Rice Biomass Prediction Model Based On Fractal Dimension And Machine Learning

Rice is the most important grain in Asia,and the study of rice yield is of great significance to the balance of grain supply in China.It is an important subject of current agricultural research to scientifically and accurately predict the growth condition of crops and predict the yield in advance.This work can provide accurate agricultural information for the government and assist the government to formulate agricultural policies with data and models to maintain the balance between supply and demand of crops,which is of great significance to macroeconomic planning.In this paper,fractal characteristic parameters can improve the performance of rice biomass prediction model.Prediction based on image feature model is a popular method of biomass prediction at present,which can realize nondestructive prediction of rice growth status,and is very important for the development of intelligent agriculture.As an important geometric feature,fractal dimension can be calculated from images,but it is rarely used in rice growth prediction.In this paper,fractal dimension is combined with traditional rice image features to improve the effect of prediction model.In this paper,the thresholding method is used to transform the cropped rice image into binary image,and the box counting method is used to calculate the fractal dimension of the image.The characteristics strongly correlated with biomass were selected by calculating correlation coefficients and ranking of importance.Random forest,support vector regression,multiple linear regression and gradient ascending tree were used to establish prediction models for dry weight,fresh weight and plant height of rice.By evaluating the prediction effect of the model,it can be concluded that:(1)Adding fractal dimension to modeling can improve the effect of biomass prediction model.As an image feature of rice,fractal dimension can be used to predict rice biomass.According to the order of scatter plot,correlation coefficient and characteristic importance,the fractal dimension was correlated with dry weight,fresh weight and plant height of rice.(2)Among the models obtained by the four methods,multiple linear regression model and gradient ascending tree model have the best comprehensive effect.The R~2 of the linear regression dry weight prediction model is 0.8697,the R~2 of the linear regression fresh weight prediction model is 0.8631,and the R~2 of the gradient lifting tree height prediction model can reach 0.9196.Meanwhile,MSE and MAPE of multiple linear regression model and gradient ascending tree model are relatively small.The rice biomass prediction model established in this paper has a better effect and has a certain guiding significance for rice research.(3)The growth period of rice will affect the prediction effect of the model,leading to the fluctuation of R~2 of the model,with an average range of about 5%.Although the growth stages of rice are different,the MAPE of the models in this paper can still remain stable,and these models have good stability as a whole.