Fig 1.
The flower is made up of sepals and petals; there is only one labellum and it may have its own separate color.
Fig 2.
RGB color model as 9 × 13 matrix M.
White is located at the right bottom cell; the blue color, columns 7–9, does not occur in our orchids.
Fig 3.
Color scheme scenario.
Fig 4.
Scenarios for Color of Flower (CF).
Fig 5.
Scenarios for Color of Labellum (CL).
Fig 6.
A selection of pictures of orchids in the dataset used in our research.
Table 1.
The depth of the network model.
Table 2.
Setting of deep learning architecture.
Fig 7.
Multi-class classifier.
Table 3.
The confusion matrices; (a) without normalization and (b) normalized.
Fig 8.
The performance of deep learning architecture on orchid flower dataset.
Table 4.
Accuracy and F1 for the multi-class classifiers.
Table 5.
Accuracy and F1 for the combined binary classifiers.
Table 6.
Accuracy of method 1 by including and excluding inconsistent results.
Table 7.
Accuracy of MLTC and MLCR method on each color scheme.
Fig 9.
Overview of the performance of the various classifiers studied.
Fig 10.
Number of images for each color related to the F1-score of the various classifiers.
Fig 11.
The confusion matrices for primary color: (a) CL1 and (b) CL2 using the ensemble classifier.
Fig 12.
The confusion matrices for primary color: (a) CF1 and (b) CF2 using ensemble classifier.
Fig 13.
The color of the orchids seen in the pictures (purple) corresponds to the literature, whereas the classifier predicted white as their color.
Fig 14.
According to the literature, the orchid should be white.
Yet, the picture clearly shows a pink-purple flower, where the classifier predicted purple as color (which includes pink in our color scheme), which was counted as being incorrect.
Fig 15.
Examples of images that are hard to classify because of photographic imperfections or non-blooming stages of the orchid.
Table 8.
Distribution of misclassified images.
Table 9.
Effect on the accuracy of the classifier after label modification and counting correct predictions in terms of set-membership.
Table 10.
The performance of the multi-class classifier using primary and secondary color after dropping the property of color commutativity.