A multispectral fluorescence imaging system was tested to identify four food products (maize, pea, soya bean and wheat). The system made it possible to record 12 images for each sample by a combination of various excitation and emission filters. Direct observation of images showed that the fluorescence of the four food products made identification possible, although more than one image was necessary to obtain satisfactory discrimination. The images were linearly combined and the most relevant images for identification were determined using a stepwise discriminant analysis and a mapping of the four food products was obtained. In the segmented images, the percentage of well-classified pixels was up to 98.8% for the four products.

The particle size distributions of wheat Ñours collected along a milling diagram were assessed. Nine wheat varieties from the genetic trait T riticum aesticum vulgare and di †ering in hardness were studied for two consecutive years. Break, sizing and middling Ñours were collected for each wheat at di †erent stages in an experimental mill, with all the milling conditions being kept constant. The particle size distributions were measured from 1É5 to 600 lm by using a laser light di †raction apparatus. The distributions were compared by principal component analysis. The method provided a global and synthetic comparison of all the Ñour fractions. In the case of the soft varieties, the distributions exhibited a Ðrst mode around 25 lm, corresponding mainly to isolated starch granules. The distribution of this mode was very low or non-existent for the hard varieties. As the milling conditions were the same for all the wheats, the di †erent proportions measured for this mode were interpreted as being directly representative of the wheat hardness. The proportion of the 25 lm mode was considered as a measure of the ability of the wheat to release starch granules and could be used to follow a grinding or milling process.1998 Society of Chemical Industry ( J Sci Food Agric 78, 237È244 (1998)