How does Computer vision compare to standard colorimeter in assessing the seed coat color of common bean (Phaseolus vulgaris L.)?

Common bean (Phaseolus vulgaris L.) exhibits a wide range of seed coat colors and this morphological trait is widely used in cultivar identification and assessment of diversity within this species. With an advancement in technology and informatics, new methods of assessing seed color are emerging in addition to traditionally used visual observation. Due to a great variety of color measuring techniques, the evaluation of the agreement between methods is needed prior to using the methods interchangeably. Seed coat color in terms of CIE L*a*b* color coordinates of 100 common bean accessions belonging to five mono-colored landraces was assessed using two methods, colorimeter and Computer vision. The percentage difference between the two methods across all samples for L* color coordinate was 5.81%, for a* color coordinate 23.32% and for b* color coordinate 44.44%. According to Bland-Altman difference plot there is a considerable lack of agreement between the two methods. However, using stepwise discriminant analysis revealed that colorimeter method correctly classified 97% of accessions into their respective landrace, while the classification success of the Computer vision was 99%.
Grah (Phaseolus vulgaris L.) pokazuje veliki raspon boja sjemene ljuske te je korištenje ove morfološke karakteristike uobičajeno u identifikaciji kultivara i procjeni raznolikosti unutar vrste. Napretkom tehnologije i informatike pojavljuju se nove metode određivanja boje sjemena uz klasičnu metodu vizualnog određivanja. Zbog velikog broja tehnika određivanja boja, potrebna je ocjena usuglašenosti između metoda prije njihovog kombiniranja ili naizmjeničnog korištenja. Boja sjemena 100 primki pet jednobojnih tradicijskih kultivara graha opisana CIE L*a*b* koordinatama procijenjena je koristeći dvije metode, kolorimetar i računalni vid. Postotna razlika između dvije metode kroz sve uzorke za koordinatu L* iznosila je 5.81%, za koordinatu a* 23.32%, a za koordinatu b* 44.44%. Bland-Altmanov graf razlike pokazuje da postoji značajan nedostatak usuglašenosti između dvije metode. Koristeći postupnu diskriminantnu analizu uspješno je klasificirano 97% primki na temelju podataka dobivenih kolorimetrom dok je klasifikacijski uspjeh računalnog vida bio 99%.