Prediction of mass and discrimination of common bean by machine learning approaches

Beans usually have similar physical attributes; thus, it is difficult to distinguish them manually. Size, shape, and mass attributes of seeds help in breeding, selection, classification, separation, and machine design. This study was conducted to determine physical attributes of 20 bean genotypes with the use of image processing techniques. Color characteristics of the present genotypes were also determined. Then, four different machine learning algorithms (MLP, RF, SVR, and k-NN) were employed to predict seed mass. Among the present genotypes, Güzelöz and Özdemir genotypes had the highest size, shape, and color characteristics. Highly significant positive correlations were encountered between projected area-equivalent diameter (r = 1.00), between geometric mean diameter—surface area and volume (r = 1.00). On the other hand, highly significant negative correlations were seen between sphericity—elongation in vertical orientation (r = − 0.98). In hierarchical cluster analysis for physical attributes, Alberto–Aslan and Aras 98–Şahin genotypes were identified as the closest genotypes. According to PCA analysis, the first two principal components (PC1 and PC2) were able to explain 73% of total variation among the genotypes. While PC1 axis included projected area (vertical), equivalent diameter (vertical), and length, PC2 axis included L*, a*, b*, sphericity, roundness (vertical), and elongation (vertical). Among the present machine learning algorithms, RF yielded the best performances in mass estimation of bean seeds. It was concluded that machine learning techniques increased the efficiency of related machinery and helped to save time and labor.