Identification of Potato Genotypes Using Digital Image Analysis

• Published in: Journal of Central European agriculture Vol. 12; no. 1; pp. 195 - 214
• Main Authors: CSÁK, Máté, HEGEDŰS, Géza, POLGÁR, Zsolt
• Format: Journal Article
• Language: English
• Published: University of Zagreb, Faculty of Agriculture 01-01-2011
• Subjects: cluster analysis; digital image analysis; fractal; genotype identification; non-hierarchic cluster analysis; potato breeding; spectral fractal dimension
• ISSN: 1332-9049; 1332-9049
• DOI: 10.5513/JCEA01/12.1.899

Based on the fractal analysis of digital images, a new classifying system has been proposed at the Potato Research Centre of Keszthely. It is a qualifying system generating objective values to distinguish potato varieties or detect quality differences within the genotype in a relatively simple way. The goal of the research project was to investigate whether Spectral Fractal Dimension (SFD) value of digital images is applicable to describe various quality characters of potato tubers and whether SFD values could be used for the identification of certain varieties – if so, which conditions were the most important to enable this process. Considering the above aims, we developed an evaluation computer program which determines the SFD values of the 4 conditions of potato tubers: skin colour; raw flesh-colour; boiled flesh-colour; greying of flesh-colour after 24 hours in RGB spectrum and in all of its sub-spectrums (R, G, B). In total 2080 digital images of 13 varieties from 4 examining periods were analysed. Based on our results we can conclude that SFD analysis can be used in potato breeding only when digital images were made under well-determined, standardized conditions. Detailed statistical analysis (hypothesis tests, principal component analysis and non-hierarchic cluster analysis) showed that SFD was not suitable for qualifying tuber characters within a genotype. When images were examined for different years and the same genotype, it became evident, that there are significant deviations between years and within same genotypes. We could conclude that the identification of genotypes should be related not to one particular SFD value, but to the control of the given year with the known value. When analyzing the differences between genotypes on yearly basis, irrespective of characteristics or the studied spectrum, we could not significantly separate genotypes, although there were some that could be separated, even though genotypes and their characteristics changed every year. It cannot be stated either that by combination of the values of different characteristics and spectrums, separation is not possible. We used non-hierarchic cluster analysis to solve this problem. As a result of the method, the separation of genotypes was successful every year, so by summarising the joint RGB SFD value of 4 characters with the values of additional spectrum the separation will be complete. The system could be utilized for research purposes and further research is needed to achieve practical applicability.