Barley Defects Identification By Convolutional Neural Networks

The right choice of ingredients, particularly barley, is a key issue in the malting and brewing industry. Nowadays, controlling barley quality involves visual inspection to identify defective or infected kernels. It requires expertise and is labour-intensive. Computer vision solutions sequentially applying attribute extraction and classification algorithms tend to be inaccurate. Deep learning networks combine the two aspects together to enable their mutual adjustment and to increase classification ability. We use this technique to identify the most common defects of malting barley. Two ways of data presentation, two implementations of convolutional neural networks and a handcrafted-features-based method are examined. The classification results are presented, compared and discussed.