Chiang Mai Journal of Science

     Coffee is among the economically-important beverage plants. In each year, a great amount of this agricultural product is traded worldwide. For this reason, inspection of coffee bean quality to match the desired level of the customers is a crucial step. Near infrared (NIR) spectroscopy is a non-destructive detection based on the measurement of the electromagnetic radiation in the region between 750-2500 nm. With the detection using a reflectance mode, a number of solid samples can be easily and quickly measured, making NIR preferably suitable for the measurement of various agricultural products, especially coffee. In this research, NIR spectra of green coffee bean samples were recorded, using a homemade NIR system. The Arabica coffee samples were obtained from Chiang Rai province in the northern part of Thailand. Three types of impurity were tested, including broken, insect damage and dried cherry beans. The coffee samples were prepared to have 0, 3, 5, 7, 10, 15, 20, 25, 30, 35, 40, 45 and 50 %w/w of the impurity levels for each test. Therefore, with the three types of the impurity tests, a total of 39 contaminated coffee samples were obtained where the NIR spectra were recorded with 140 replicates to provide an average spectrum of each sample. The spectral data were exploratorily analyzed using principal component analysis (PCA) to investigate their variation. After that, partial least square (PLS) models were established to estimate the impurity levels of the coffee samples. From the PCA score plot, the developed NIR sensor system could be well employed to identify the difference among the contaminated coffee. The PLS models could be used to accurately quantify the impurity levels with acceptable degree of error, demonstrating that the developed NIR sensor system could be used for screening the impurity in the coffee bean products