Document Type : Research Paper
Authors
1 Department of Food Science and Technology, Tuyserkan Faculty of Engineering and Natural Resources, Bu-Ali Sina University, Hamedan, Iran
2 Department of Mechanical engineering of Biosystems, University of Jiroft, Jiroft, Iran
3 Department of Energy Engineering, National University of Distance Education, Madrid, Spain
4 Department of Agriculture and Medicinal Plants, Nahavand Higher Education Complex, Bu-Ali Sina University, Hamedan, Iran
Abstract
Keywords
Main Subjects
EXTENDED ABSTRACT
Medicinal plants have gained considerable attention as valuable natural resources in scientific and industrial communities. Iran, with its high diversity of medicinal plants, particularly Bunium persicum (Persian cumin), holds a significant position in this field. Bunium persicum, a herbaceous plant, grows in various regions of Asia and Europe. One of the main challenges in the cumin market is fraud, where authentic cumin is often mixed with lower-value species. Therefore, there is an increasing need for precise methods to distinguish genuine cumin from counterfeit ones. Image processing, as a non-destructive and innovative method, has been widely used in the food industry for quality evaluation and product differentiation. Color and texture features play a crucial role in food quality assessment, and artificial neural networks (ANNs) are suitable tools for classifying these complex features due to their capacity for data analysis.
In this study, nine cumin samples were collected from the local market of Kerman, consisting of five authentic Bunium persicum samples and four counterfeit samples (including European black cumin, artificially colored green cumin, and two types of low-quality mixed cumin). The images of the samples were captured using a digital camera, fluorescent light source, computer, and image processing software in a controlled environment. After image acquisition, the samples were converted into different color spaces (RGB, HSI, HSV), and their color and texture features were extracted. The color features included the mean, variance, standard deviation, and range of color components. The texture features were extracted using the Gray-Level Co-occurrence Matrix (GLCM) and included contrast, homogeneity, correlation, energy, and entropy. A total of 36 color features and 108 texture features were considered for analysis. To reduce complexity and enhance model accuracy, important features were selected using factor analysis. Then, a Multilayer Perceptron (MLP) neural network with the Levenberg-Marquardt backpropagation method in MATLAB software was used for classification and recognition. The data were divided into three subsets: training (60%), validation (20%), and testing (20%).
The classification accuracy using color features varied. The highest accuracy (93.51%) was achieved using the sigmoid transfer function in the hidden layer and a linear output function. The reduced color features, selected using factor analysis, resulted in improved classification accuracy. For texture features, the highest classification accuracy (95.86%) was achieved with a network using 108 texture features. The network with the sigmoid transfer function in the hidden layer and a linear output function showed the best performance. Texture features outperformed color features in classification accuracy. The combination of color and texture features resulted in better performance, with the classification accuracy reaching 95.95%. The reduced combined features, obtained through factor analysis, provided greater accuracy than using color or texture features alone. Factor analysis was employed at all stages to reduce the number of features and eliminate redundant features, which not only reduced computational complexity but also improved model accuracy.
This study evaluated the use of a combination of image processing and artificial neural networks (ANNs) for the identification and classification of authentic and counterfeit Bunium persicum. The results demonstrated that texture features provided the highest accuracy in distinguishing the two types of cumin, with overall classification accuracies of 93.51%, 95.86%, and 95.59% for color, texture, and combined features, respectively. Additionally, after feature selection using factor analysis, the best average classification accuracy with color, texture, and combined features was 93.81%, 95.65%, and 94.67%, respectively. The classification accuracy increased with the reduction of color features, while a decrease in texture and combined features led to a reduction in classification accuracy. These findings highlight the significant potential of image processing and ANNs in identifying and classifying genuine cumin from counterfeit ones. The application of these methods in the design of intelligent systems for identifying authentic and counterfeit cumin in the food industry could be a valuable step towards improving food quality and safety.
Conceptualization, M.D., I.G. A.N.; methodology, M.D. and A.N.; software, I.G. and A.M.; validation, A.N., M.D., I.G. and A.M.; formal analysis, I.G.; investigation, A.N. and M.D.; resources, M.D. and A.M.; data curation, M.D. and A.M.; writing-original draft preparation, A.M. and M.D; writing-review and editing, A.M. and M.D.; visualization, A.M. and I.G..; supervision, M.D.; project administration, M.D.
All authors have read and agreed to the published version of the manuscript.” All authors contributed equally to the conceptualization of the article and writing of the original and subsequent drafts.
All data generated or analyzed during this study are available from the corresponding author on request
The authors would like to extend their sincere appreciations for financial support provided by the University of Jiroft. The authors avoided data fabrication, falsification, plagiarism, and misconduct.
The author declares no conflict of interest.
)