Application of Near-Infrared Spectroscopy for Rice Characterization Using Machine Learning
- PDF / 1,244,553 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 101 Downloads / 175 Views
ORIGINAL CONTRIBUTION
Application of Near-Infrared Spectroscopy for Rice Characterization Using Machine Learning Shagufta Rizwana1 • Manuj Kumar Hazarika1
Received: 24 February 2020 / Accepted: 12 July 2020 Ó The Institution of Engineers (India) 2020
Abstract Near-infrared (NIR) spectroscopy was investigated to relate the intrinsic properties of rice to its extrinsic properties, and thereby, to provide a better solution at the consumer’s level for identification of rice characteristics. Spectral data in the wavelength range of 740–1070 nm are collected with the help of a portable NIR sensor and processed with machine learning techniques were used to develop a rapid predictive model for on-site evaluation of rice quality. Rice properties like glycemic index (GI), amylose content (AC) and viscogram, obtained from laboratory measurements, were mapped to the spectral data employing the machine learning techniques like principal component analysis, linear discriminant analysis, random forest classifier and partial least square (PLS). The regression coefficient and root mean squared error of the PLS model for AC estimation are 0.715 and 1.736; however, a lower value for regression coefficient was obtained for the GI model. Similarly, a confusion matrix of 100% true value prediction was obtained at lower AC values, 83% at high AC values; however, at intermediate range of AC confusion matrix yielded 60% true value prediction. A comparison of classification of rice for parboiling, based on the viscogram and NIR spectral data, revealed that the NIR data produce better clusters with Euclidean distance of 5.46 units between the centroid of the closest clusters, viz., open parboiled and pressure parboiled. The developed model was used to develop a smartphone-based applet for the estimation of AC in rice.
& Manuj Kumar Hazarika [email protected] 1
Department of Food Engineering and Technology, Tezpur University, Assam 784028, India
Keywords Near-infrared spectroscopy Machine learning Rice characterization Abbreviations NIRS Near-infrared spectroscopy ML Machine learning PLS Partial least square ANOVA Analysis of variance PCA Principal component analysis RF Random forest LDA Linear discriminant analysis GI Glycemic index AC Amylose content m.c Moisture content w.b Wet basis
Introduction Rice is a staple food for half of the population of the globe [1]. It is cooked or processed in different ways, and every step and conditions have a certain impact on its eating quality. Also, geographical location has impacts on the rice characteristics [2]. Rice grown in the state of Assam or northeastern states (India) is mostly of medium size and bold or slender in shape distinguishing themselves from the rice preferred and grown in other parts of the country, which are mostly long in size and slender in shape. Dimensions are visually apparent to the consumer and are perceived well to relate to the extrinsic properties. However, there are intrinsic properties in the uncooked form of the rice which have a major impact on the c
Data Loading...
