Classification of Cotton Trash with Features Extracted from Fluorescent Images

A fluorescent imaging setup consisting of blue LED and the UV LED excitation sources with an SLR camera was constructed based on the excitation/emission analysis by fluorescence spectroscopy. Under the blue LED excitation light the following cotton trash types were imaged: bark, brown leaf, bract, green leaf, and hull. Under the UV LED excitation light the following cotton trash types were imaged: paper, plastic packaging, seed, seed coat (inner), and seed coat (outer).  Images of botanical and non-botanical cotton trash on top a lint layer were acquired and subjected to a series of image processing steps to extract the information from the regions of interest. The analysis of images considered two color models: RGB and HSV. From each of the color models, specific image ratios and channels were extracted and subjected to statistical analysis to determine their potential for classification of cotton trash. Linear Discriminant Analysis (LDA) was applied in order to see whether image ratios and channels can be used as classification inputs. Classification rates of 100 % were achieved for paper and plastic packaging, and rates of at least 80 % were achieved for green leaf, hull, and outer portion of the seed coat.