Prediction of grey mould disease from greenhouse tomato based on radical basis function neural network
Zhi Zhang,Xiaohui Hu,Zhirong Zou
DOI: https://doi.org/10.3785/j.issn.1008-9209.2014.02.012
2014-01-01
Abstract:Summary Tomato is one of the main vegetables cultivated in greenhouse for its rich nutrition and good taste. In recent years,the occurrence of tomato disease brought a big threat to the yield and quality of tomato. Therefore,it is becoming more important to forecast and prevent the disease occurrence. In greenhouse, environmental factors,such as temperature and humidity,have great effect on occurrence of grey mould disease. Environmental factors are interactional in greenhouse,and the relationships between these factors and disease occurrence are complex,nonlinear and not easy to articulate.The traditional methods of mathematical statistics have some limitations in modeling the effects of environmental factors on the disease occurrence.Radical basis function (RBF) neural network is an ideal tool that could be applied to predict the grey mould disease from greenhouse tomato.It has fine characteristics of approximation performance and the global optimum which can overcome the limitations. <br> It is difficult to build up the prediction model with all the involved factors,so the most correlated factors with the disease occurrence should be determined as predictors. A good prediction can strengthen foresight for preventing and treating diseases,which will provide scientific basis to formulate the most reasonable scheme for control diseases.In this paper,we used a neural network toolbox provided by Matlab to establish the RBF neural networks.The data of grey mould disease from greenhouse tomato and corresponding environmental data used in the experiments had been collected from one of the greenhouses in Xintiandi demonstration garden (Yangling, shaanxi Province) during the dates from 2009 04 01 to 2010 05 18.There were 102 sets of sample data in all,among which 90 sets were used to train the neural network models,and the other 12 sets were used to test. The RBF network consisted of input layer,hidden layer and output layer.Grey mould disease grade of tomato was target output,which was expressed by four node vectors to represent four grades.Input vectors had been pretreated with the normalization function ??premnmx”.Mean temperature,minimum temperature and mean relative humidity were selected as input vectors.The highest accuracy and the lowest mean square error from all the variable combination were obtained while the input layer was made up of the three variables. <br> The newrb function was used to establish the RBF neural network model.The spread value was set at 1.0 and the maximum number of neurons was 20.Through cycle test,training accuracy was the highest when spread value was 1.3,which could reach 0.966 7.The curve showed that the predicted output was similar to the expected output.The results of fitting testing of experimental data from other 12 samples with the obtained model showed that the predicted values of 10 samples about grey mould disease grade were the same with their actual values, while one grade higher than actual value could be seen in other two samples.The prediction accuracy was 0.833 3 and mean square error was 0.082 1,respectively. <br> It is concluded that the RBF neural network may be considered as an effective method for forecasting disease grade in greenhouse.And also,the performance of the experiments could be improved if there were more sample data.