Search: image comparison

A scanning device is utilized to obtain a high resolution image of a sample of barley kernels. Algorithms are then applied to detect and segment each individual kernel captured in the image. Subsequently, each individual kernel is analyzed by a Convolutional Neural Network (CNN) with a layer structure that has been specifically selected and developed for analyzing and classifying agricultural commodities. The CNN is trained with verified information (also known as "ground truth") so that it can differentiate barley varieties. The ground truth consists of pure samples of kernels from different barley varieties that were previously digitized using the device and comprises the full data set (artificial intelligence models). In order to obtain accurate artificial intelligence models, the algorithms must be trained to recognize the wide range of variability present in the pure samples, such as those collected from varieties grown in different regions and under varying conditions as well as from various crop years. The purpose of training is to teach the algorithms to understand and detect the patterns unique to each variety that can be used to distinguish it. Once trained, the algorithms are capable of accurately predicting the varietal purity of an unknown sample of barley kernels, provided that the variety has been integrated into the artificial intelligence models.

The method can be employed to reveal slight differences between two samples, e.g., the type and expression of specific attributes or the overall impression as well as for training and selection of individuals for a tasting panel.

Reductones reduce a certain amount of Tillmann reagent (2,6-dichlorophenolindophenol, DPI) within a certain period of time. In the ITT, the time to 80 % decolorization of the indicator solution is determined by visual color comparison.