Design and Optimization of the Classifier Based on Artificial Neural Network
|School||Anhui Agricultural University|
|Course||Applied Computer Technology|
|Keywords||Neural Network MIV Fruit Fly Optimization Algorithm MapReduce Hadoop|
Classification is an important task of agricultural data mining. The design of theclassifier deeply affects the performance of the classification system. With thedevelopment of digital agriculture, Agriculture presents the diversification and the trend ofregionalization, Then this situation caused the characteristics of agricultural data withmultidimensional, dynamic, nonlinear and instability. The complexity of agriculturalclassification problem is increasing. The classifier established by traditional method is notcomprehensive, science and nature enough to react the complexity, especially in thetreatment of small scale and large scale agricultural dataset. It seriously restricts theclassification and progress of agricultural data development that causes the insufficientamount of information loss and classification accuracy. Therefore, according to thecharacteristics of different scale agriculture data, construct more efficient and reasonable,targeted classifier, to achieve accurate classification of agricultural data, has a positive rolein promoting.The design and optimization method of neural network classifier is proposed witchbased on the two special cases in scale of agricultural dataset, in the analysis of agriculturaldata classification are focused on, data preprocessing of the variable selection, classifierdesign for small-scale dataset, classifier design for large-scale dataset. The methodcombined neural network and swarm intelligence algorithm, neural network and cloudcomputing, realized the classify of different agricultural dataset, experiments is carried outto verify the correctness and effectiveness of the method.The main content and the research achievements are as follows:(1) A MIV variable selection method of neural network is proposed. Throughcalculating the each attribute of data set for the mean impact value of the neural networkclassification accuracy, Select the obvious attributes of the neural network modelingresults，use as the network input variables, and Remove other attributes. To achieve theeffect of variable selection, redundancy elimination, improve the classification accuracy.(2) A neural network classification method of small-scale agricultural dataset isproposed. It based on the shortage of sample data quantity. With the combination of FruitFly Optimization Algorithm and GRNN neural network, the research achieved GRNNneural network implementation key modeling parameters through the foraging behaviorsimulation of the Fruit Fly. It enhanced the accuracy of small-scale dataset classification bycompleted the optimization of the GRNN neural network with the definition of smoothfactor. (3) A neural network classification method of large-scale agricultural dataset isproposed. With the analysis of the traditional BP-Adaboost algorithm, combined withcloud computing platform, a BP-AdaBoost algorithm using MapReduce programmingmode was proposed. The new mode operates on the Hadoop cluster, After the twocomparison experiments on the Hadoop cluster，the practicality of this model is assured. Itcan not only deal with massive data, but also reduce the time complexity of algorithm, witha better linear speedup ratio and accuracy.(4) A neural network classification system for agricultural data is developed. Basedon the Matlab2012(a) platform, the correctness and effectiveness of the proposed methodis validated with programming to achieve the main function, and achieved good results.The research results have certain research value and practical significance forintensive study of classification theory and method in the field of agricultural data. Theestablishment of a more precise and effective agricultural data classifier promotes thedevelopment of digital and precise agriculture.