Method of evaluating diversity of carrot roots using a self-organizing map and image data

Carrot cultivars differ in nutritional value, and the quality of individual pieces may differ from the average, which can be troublesome for companies that process carrots. What is needed is a tool for quick confirmation of the existence and definition of the nature of the differences between the carrot roots. The aim of the experiment was to test whether the simple image parameters, such as colour, will determine the variation within populations of carrot and what kind of information about this differentiation may be provided by a self-organizing map. We indicated that the image data in combination with the self-organization algorithm can provide useful information about carrot roots. However analysis of each individual neural neuron could produce too many summaries to be useful. Therefore, the map was segmented into clusters using k-means method. Segmentation of the self-organizing map allowed to make a comprehensive evaluation of the roots. From the practical point of view such segmentation could help in sorting material, for example, taking into account suitability for processing. The proposed method is sensitive to changes in the processing features of the raw material and is able to locate it in the appropriate area of the topological map. This makes the processor can quickly check whether the raw material he received meets the standards established by a food processing plant. This information is important for the processing industry, in which final product parameters depend on the quality of raw material.

► We assigned carrot roots into classes of similar processing suitability.
► Classification was based on image data related to the colour of the roots.
► Colour was a reflection of information about the chemical composition of carrot.
► We used self organizing map to locate the actual food processing features of the carrot.
► Method is useful in setting the destiny of raw material in the food processing chain.