Predicting the Heating Time of Palm oil using Optimal Selection of Color Parameters and Machine Learning

Abstract

The heating time of palm oil can affect its quality indicators such as free fatty acids (FFA), smoke point (SP), anisidine value (AnV), induction period (IP), polar compounds, color, etc. Prediction of the heating time of palm oil in high temperatures is guidance for monitoring its quality. This paper proposes a computer vision model that can rapidly predict palm oil's heating time at a typical frying temperature (180℃). Firstly, we use YOLOv3 to detect palm oil samples in the images. Secondly, we extract the color parameters of palm oil and construct five kinds of feature vectors: (), (), (), () and (). Thirdly, we use Random Forest Regressor, Random Forest Classifier, SVR, SVC, BP Neural Network to construct heating time prediction models and make a comparison. Finally, we select the best prediction model combined with YOLOv3 to detect palm oil samples and predict their heating time. The results show that when (,,) is used as the color parameter and SVC is used as the heating time prediction model, the prediction accuracy is the highest, reaching 97.2%.