planning of each functional module,And the exploration of key technologies such as visual display technology and big data platform construction and the realization of unit technology points;The research on key technologies such as perception layer data access technology,crawler technology,intelligent streaming media technology and application layer page rendering has been carried out,and the mobile end online monitoring system and shallow water ecological aquaculture intelligent service platform have been constructed,basically realizing the real-time monitoring of ecological aquaculture,production process management,information release,environmental early warning The system design and business process planning of intelligent decision-making service and other functions provide the demonstration area with the whole process information platform support of collaborative linkage between mobile terminal and platform terminal. |
