Development of Corn Moisture Content Prediction Model Using Portable Spectrometer and Machine Learning
Dimas Firmanda Al Riza1*, Harki Himawan1, Choirul Umam2, Nurwahyuningsih3, and Zaqlul Iqbal1

1 Department of Biosystems Engineering, Faculty of Agricultural Technology, Universitas Brawijaya, Jl. Veteran 65145, Malang, Indonesia
2Faculty of Agriculture, Universitas Trunojoyo Madura, Indonesia
3Food Engineering Technology Study Program, Department of Agricultural Technology, State Polytechnic of Jember, Indonesia
*dimasfirmanda[at]ub.ac.id

Abstract

Corn moisture content is one of the parameters used to determine harvest time. The standard method of measuring corn moisture content is destructive (oven method), which has limitations in terms of time, labor, and measurement accuracy. A better method is to use infrared wave optical sensors, but the cost is expensive. There were several stages carried out, the first stage as many as 50 corn samples were dried using a drying house. The drying process was carried out for two weeks, and data acquisition was carried out once every two days from day 0 to day 14. Data acquisition was carried out which included measurement of corn moisture content and corn spectrum using a portable spectrometer. Then, a prediction model was developed from the data using Partial Least Square Regression (PLSR) and Artificial Neural Network (ANN) methods. The best model from the PLSR analysis results show that preprocessing Moving Average (MA) with R2 Train and R2 Test of 0.9843 and 0.9826 and RMSE error values from each Train and Test of 1.8661 and 1.9743. Then the best results from ANN show R2 Train and R2 Test of 0.951 and 0.905 with RMSE error values of 3.234 and 4.853. Based on the results of the study, the portable spectrometer has the potential to be developed into a detection tool for predicting corn moisture content in the field.

Keywords: Machine Learning, Moisture content, Non-destructive, Spectroscopy

Topic: Agricultural Engineering