Can you make morphometrics work when you know the right answer? Pick and mix approaches for apple identification

FruitPhenoBox - a device for rapid and automated fruit phenotyping of small sample sizes

BACKGROUND

Fruit appearance of apple (Malus domestica Borkh.) is accession-specific and one of the main criteria for consumer choice. Consequently, fruit appearance is an important selection criterion in the breeding of new cultivars. It is also used for the description of older varieties or landraces. In commercial apple production, sorting devices are used to classify large numbers of fruit from a few cultivars. In contrast, the description of fruit from germplasm collections or breeding programs is based on only a few fruit from many accessions and is mostly performed visually by pomology experts. Such visual ratings are laborious, often difficult to compare and remain subjective.

RESULTS

Here we report on a morphometric device, the FruitPhenoBox, for automated fruit weighing and appearance description using computer-based analysis of five images per fruit. Recording of approximately 100 fruit from each of 15 apple cultivars using the FruitPhenoBox was rapid, with an average handling and recording time of less than eleven seconds per fruit. Comparison of fruit images from the 15 apple cultivars identified significant differences in shape index, fruit width, height and weight. Fruit shape was characteristic for each cultivar, while fruit color showed larger variation within sample sets. Assessing a subset of 20 randomly selected fruit per cultivar, fruit height, width and weight were described with a relative margin of error of 2.6%, 2.2%, and 6.2%, respectively, calculated from the mean value of all available fruit.

CONCLUSIONS

The FruitPhenoBox allows for the rapid and consistent description of fruit appearance from individual apple accessions. By relating the relative margin of error for fruit width, height and weight description with different sample sizes, it was possible to determine an appropriate fruit sample size to efficiently and accurately describe the recorded traits. Therefore, the FruitPhenoBox is a useful tool for breeding and the description of apple germplasm collections.

Morphological Traits Evaluated with Random Forest Method Explains Natural Classification of Grapevine (Vitis vinifera L.) Cultivars

There are hundreds of morphologic and morphometric traits available to classify and identify grapevine (Vitis vinifera L.) genotypes, while statistical evaluation of those has certain limitations, especially when we have no information about the traits that are discriminative to a certain sample set. High numbers of investigated characters could cause redundancy, while reducing those numbers may result in data loss. Grapevine is one of the most important horticultural crops, with many cultivars in production. The characterization of the genotypes is of undeniably high importance. In this study, we analyzed a dataset of scientific and historical importance with 125 morphological traits of 97 grapevine cultivars described by Németh in 1966. However, the traits are not independent in a set of a large number of categorical traits with too few cultivars. Therefore, the number of traits was first reduced using a simple and effective algorithm to eliminate traits with redundant information content using the asymmetric measure of association Goodman and Kruskal's λ. We reduced the number of traits from 125 to 59 without any information loss. For the classification, we applied a random forest (RF) method. In this way, 93% of the cultivars were correctly classified using only four traits of the data set. To our knowledge, only a few studies applied a trait elimination algorithm similar to ours in ampelography that can be used for other biological data sets of similar structure. The classification results give a morphological explanation to several cultivars from the Carpathian Basin, a territory where all three Vitis vinifera L. geographical groups, occidentalis, orientalis and pontica, are represented. We found that the information-loss-avoiding data reduction method we applied in our study solved the redundancy-caused interdependencies and provided a suitable dataset for classifying grapevine genotypes. For example, this method may successfully be applied in digital image analysis-based traditional morphometric investigations in ampelography.

When do apples stop growing, and why does it matter?

Apples in the commercial food chain are harvested up to two weeks before maturity. We explore apple fruit development through the growing season to establish the point at which physical features differentiating those cultivars become evident. This is relevant both for the understanding of the growing process and to ensure that any identification and classification tools can be used both on ripened-on-tree and stored fruit. Current literature presents some contradictory findings on apple growth, we studied 12 apple cultivars in the Brogdale National Fruit Collection, UK over two seasons to establish patterns of growth. Fruit were sampled at regular time points throughout the growing season and four morphometrics (maximum length, maximum diameter, weight, and centroid size) were collected. These were regressed against growing degree days in order to appropriately describe the growth pattern observed. All four morphometrics were adequately described using log-log linear regressions, with adjusted R2 estimates ranging from 78.3% (maximum length) to 86.7% (weight). For all four morphometrics, a 10% increase in growing degree days was associated with a 1% increase in the morphometric. Our findings refine previous work presenting rapid early growth followed by a plateau in later stages of development and contrast with published expo-linear models. We established that apples harvested for commercial storage purposes, two weeks prior to maturity, showed only a modest decrease in size compared with ripened-on-tree fruit, demonstrating that size morphometric approaches are appropriate for classification of apple fruit at point of harvest.

Machine Learning Applications and Optimization of Clustering Methods Improve the Selection of Descriptors in Blackberry Germplasm Banks

Machine learning (ML) and its multiple applications have comparative advantages for improving the interpretation of knowledge on different agricultural processes. However, there are challenges that impede proper usage, as can be seen in phenotypic characterizations of germplasm banks. The objective of this research was to test and optimize different analysis methods based on ML for the prioritization and selection of morphological descriptors of Rubus spp. 55 descriptors were evaluated in 26 genotypes and the weight of each one and its ability to discriminating capacity was determined. ML methods as random forest (RF), support vector machines, in the linear and radial forms, and neural networks were optimized and compared. Subsequently, the results were validated with two discriminating methods and their variants: hierarchical agglomerative clustering and K-means. The results indicated that RF presented the highest accuracy (0.768) of the methods evaluated, selecting 11 descriptors based on the purity (Gini index), importance, number of connected trees, and significance (p value < 0.05). Additionally, K-means method with optimized descriptors based on RF had greater discriminating power on Rubus spp., accessions according to evaluated statistics. This study presents one application of ML for the optimization of specific morphological variables for plant germplasm bank characterization.

Apples before the fall: Does shape stability coincide with maturity?

Fruit shape is the result of the interaction between genetic, epigenetic, environmental factors and stochastic processes. As a core biological descriptor both for taxonomy and horticulture, the point at which shape stability is reached becomes paramount in apple cultivar identification, and authentication in commerce. Twelve apple cultivars were sampled at regular intervals from anthesis to harvest over two growing seasons. Linear and geometric morphometrics were analysed to establish if and when shape stabilised and whether fruit asymmetry influenced this. Shape stability was detected in seven cultivars, four asymmetric and three symmetric. The remaining five did not stabilise. Shape stability, as defined here, is cultivar-dependent, and when it occurs, it is late in the growing season. Geometric morphometrics detected stability more readily than linear, especially in symmetric cultivars. Key shape features are important in apple marketing, giving the distinctness and apparent uniformity between cultivars expected at point of sale.