Estimation of Weights in Growth Stages of Onions Using Statistical Regression Models and Deep Learning Algorithm

Abstract

Generally, the problem of predicting yields of onions grown in a year is of utmost concern to both the farmers who grow vegetables and the government departments that manage them. In this study, we first considered five environmental variables, Mean Wind Speed (MWS), Mean Temperature (MT), Mean Ground Temperature (MGT), Mean Humidity (MH), Daily Sunshine (DS) and Daily Rainfall (DR), which have high influence on onion weight at different stages of growth. Second, we use the partial least square (PLS) regression, support vector machine (SVM) regression, multilayer perceptron (MLP) network as statistical prediction model and LSTM network as deep learning algorithm in order to predict the weight of onion using the collected data. Third, we conducted an experiment to investigate the performance of four prediction models for its weight and the influence of six environmental variables on onion growth. Finally, from the experimental results, we first note that the optimal cultivation strategy to increase onion growth is to lower the MWS, MGT and DR below a certain level and at the same time increase the MT, MS and DS values above a certain level. Secondly, we note that for raw data, the weight of onions is not well predicted at the stages of growth by four prediction methods, but for log transform data, it is well predicted during the growth stages. Thirdly, we can also see that the SVMR method is slightly more predictive than the other three methods, PLS, MLP, and LSTM for both raw data and transformation data.