Evaluation of physiological status of potato tubers using hyperspectral imaging

Authors: RADY, AHMED - Michigan State University; GUYER, DANIEL - Michigan State University; Lu, Renfu

Submitted to: Food and Bioprocess Technology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/10/2014
Publication Date: 1/13/2015
Citation: Rady, A., Guyer, D.E., Lu, R. 2015. Evaluation of physiological status of potato tubers using hyperspectral imaging. Food and Bioprocess Technology. 8(5):995-1010.

Interpretive Summary: In potato processing, the color of chips is closely related to the level of reduced sugars in potato tubers, which include glucose and fructose. If the level of reduced sugars exceeds a certain level, the browning phenomenon will take place in potato chips, which may render the chip product unmarketable. Currently, laboratory methods, such as gas chromatography, high performance liquid chromatography and gas chromatography-mass spectrometry, are routinely used to measure these sugars. These methods, however, are time consuming and require expensive instrumentation, which cannot meet the rapid, real-time quality control need for the potato processing industry. Other less expensive methods are available for assessing the glucose, but they are suitable for only testing a few samples. Over the past decade, hyperspectral imaging has emerged as a powerful technique for rapid, nondestructive quality assessment of agricultural and food products. The technique allows us to acquire spectral information in the spatial dimensions, which would be potentially useful for rapidly measuring sugars in potato. This research was therefore intended to evaluate the feasibility of hyperspectral imaging in reflectance mode for measuring glucose and sucrose for two cultivars of processing potatoes (i.e., Russet Norkotah and Frito Lay 1879). Potato tubers were samples in the 2009 season and one tuber slice was cut from each tuber. Hyperspectral reflectance images were acquired for each tuber slice and multiple features were then extracted from these images. Thereafter, wet laboratory chemistry tests were used to measure the glucose and sucrose contents of potato tubers. Mathematical models were developed relating the hyperspectral image features to the glucose and sucrose content. Excellent correlations were found for the glucose of Russet Norkotah with correlation values as high as 0.97, whereas those values were only 0.81or lower for Frito Lay 1879. Lower correlations were found for the sucrose content of both cultivars. This research showed that hyperspectral imaging is potentially useful for rapid measurement of glucose content and, possibly, sucrose content in potato tubers.

Technical Abstract: Visible and near-infrared hyperspectral reflectance imaging was evaluated as a rapid technique to predict the glucose and sucrose percentages in two common fresh use and chipping potato cultivars. Tubers were sampled in the 2009 season and held in multiple storage temperatures in attempt to develop broad constituent distributions. Each tested sample was a 12.5 mm thick slice cut uniformly from all tubers. Multiple features were extracted from the hyperspectral reflectance images of potato samples, including mean reflectance spectra and feature parameters obtained from an exponential model that fitted the reflectance scattering profiles. The glucose and sucrose contents were measured using the Megazyme sucrose and D-glucose assay procedures, respectively. Partial least squares regression (PLSR), feed forward neural networks, radial basis functions neural networks, and exact design radial basis functions neural networks were used for building prediction models. PLSR models built using mean reflectance spectra showed excellent correlations for the glucose of Russet Norkotah (RN) with correlation (R) values as high as 0.97, whereas those values were only 0.81or lower for Frito Lay 1879 (FL). Sucrose models showed lower correlations with R values as high as 0.60 for FL and 0.38 for RN. Wavelengths selection using interval partial least squares (IPLS) and genetic algorithm (GA) was conducted on the data. These methods yielded results close to the full wavelength models for glucose and sucrose for both cultivars, with IPLS being preferred as it used fewer variables than GA. The results from the research showed the potential of using hyperspectral imaging for rapidly measuring glucose and, perhaps, sucrose in potato tubers for quality control and monitoring in the potato industry.