Font Size: a A A

Research On Cauliflower Weight Prediction Based On Kinect V2

Posted on:2019-03-25Degree:MasterType:Thesis
Country:ChinaCandidate:X X GuoFull Text:PDF
GTID:2393330566971375Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
As a kind of important economic crops in China,the best harvest time of mature of cauliflower need experienced farmers to be discriminated by means of artificial selection.However,this way not only results in yield decrease,but also increased the investment of the farmer`s human capital.Therefore,it is an urgent problem to reduce labor cost and use non-destructive method to predict the production of prenatal crops.At present,the domestic research on cauliflower yield mainly concentrated in the point of biology,irrigation and fertilization based on varieties of cauliflower,these traditional planting method due to can`t accurate prediction of cauliflower weight,to some extent,it can't completely into economic benefits,directly affects the economic benefits of farmers.The3 D model of cauliflower was obtained by using Kinect V2 scanning equipment,the length,width,height,maximum cross sectional area and volume of the model were selected as features.Then,compared and analyzed the weight prediction accuracy of multiple linear regression model,ridge regression model and Lasso model.In order to further improve the weight prediction accuracy of cauliflower,predicted the weight by the support vector regression model,and compared and analyzed.The prediction accuracy of the models with different kernel functions.The method of using sensor to gain 3D model of cauliflower and then predicted the weight by machine learning can satisfies the requirement of using non-destructive method for production forecast of prenatal.These main works of the thesis are as follows:(1)In order to avoid the influence of the scene around on scan data,the data acquisition scene is designed;on the basis of analysis of the technical parameters and depth measurement principle of Kinect V2,the depth camera of Kinect V2 was calibrated.ICP algorithm is adopted to accomplish 3D point cloud registration,through TSDF point cloud fusion algorithm completed the 3D reconstruction of cauliflower,and extract the characteristic information of the cauliflower 3D model,such as length,width,height,maximum cross-sectional area and volume etc..(2)Firstly,established a model for the weight prediction model of the cauliflower by multiple linear regression,ridge regression model and Lasso model are obtained by regularization of the model,then compared and analyzed the multiple linear regression model,ridge regression and Lasso model;using the stochastic gradient descent method to optimized multiple linear regression,ridge regression and Lasso model.The results showthat the generalization ability of the regularized model is stronger,and the prediction accuracy of the Lasso model of cauliflower weight prediction is 73%.(3)In order to improve the prediction accuracy of cauliflower weight,the support vector regression model was established.In support vector regression,the selection of kernel parameter and C parameter had an important influence on the results.The optimal parameters are obtained through cross-validation,and according to different kernel functions to predict the weight of cauliflower.The results show that using gaussian kernel function of support vector regression is the best in predicting cauliflower weight,reaching93.2%.
Keywords/Search Tags:3D reconstruction, Multiple linear regression, Support vector regression, Cauliflower, Weight prediction
PDF Full Text Request
Related items