Varietal classification of barley by convolutional neural networks

Assuring high quality of beer is a key issue in the brewing industry. This can be achieved by controlling the brewing process and the quality of the ingredients. The currently employed visual inspection of malting barley requires expert knowledge and is time-consuming as well as expensive. It has been demonstrated that computer vision algorithms can replace human expertise in the detection of defective grains or in recognition of barley varieties. However, varietal classification based on colour, texture and morphological attributes returned accuracy of less than 75%. In this paper, nine implementations of convolutional neural networks are examined in their application to varietal classification. The comparison covers deep learning models and transfer learning models with regard to learning and classification times, computational requirements and classification accuracy. The use of convolutional neural networks enables barley classification with accuracy exceeding 93%. (C) 2019 IAgrE. Published by Elsevier Ltd. All rights reserved.