Visual quality assessment of malting barley using color, shape and texture descriptors

Abstract An essential part of the commercialization process of malting barley is the assessment of crop quality. While a traditional indicator of quality is the moisture content, visual inspection is normally carried out as well by an expert. Some indicators of sub-par crops are broken grains, contamination by foreign objects or other agricultural products, discoloration or abnormal pigmentation. We present a complete pipeline of vision and machine learning algorithms aimed at assessing the quality of malting barley grain. Previous solutions based on computer vision have solved this problem to varying degrees, and often require the grains under inspection to be clearly separated. We compare several feature vectors, combined with a nonlinear classifier. Our results show that the Local Phase Quantization descriptor, combined with color and shape features, provides the best results even against improved local descriptors like the Median Ternary Pattern or Median Robust Extended Local Binary Pattern. Our approach is fast, tolerates touching grains and provides an assessment that complies with local industry regulations.