Identifying maize cultivars by single characteristics of ears using image analysis

Definitions and screens of novel organ characteristics are among main concerns in differentiation of similar cultivars in maize (Zea mays L.) DUS (Distinctness, Uniformity and Stability) test. To screen image features of ears in maize for potential traits of DUS testing, RGB images for fifty well-developed upper ears of 4 maize cultivars were photographed. Features of 145 ear traits were exacted using image analysis from into 4 categories of size, shape, color and texture. The correct identification rate (CIR) derived from statistical discrimination analysis using single image features was taken as the identifaction power. CIR’s of 145 extracted ear traits varied from 0.244 to 0.634 with the mode of 0.35. The 17 traits out of 145 ones performed well with over 0.5 correct rate in discriminating cultivars, and could be promising candidate traits for DUS test. They consisted of 6 shape features of ear length to width ratio, compactness, shape factor, eccentricity, surface sphericity and body sphericity, and 8 textures of the first, second and sixth invariant moments, uniformity, and 4 mean or variance statistics of Fourier Transform, and 3 colors of ear green skewness, kernel crown red mean and row gap proportion. One of these 17 traits is in the list of the National Guide, but none of them belong to size category. The generalized heritability (GH) of tested traits ranged from 0.00 to 0.66. Robust regression analysis revealed a highly positive relationship between CIR and GH (CIR = 0.29 + 0.44*GH, r = 0.897, P<0.01). Conclusions are that novel image features of maize ears may be promising candidate characteristics for DUS test, and that potentials of 4 categories for variety identification power rank as shape > texture > color > size.