Research on Machine Learning Models for Maize Hardness Prediction Based on Indentation Test

Abstract

Hardness is a critical mechanical property of grains. Accurate predictions of grain hardness play a crucial role in improving grain milling efficiency, reducing grain breakage during transportation, and selecting high-quality crops. In this study, we developed machine learning models (MLMs) to predict the hardness of Jinsui No.4 maize seeds. The input variables of the MLM were loading speed, loading depth, and different types of indenters, and the output variable was the slope of the linear segment. Using the Latin square design, 100 datasets were generated. Four different types of MLMs, a genetic algorithm (GA), support vector machine (SVM), random forest (RF), and long short-term memory network (LSTM), were used for our data analysis, respectively. The result indicated that the GA model had a high accuracy in predicting hardness values, the R2 of the GA model training set and testing set reached 0.98402 and 0.92761, respectively, while the RMSEs were 1.4308 and 2.8441, respectively. The difference between the predicted values and the actual values obtained by the model is relatively small. Furthermore, in order to investigate the relationship between hardness and morphology after compression, scanning electron microscopy was used to observe the morphology of the maize grains. The result showed that the more complex the shape of the indenter, the more obvious the destruction to the internal polysaccharides and starch in the grain, and the number of surface cracks also significantly increases. The results of this study emphasize the potential of MLMs in determining the hardness of agricultural cereal grains, leading to improved industrial processing efficiency and cost savings. Additionally, combining grain hardness prediction models with the operating mechanisms of industry machinery would provide valuable references and a basis for the parameterization of seed grain processing machinery.