A feasibility test for detection of atypical cane samples using near infrared spectroscopy.

MILL RESEARCHERS HAVE noted that in any given season, 1?5% of samples often have unusually low laboratory estimates of Pol in juice (Pij) given the recorded Brix in juice (Bij) value. These ?atypical? samples are of particular concern as they may represent deteriorated or contaminated cane samples. Deteriorated or contaminated cane has a number of negative impacts on the cane milling process. Deterioration in particular can lead to higher viscosity, longer crystallisation times and overall lower cane purity. Many indicators for cane deterioration have been proposed but most are considered expensive, time consuming or unreliable, making them impractical for use during the milling process. Near Infra Red Spectroscopic (NIRS), analysis has been implemented in many Australian sugarcane mills to replace or supplement laboratory analysis of cane quality. However, there is little evidence in the literature that NIRS has been used to classify atypical samples. The purpose of this research was to test the feasibility of predicting possible atypical cane samples using NIRS analysis. Data were collected from a single Australian sugarcane mill from 2006 to 2009. In total, 13 014 samples were collected with Bij, Pij, apparent purity (AP) and NIR spectroscopic data. Atypical samples were defined based on laboratory Bij and Pij values as cane deterioration/contamination data are not routinely measured. A partial least squares discriminant analysis (PLS-DA) was then used to build an NIRS model to identify the defined atypical cane samples. On a test set, the PLS-DA analysis had a correct classification rate of 91.6% of all samples with 86.6% of atypical samples correctly classified and 91.8% of ?typical? samples correctly classified. These preliminary results suggest that it is feasible to predict ?atypical? samples using NIRS. The ability to identify atypical samples in a rapid and non-invasive manner can be useful in quality control measures within the mill and could lead to improved NIRS models specific to these particular samples.