Abstract
In Malaysia, rice, ranked as the third most crucial crop, faces challenges due to domestic consumption outpacing production, resulting in increased instances of rice adulteration. This underscores the imperative of maintaining integrity and quality standards across the entire supply chain. This study uses an electronic nose, comprising four metal oxide semiconductor (MOS) gas sensors, and employing temperature modulation, Principal Component Analysis (PCA) and supervised machine learning (classification models) to distinguish rice varieties such as Bario, Bajong, Borneo Fragrant, Biris, and Jasmine. The study evaluated 30 classifiers based on their classification and validation accuracy. Sensor data was first extracted from the transient response of sensors output voltage, yielding a 12-dimension dataset with response times of 30 s, 50 s, and 95 s. Classification models trained from this dataset achieved classification (training) accuracy of up to 100% and validation accuracy of up to 96%, where the best performing models are subspace discriminant and kernel naïve bayes classifiers. An attempt was also made to analyze the sensor data frequency response for rice classification. Comparison between the prediction results in the transient and frequency domains showed that transient response is better suited for the classification of rice.
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call