RESUMO
Near infrared (NIR) spectroscopy was investigated to predict trash content and classify types of ginned cotton by using a fiberoptic in diffuse reflectance mode. Different spectra preprocessing methods were compared, and partial least-squares (PLS) regression was established to predict the trash content of ginned cotton. Discriminant analysis (DA) was used to classify various types of lint and content level of trash. The correlation coefficient r was 0.906 for optimal PLS model using three factors based on first-order derivative spectra, and RMSEC and RMSEP was 0.440 and 0.823 respectively. To classify ginned cotton with and without plant trash, the accuracy rate reached 95.4% using 15 principal components (PCs) via DA, whereas the prediction accuracy rate was only 80.9% for the classification of sample types due to containing foreign fiber, and the classification result for the content level of trash in lint was not good for the samples without any preprocessing. The result indicated that the NIR spectra of sample can be used to predict trash content in ginned cotton, which is often disturbed by type, content and distribution of foreign matters, and the accuracy of some prediction model is unsatisfactory. In order to improve the prediction accuracy, some methods would be applied in future research, such as pretreatment according to acquisition request of solid sample, or using transmission mode.