The use of seed texture features for discriminating different cultivars of stored apples
Introduction
Apples (Malus domestica) belong to the family Rosaceae and are widely cultivated around the world (Yu et al., 2007). Apples reproduce through seeds, and they are self-incompatible. The offspring may not resemble the parent tree, and variability is often observed in the progeny (Cornille et al., 2014). Apple fruits contain numerous seeds which are often regarded as waste material. However, seeds contain oil that is abundant in oleic and linoleic acids as well as phytosterols such as β-sitosterol. Apple seed oil has many applications in the food, oleochemical and cosmetic industries. The content and chemical composition of oil in apple seeds are influenced by cultivar, climate and genetic factors. Cultivar, ripening stages, maturity and environmental factors can modify the composition of volatile organic compounds in apple seed oil (Abbas et al., 2019). According to Yu et al. (2007), apple seeds are a potential source of edible oil and supplementary proteins. The above authors demonstrated that apple seeds contain fat (27.5–28%), protein (33.8–34.5%), linoleic acid (49.6%), oleic acid (39.7%), palmitic acid (7.1%), stearic acid (2.4%), as well as magnesium, potassium, phosphorus, iron and calcium. Xu et al. (2016) found that apple seeds are abundant in phenolic compounds with antioxidant properties and that phenolic concentrations vary across apple cultivars. They concluded that apple seeds are a potential source of functional food ingredients. However, apple seeds contain also different amounts of amygdalin, depending on cultivar, a compound that can be toxic to humans (Bolarinwa et al., 2015). Cultivar influences the properties of apple seeds, which is why fast, inexpensive and non-destructive methods are required to test the authenticity and adulteration of apple seeds.
Computer vision systems support objective inexpensive, reproducible and accurate evaluations of food quality (Abdullah, 2016; Priyadharshini and Akila, 2016). Image analysis is a non-destructive method for identifying seed cultivars, evaluating seed quality and selecting aberrant seeds, and it generates useful information for consumers, producers and distributors (Kapadia et al., 2017). The textural features of seeds can be inferred from the acquired images as a function of the spatial variation of the brightness intensity of the pixels. Images are characterized by repeated subpatterns of dispersion and distribution of pixel brightness, which represent the smoothness, roughness, color, size, brightness, directivity and granulation of seed textures. Textures can carry important information about the structure of the examined physical objects, and quantitative analyses of textural attributes provide valuable insights about product quality (Strzelecki et al., 2013; Armi and Fekri-Ershad, 2019). Image analyses involve the determination of regions of interest (ROIs) where textural features are calculated. Textural parameters can be calculated with the use of various approaches, including the co-occurrence matrix, run-length matrix, gradient, image histogram, Haar wavelet, and the autoregressive model. However, only selected features (texture parameters) contain information that is important for discriminant analyses. Therefore, some features have to be eliminated to ensure that only the most desirable textural parameters are chosen (Szczypiński et al. 2007, 2009; Strzelecki et al., 2013).
The aim of this study was to identify the textural features of apple seeds with the highest discriminatory power for distinguishing the seeds of different apple cultivars with the use of discriminative classifiers. A flatbed scanner was used to acquire seed images and discriminate between apple cultivars in an inexpensive, non-destructive and objective manner.
Section snippets
Materials
The seeds of apple cultivars Gala, Jonagold and Idared were analyzed in the experiment. The apples were purchased in a supermarket in Poland. Before delivery to the supermarket, the harvested fruits had been cold stored for several months. Two hundred and fifty seeds were manually extracted from each apple cultivar. The extracted seeds were cleaned and air-dried. The prepared seeds were subjected to image analysis. The mean seed size was 8.1 × 4.7 mm for cv. Gala, 9.2 × 4.4 mm for cv. Jonagold,
Results and discussion
In the first step, a discriminatory analysis of the seeds of the examined apple cultivars (Gala, Idared, Jonagold) was performed based on selected textures from color spaces Lab, RGB, YUV, HSI and XYZ and color channels L, a, b, R, G, B, Y, U, V, H, S, I, X, Y and Z. Classification accuracies were highest for color spaces Lab and YUV, and color channels b and U. Therefore, only the results of discriminative models based on the textural features from the above spaces and channels were discussed
Conclusions
Selected texture parameters calculated from the images acquired with a flatbed scanner were used to develop highly accurate models for discriminating apple cultivars. The seeds of apple cvs. Gala and Idared were discriminated with the highest accuracy of up to 100%. These results indicate that the seeds of different apple cultivars can be distinguished with practically no error. The seeds of apples cvs. Idared and Jonagold were distinguished with 94% accuracy, apples cvs. Gala and Jonagold were
CRediT authorship contribution statement
Ewa Ropelewska: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing - original draft, Writing - review & editing, Visualization, Supervision.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (29)
Image acquisition systems
- et al.
Determination of amygdalin in apple seeds, fresh apples and processed apple juices
Food Chem.
(2015) - et al.
Classification of cereal grains using wavelet, morphological, colour, and textural features of non-touching kernel images
Biosyst. Eng.
(2008) - et al.
Identification of wheat classes using wavelet features from near infrared hyperspectral images of bulk samples
Biosyst. Eng.
(2009) - et al.
The domestication and evolutionary ecology of apples
Trends Genet.
(2014) - et al.
Wheat class identification using monochrome images
J. Cereal. Sci.
(2008) - et al.
Application of denaturing high-performance liquid chromatography for rice variety identification and seed purity assessment
Mol. Breed.
(2016) - et al.
Identification of nine Iranian wheat seed varieties by textural analysis with image processing
Comput. Electron. Agric.
(2012) - et al.
Discriminating power of selected physical properties of seeds of various rapeseed (Brassica napus L.) cultivars
J. Cereal. Sci.
(2017) - et al.
A software tool for automatic classification and segmentation of 2D/3D medical images
Nucl. Instrum. Methods Phys. Res.
(2013)
MaZda - a software package for image texture analysis
Comput. Methods Progr. Biomed.
Variation in phenolic compounds and antioxidant activity in apple seeds of seven cultivars
Saudi J. Biol. Sci.
Apple (Malus pumila) seed oil. Chapter 24
Prospects of fatty acid profile and bioactive composition from lipid seeds for the discrimination of apple varieties with the application of chemometrics
Grasas Aceites
Cited by (19)
The evaluation of the usefulness of textures from cross-section images obtained using a digital camera and a flatbed scanner for cultivar discrimination of quince (Cydonia oblonga Mill.)
2022, Food ControlCitation Excerpt :The Best first search method allows for the selection of the attributes in a way similar to the greedy hill-climbing using backtracking. The CFS subset evaluator assesses the predictive values of the attributes and the degree of redundancy (Ropelewska, 2020; Witten & Frank, 2005). The applied Multilayer Perceptron was based on a neural network and involved backpropagation.
Cultivar discrimination of stored apple seeds based on geometric features determined using image analysis
2021, Journal of Stored Products ResearchCitation Excerpt :Seed image analysis for the identification of seeds of different cultivars can be applied in taxonomic studies for characterization and discrimination even very close taxa (Bacchetta et al., 2011). Additionally, based on the properties of seeds, the cultivar of apples may be identified (Ropelewska, 2020; Sau et al., 2019). The aim of this study was to develop the discriminative models based on geometric features from images acquired using a flatbed scanner to distinguish seeds of apple belonging to different cultivars.
Computer-vision classification of corn seed varieties using deep convolutional neural network
2021, Journal of Stored Products ResearchCitation Excerpt :In this regard, many studies present techniques for non-destructive seed classification, such as magnetic resonance imaging, electronic tongue, acoustic, electronic nose, and computer vision (Xia et al., 2019). Among these methods, computer vision and image processing can classify crops at a low cost with high analytical and computational power (Patrício and Rieder, 2018; Ropelewska, 2020). Computer vision-based classification consists of four blocks: image preprocessing, segmentation, feature extraction, and classification (Sharif et al., 2018), where feature extraction has a significant effect on classification accuracy (Iqbal et al., 2018).