Study On Data Preprocess Of Fishery Analysis And System Implementation

In recent years, data mining has aroused great interest in the information industry.The main reason is that in reality there is a large amount of data, which cannot be fullyused, and there is an urgent need to translate these data into useful information. Datapreprocessing is a basic step in completing data mining process, so the datapreprocessing has a direct affection the data mining results. Meanwhile, with thedevelopment of information and space technology, applications of satellite remotesensing in marine fisheries research has drawn people’s attention. Fishery knowledgebased on satellite remote sensing data and marine fisheries yield data have graduallybecome important in fishery industries.Now, experts predict fisheries mainly using correlation analysis, linear regresion,multivariate linear regression, nonlinear regression and various intelligent algorithmincluded neural network. When experts get the data from browser, the operation iscomplicated in the situation of downloading a lot of data; data processing is to usevarious software such as Excel, SPSS, ArcGis. This practice is not really using thestrengths of software, but increases the complexity of data processing.Based on the characteristics of fishery production data and marine satellite remotesensing data, this paper works out a comprehensive preprocessing method of this twokinds of data, which will improve the efficiency to business application level. Includinggetting the data regarding as a step in the process of data preprocessinginnovatively,propose automatic method for getting remote sensing data. The method canbe customized in time and space, to help users get data more easily; improving kriginginterpolation algorithm to enhance the precision for satellite remote sensing data;extracting data from remote sensing image using BP neural network, compared withordinary method, significantly save time and increase accuracy; summarizing marine remote sensing environmental data and fisheries production data at different granularityof time and space scales using concept of tree attributes increase data preprocessingspeed. In addition, the paper adds automatic calculations for commonly used index infishery konwledge and fishery forecasting.Research methods in the paper could be stated as follows:(1) By analyzing the structure of the site, downloading connection string and otherprocedures, automatedly get marine remote sensing environmental data. Datainclude sea surface temperature(SST), sea surface height(SSH), Chlorophyll-a,ect. Data formats include txt, nc, jpg, etc. Oneline data sources includeColombia, microware website, Japan Meteorological Agency and so on.(2) Kriging interpolation algorithm is improved to fill the missing data caused bycloud cover and other reasons. Some kinds of mutation functions are tested inKriging algorithm in order to get the best variogram model, and further improvethe accuracy.(3) Training standard environment color card in remote sensing images by BPneural network to get model data. Using this model to extract environmentalimages solute the problems of matching wrong RGB values and long timecaused by using ordinary method.(4) Using concept of tree attributes summarize marine remote sensingenvironmental data and fisheries production data at different granularity of timeand space scales. Storing marine remote sensing environmental data andfisheries production data at lowest time and space level can quickly obtain anyother time resolution and spatial resolution data.(5) The paper achieve data preprocess of fishery analysis system to businessapplication. Considering the practical applications in fishery knowledge andfishery forescating, the paper adds automatic calculations for commonly usedindex in fisheries, such as GSST, CPUE, HIS etc. At last, using1995-2007big-eye tuna purse-seine data varify the corretness and effectiveness of thesystem, produce a variety of data for users.