A computational method for classifying a natural cork stopper

Abstract

The present disclosure relates to the non-invasive and non-destructive testing of natural cork stoppers, for the classification of natural cork stoppers suitable to be provided in bottles such as wine bottles. It includes a computational method (100) for classifying a natural cork stopper comprising obtaining electromagnetic information of the natural cork stopper, the electromagnetic information being obtained from a non-destructive testing imaging technique performed on the natural cork stopper (101), obtaining one or more features of the natural cork stopper from the electromagnetic information (102), accessing a database correlating electromagnetic information features of natural cork stoppers with corresponding oxygen transmission rates (103), and comparing the obtained one or more features with the electromagnetic information features of the database (104) and therefrom classifying the natural cork stopper (105).

Description

A COMPUTATIONAL METHOD FOR CLASSIFYING A NATURAL CORK STOPPER TECHNICAL FIELD The present disclosure relates to the non-invasive and non-destructive testing of natural cork stoppers, for the classification of natural cork stoppers suitable to be provided in bottles such as wine bottles. PRIOR ART Oxygen management is one of the most challenging tasks in wine making. Starting from the initial juice to the maturation process, several critical steps, relating to the oxygen exposure, can be found, where the quantities of oxygen supplied will have a major impact on the organoleptic characteristics of the finished product. Premature oxidation during bottle aging, for white wines significantly impacts the sensory quality of the product. Research identifying oxidation in whites as a major organoleptic fault commenced in the 1990s, initially focusing on changes in wine color. Since the year 2000, research evolved into two distinct approaches: one aimed at understanding the key factors influencing wine's sensory perception, and the other adopting a molecular perspective to elucidate the mechanisms behind wine oxidation. The primary factors driving oxidation in bottled wine, can be categorized into several classes: (i) wine matrix, antioxidant fraction, transition metals ration for each specie, iron and cupper or SO2; (ii) the wine bottling process; (iii) the bottleneck variability; (iv) surface treatments present at the cork-glass interface; and (v) oxygen transmission rate (OTR) through the natural cork. Those factors are highly amplified by the time, Temperature, pH or bottle position during storage. The Oxygen Transmission Rate (OTR) refers to the measure of permeability of a natural cork stopper to oxygen. It quantifies the amount of oxygen that passes through the cork material over a specific period under certain conditions of temperature and humidity. The OTR is critical in determining the cork stopper's efficacy in preserving the quality and longevity of bottled contents, particularly in the case of wine. The natural variability in cork's cellular structure can lead to variations in OTR, which in turn can influence the aging process of wine or other bottled products. The measurement of the OTR in natural cork stoppers is important for ensuring consistent product performance, including applications where controlled oxygen exposure is relevant for the maturation or preservation of the bottled contents. Industrial solutions known in the art do not allow to individually measure the OTR of high numbers natural cork stoppers, resulting that tested stoppers are no longer suitable for usage after testing. The present disclosure provides a non-invasive and non-destructive solution for the testing of natural cork stoppers which enables the determination of the OTR of an individual cork stopper and the consequent classification of the cork stopper. This determination is performed in a swift and thereby industrially viable manner. SUMMARY OF THE DISCLOSURE The present disclosure includes a computational method, specifically a computational method for classifying a natural cork stopper. The method comprises the following steps: obtaining electromagnetic information of the natural cork stopper, the electromagnetic information being obtained from a non-destructive testing imaging technique performed on the natural cork stopper, obtaining one or more features of the natural cork stopper from the electromagnetic information, accessing a database correlating electromagnetic information features of natural cork stoppers with corresponding oxygen transmission rates, comparing the obtained one or more features with the electromagnetic information features of the database and therefrom classifying the natural cork stopper. The electromagnetic information may comprise one or more images obtained from a non-destructive testing imaging technique performed on the natural cork stopper, and the one or more features of the natural cork stopper obtained from the electromagnetic information are obtained from the one or more images. Comparing the obtained one or more features with the electromagnetic information features of the database may comprise automatically determining an oxygen transmission rate associated with the obtained one or more features and therefrom classifying the natural cork stopper. Comparing the obtained one or more features with the electromagnetic information features of the database may further comprise comparing the associated oxygen transmission rate with a predefined threshold and therefrom classifying the natural cork stopper. Classifying the natural cork stopper may comprise classifying the natural cork stopper as compliant or as non-compliant. The non-destructive testing imaging technique comprises x-ray imaging. Obtaining one or more electromagnetic images of the natural cork stopper may comprise obtaining a plurality of images, the one or more obtained features being provided in several of the images, the images being optionally more than 5, optionally more than 10, optionally more than 15, optionally more than 20, optionally more than 30, optionally more than 40, optionally 24 or 50. The several images may correspond to different views of the natural cork stopper, particularly different rotated views, the corresponding axis rotation being longitudinal or axial in relation to the natural cork stopper. Obtaining the one or more features may comprise, for each image: applying a threshold for binarization, segmenting along a dimension, such as the longitudinal dimension, performing the sum of all pixels in the same dimension, determining the maximum values of the pixels, calculating a metric of the maximum values, the metric including the mean or the median, and may further comprise: determining the maximum value from the obtained metrics. The computational method for classifying a natural cork stopper may be implemented by an artificial intelligence model trained with data of: electromagnetic information of a plurality of natural cork stoppers, the electromagnetic information being obtained from a non-destructive testing imaging technique performed on the natural cork stopper, one or more features obtained from the electromagnetic information of each natural cork stopper, the oxygen transmission rate of each natural cork stopper, a classification of the natural cork stopper. The training of the artificial intelligence model may specifically be such that the electromagnetic information of a plurality of natural cork stoppers with which the artificial intelligence model is trained comprises one or more electromagnetic images of a plurality of natural cork stoppers, the one or more electromagnetic images being obtained by a non-destructive testing imaging technique, the one or more features with which the artificial intelligence model is trained comprise one or more features obtained from the one or more images of each natural cork stopper. The artificial intelligence model may be supervised or unsupervised. The artificial intelligence model may be a deep learning model, optionally comprising convolutional neural networks. The present disclosure may further comprise a computational method for training an artificial intelligence model to automatically classify natural cork stoppers. The computational method for training an artificial intelligence model to automatically classify natural cork stoppers may comprise training the artificial intelligence model with data of: electromagnetic information of a plurality of natural cork stoppers, the electromagnetic information being obtained from a non-destructive testing imaging technique performed on the natural cork stopper, one or more features obtained from the electromagnetic information of each natural cork stopper, the oxygen transmission rate of each natural cork stopper, a classification of the natural cork stopper. The computational method for training an artificial intelligence model to automatically classify natural cork stoppers may be such that the electromagnetic information of a plurality of natural cork stoppers with which the artificial intelligence model is trained comprises one or more electromagnetic images of a plurality of natural cork stoppers, the one or more electromagnetic images being obtained by a non-destructive testing imaging technique, the one or more features with which the artificial intelligence model is trained comprise one or more features obtained from the one or more images of each natural cork stopper. The artificial intelligence model may be supervised or unsupervised. The artificial intelligence model may be a deep learning model, optionally comprising convolutional neural networks. The present disclosure further comprises a non-destructive and non-invasive method for classifying a natural cork stopper. The non-destructive and non-invasive method for classifying a natural cork stopper may comprise: applying a non-destructive testing technique to obtained electromagnetic information from a natural cork stopper, implementing the computational method for classifying a natural cork stopper of the present disclosure, classifying the natural cork stopper and automatically sorting the natural cork stopper based on the attributed classification. The present disclosure further comprises a system for the non-destructive and non-invasive classification of a natural cork stopper. The system for the non-destructive and non-invasive classification of a natural cork stopper may comprise: imaging means for applying a non-destructive testing imaging technique to a natural cork stopper, computational means configured to implement the computational method for classifying a natural cork stopper of the present disclosure, automatic sorting means configured to sort the according to the attributed classification. The present disclosure may further comprise a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the computational method for classifying a natural cork stopper of the present disclosure. The present disclosure may further comprise a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the computational method for training an artificial intelligence model to automatically classify natural cork stoppers of the present disclosure. DESCRIPTION OF THE DRAWINGS Figure 1 - a representation of the computational method for classifying a natural cork stopper of the present disclosure. Figure 2a - a representation of a first part of a non-artificial intelligence-based image processing data flow according to the present disclosure, from an image to which binarization is applied to obtaining maximum values of mean. Figure 2b - a representation of a second part of a non-artificial intelligence-based image processing data flow according to the present disclosure, following the steps of Figure 2a and from obtaining mean values to obtaining maximum values of the determined mean values. Figure 3 - a representation of images obtained from natural cork stoppers based on imaging means and processed according to the present disclosure. Figure 4 - representation of oxygen transmission rates of different natural cork stoppers along time, for treated natural cork stoppers and in which five outliers are visible (88A, 2041A, 1119A, 2024A and 1698A). Figure 5 - images of natural cork stoppers obtained according to the methods of the present disclosure corresponding to the oxygen transmission rates of Figure 4, whereby images contain the features generating the outlying OTR. Figure 6 - representation of oxygen transmission rates of different natural cork stoppers along time, for untreated natural cork stoppers and in which four outliers are visible (3F, 2D, land 2). Figure 7 - images of natural cork stoppers obtained according to the methods of the present disclosure corresponding to the oxygen transmission rates of Figure 6, whereby images contain the features generating the outlying OTR. DETAILED DESCRIPTION As described in the Summary of the invention, the present disclosure comprises a computational method for classifying a natural cork stopper and a non-destructive and non-invasive method for classifying a natural cork stopper. The solutions of the present disclosure provide a significant advancement in the field of cork quality assessment, offering a rapid, accurate, and non-destructive means of classifying the oxygen transmission rate of natural corks, thereby enhancing the quality control process in industries reliant on cork sealing. A natural cork stopper is typically an at least partly cylindrically shaped closure device specifically designed for sealing containers, such as glass bottles for containing beverages such as wine. The capacity or volume of wine bottles is typically in the range of 37.5 cl to 3 I, being the most common volume of 75 cl. Natural cork stoppers may also be used in higher capacity containers, having volumes of more than 3 I, such as 5 I or 15 I containers. A natural cork stopper involves the usage of cork oak tree bark to produce the stopper. The cellular structure of cork, characterized predominantly by closed cells filled with an air-like gas and primarily composed of suberin, lignin, and polysaccharides, imparts mechanical and chemical properties to the stopper. The process of manufacturing a natural cork stopper typically involves harvesting the cork bark, boiling to increase elasticity and sterilization, followed by cutting and shaping into standardized dimensions suitable for insertion into the necks of bottles. The stopper's dimensions may be calibrated to achieve a balance between sufficient compression for insertion into the bottle neck and adequate expansion to form an airtight and liquid-tight seal, thereby preserving the contents of the bottle. Further, porosity of a natural cork stopper allows for controlled oxygen transmission rate (OTR), an important factor in the aging of wine and other bottled beverages. The solution of the present disclosure provides a non-invasive and non-destructive solution for the determination of the OTR of an individual natural cork stopper, and the consequent classification of the natural cork stopper. This determination is performed in a swift and thereby industrially viable manner. The computational method for classifying a natural cork stopper of the present disclosure comprises obtaining electromagnetic information of the natural cork stopper, the electromagnetic information being obtained from a non-destructive testing imaging technique (NDT technique) performed on the natural cork stopper. The electromagnetic information may consist of one or more signals based on an NDT technique, enabling the computational processing of such signals. Furthermore, the electromagnetic information may comprise one or more images obtained from a non-destructive testing imaging technique performed on the natural cork stopper. An example of an NDT technique includes x-ray imaging. The NDT technique provides imaging information on the inside of the natural cork stopper under testing / classification. For electromagnetic information comprising one or more images, the one or more features of the natural cork stopper obtained from the electromagnetic information are obtained from the one or more images. Still according to the computational method for classifying a natural cork stopper of the present disclosure, a database correlating electromagnetic information features of natural cork stoppers with corresponding oxygen transmission rates is accessed. Such database correlates electromagnetic information features and oxygen transmission rates, which provide a basis for comparison with the features obtained from the natural cork stopper under testing. The database may be locally or remotely provided. Finally, the obtained one or more features are compared with the electromagnetic information features obtained from the database. Based on such comparison, the natural cork stopper is classified. Such method does not violate the integrity of the natural cork stopper under testing, and still provides a reliable analysis of the stopper. Moreover, such testing is sufficiently quick to enable industrialization and provide total quality control. Comparing the obtained one or more features with the electromagnetic information features of the database may comprise automatically determining an oxygen transmission rate associated with the obtained one or more features and therefrom classifying the natural cork stopper. Comparing the obtained one or more features with the electromagnetic information features of the database may further comprise comparing the associated oxygen transmission rate with a predefined threshold and therefrom classifying the natural cork stopper. Classifying the natural cork stopper may comprise classifying the natural cork stopper as compliant or as non-compliant. The natural cork stopper may also be classified according to other virtual classifications, which may be designated as baskets. Such baskets may represent different quality levels, a natural cork stopper being classified with one of such baskets or quality levels. Obtaining one or more electromagnetic images of the natural cork stopper may comprise obtaining a plurality of images. The obtained features are present in several of the images. The images may be two-dimensional images. The images acquired may be stacked to make a sequence of images. In such case, the images correspond to slices of the tested natural cork stopper, wherein the features are spread or identifiable in the several images. Moreover, the several obtained images may correspond to different views of the natural cork stopper, particularly different rotated views, the corresponding axis rotation being longitudinal or axial in relation to the natural cork stopper. The method may involve that the obtained images result from the rotating the natural cork stopper and subsequently obtaining of the images. The method may alternatively involve that the obtained images result from the rotating of means for obtaining such images around the natural cork stopper, and the subsequent obtaining of images. In an embodiment, images are analyzed using an artificial intelligence model. More specifically, the artificial intelligence model may involve the usage of supervised or unsupervised machine learning, with the usage of segmented or non-segmented techniques. The implemented algorithms may include deep learning algorithms, more specifically convolutional neural networks. In another, alternative, embodiment, images - or the stack of images (or frames) - may be received and the following data flow is applied: - applying a threshold for binarization to each image, - binarizing each image / frame on the stack, - segmenting each image / frame in a longitudinal dimension of the natural cork stopper, providing a number of segments, such as 5 segments, - performing the sum of all pixels in the longitudinal dimension, - In each of the segments, identifying the maximum, - calculating the average of the five maximum values, - repeat for the total images or frames of the stack, which preferably consist of 20 or 24, and determining the mean for all frames, - determining the maximum of the determined mean values, - classifying samples above a predefined threshold as non-compliant (NOT OK) and samples below the predefined threshold as compliant (OK). Figures 2a and 2b contain a schematic representation of such data flow, based on a single image or frame. The original image of Figure 2a (on the left, top), the representation obtained upon binarization show a distinctive feature provided in the middle. The identifiable feature will result in, upon the application of the above-described set of steps, a maximum media value to which will correspond a classification of NOT OK. Figure 3 shows, firstly (graphic on the top, right side), the OTR is independent of optical density, thereby providing adaptability to the method of the present disclosure. In addition, secondly (graphic on the bottom, right side), the false positive rate of the method of the present disclosure is considerably low. Figure 4 contains a graphic representing different oxygen transmission rates of different treated natural cork stoppers, as calculated along several days. As can be seen, there are five stoppers with outlying values, particularly identifiable after day 10. Direct testing of the oxygen transmission rates along time would be absolutely unviable for an industrial application. Figure 5 shows x-ray imaging frames obtained from the same stoppers with outlying values. Distinctive features are clearly identifiable in these stoppers. The identification of these features through an NDT technique allows to individually test each stopper and classify it, by the indirect determination of its OTR. This procedure is industrially viable and enables total quality control. The same rationale is applicable for figures 6 and 7. Figure 6 contains a graphic representing different oxygen transmission rates of different untreated natural cork stoppers, as calculated along several days. As can be seen, the natural cork stoppers with outlying values are identifiable much quicker than for untreated cork stoppers. There are four stoppers with outlying values, particularly identifiable after day 2 / 3. Again, distinctive features are visually perceptible in the corresponding frames of Figure 7. The method of the present disclosure allows to automatically and reliably identify the features resulting which would result in outlying OTR, if tested during a long period of time, as provided in Figures 5 and 7. Regarding another aspect of the present disclosure, the computational method for classifying a natural cork stopper and the training method of the present disclosure may involve training an artificial intelligence model based on electromagnetic information and / or images correlated with analytical OTR measurements from reference methodologies. Further embodiments are described subsequently. The algorithms are trained on a dataset of OTR calibrated corks and the image projections obtained are annotated. A first step of such methodology involves feature definition and extraction, correlated with OTR values. Segmented and non-segmented images techniques are used for training. Deep learning algorithms are trained to recognize the feature. Deep learning algorithms may be used on stacks of at least 20 images / frames and the sample is classified as NOT OK if at least 5 images present the feature and is rejected. Otherwise, the natural cork stopper sample is classified as OK and is not rejected. These annotations are based on analytical measurements of OTR using established reference methodologies. The system identifies specific features correlated with OTR levels. Real-Time Classification is applied: upon image analysis, the system classifies the OTR of each cork in real-time. The classification is non-destructive and non-invasive, preserving the integrity of the cork. The system and methods can be integrated into cork production lines for quality control. The system and methods provide a reliable and efficient method for selecting corks with desired OTR levels for various applications. The non-destructive and non-invasive method of the present disclosure involves the usage of electromagnetic radiation such as X-ray radiation to capture high-resolution images of natural corks. The X-ray radiation threshold is adjusted and improve the image quality and visualization of defects, assuring the correct performance of the process. The light source may be set specifically in the THz range, preferentially X-ray source can be used. The present disclosure may further comprise creating a database a database correlating electromagnetic spectral information features of natural cork stoppers with corresponding oxygen transmission rates. Creating such database includes the steps of: obtaining electromagnetic information, for instance one or more electromagnetic images, of a natural cork stopper, the electromagnetic information being obtained by a non-destructive testing imaging technique, obtaining one or more features of the natural cork stopper from the electromagnetic information, measuring the oxygen transmission rate of the same natural cork stopper, classifying the measured oxygen transmission rate as compliant or non-compliant, relating the obtained one or more features, the oxygen transmission rate and the corresponding classification in the database. Although the present disclosure is mainly described in terms of methods and systems, the skilled person understands that it is also directed to various devices or apparatuses, which may include one or more computational devices, installed locally or remotely. For instance, the method of the present disclosure may be implemented by the system of the present disclosure, including an electronic or computational device installed locally or remotely. The system of the present disclosure may comprise an electronic or computational device, having computational means installed locally or remotely. The system may include components to perform at least some of the example features and features of the methods described, whether through hardware components (such as memory and / or processor), software or any combination thereof. An article for use with the system, such as a pre-recorded storage device or other similar computer-readable medium, including program instructions recorded on it, or a computer data signal carrying readable program instructions computer can direct a device to facilitate the implementation of the methods described herein. It is understood that such apparatus, articles of manufacture and computer data signals are also within the scope of the present disclosure. A "computer-readable medium" means any medium that can store instructions for use or execution by a computer or other computing device, including read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, random access memory (RAM), a portable floppy disk, a drive hard drive (HDD), a solid state storage device (for example, NAND flash or synchronous dynamic RAM (SDRAM)), and / or an optical disc such as a Compact Disc (CD), Digital Versatile Disc (DVD) or Blu-Ray ™ Disc. As will be clear to one skilled in the art, the present disclosure should not be limited to the embodiments presented herein, and a number of changes are possible which remain within the scope of the present disclosure. Of course, the disclosed embodiments are combinable, in the different possible forms, being herein avoided the repetition all such combinations.

Claims

1. A computational method for classifying a natural cork stopper comprising the steps of:obtaining electromagnetic information of the natural cork stopper, the electromagnetic information being obtained from a non-destructive testing imaging technique performed on the natural cork stopper,obtaining one or more features of the natural cork stopper from the electromagnetic information,accessing a database correlating electromagnetic information features of natural cork stoppers with corresponding oxygen transmission rates,comparing the obtained one or more features with the electromagnetic information features of the database and therefrom classifying the natural cork stopper.

2. A computational method according to the previous claim wherein the electromagnetic information comprises one or more images obtained from a non-destructive testing imaging technique performed on the natural cork stopper, andthe one or more features of the natural cork stopper obtained from the electromagnetic information being obtained from the one or more images.

3. A computational method according to any of the preceding claims wherein comparing the obtained one or more features with the electromagnetic information features of the database comprises automatically determining an oxygen transmission rate associated with the obtained one or more features and therefrom classifying the natural cork stopper.

4. A computational method according to the previous claim wherein comparing the obtained one or more features with the electromagnetic information features of the database further comprises comparing the associated oxygen transmission rate with a predefined threshold and therefrom classifying the natural cork stopper.

5. A computational method according to any of the preceding claims wherein classifying the natural cork stopper comprises classifying the natural cork stopper as compliant or as non-compliant.

6. A computational method according to any of the preceding claims wherein the nondestructive testing imaging technique comprises x-ray imaging.

7. A computational method according to any of the claims 2-6 wherein obtaining one or more electromagnetic images of the natural cork stopper comprises obtaining a plurality of images, the one or more obtained features being provided in several of the images, the images being optionally more than 5, optionally more than 10, optionally more than 15, optionally more than 20, optionally more than 30, optionally more than 40, optionally 24 or 50.

8. A method according to the previous claim wherein the several images correspond to different views of the natural cork stopper, particularly different rotated views, the corresponding axis rotation being longitudinal or axial in relation to the natural cork stopper.

9. A method according to any of the claims 2-8 wherein obtaining the one or more features comprises, for each image:applying a threshold for binarization,segmenting along a dimension, such as the longitudinal dimension, performing the sum of all pixels in the same dimension, determining the maximum values of the pixels,calculating a metric of the maximum values, the metric including the mean or the median,and further comprises:determining the maximum value from the obtained metrics.

10. A method according to any of the claims 1-9 wherein it is implemented by an artificial intelligence model trained with data of:electromagnetic information of a plurality of natural cork stoppers, the electromagnetic information being obtained from a non-destructive testing imaging technique performed on the natural cork stopper,one or more features obtained from the electromagnetic information of each natural cork stopper,the oxygen transmission rate of each natural cork stopper, a classification of the natural cork stopper.

11. A method according to the previous claim wherein:the electromagnetic information of a plurality of natural cork stoppers with which the artificial intelligence model is trained comprises one or more electromagnetic images of a plurality of natural cork stoppers, the one or more electromagnetic images being obtained by a non-destructive testing imaging technique,the one or more features with which the artificial intelligence model is trained comprise one or more features obtained from the one or more images of each natural cork stopper.

12. A method according to any of the claims 10-11 wherein the artificial intelligence model is supervised or unsupervised.

13. A method according to any of the claims 10-12 wherein the artificial intelligence model is a deep learning model, optionally comprising convolutional neural networks.

14. A computational method for training an artificial intelligence model to automatically classify natural cork stoppers, the method comprising training the artificial intelligence model with data of:electromagnetic information of a plurality of natural cork stoppers, the electromagnetic information being obtained from a non-destructive testing imaging technique performed on the natural cork stopper,one or more features obtained from the electromagnetic information of each natural cork stopper,the oxygen transmission rate of each natural cork stopper,a classification of the natural cork stopper.

15. A method according to the previous claim wherein:the electromagnetic information of a plurality of natural cork stoppers with which the artificial intelligence model is trained comprises one or more electromagnetic images of a plurality of natural cork stoppers, the one or more electromagnetic images being obtained by a non-destructive testing imaging technique,the one or more features with which the artificial intelligence model is trained comprise one or more features obtained from the one or more images of each natural cork stopper.

16. A method according to any of the claims 14-15 wherein the artificial intelligence model is supervised or unsupervised.

17. A method according to any of the claims 14-16 wherein the artificial intelligence model is a deep learning model, optionally comprising convolutional neural networks.

18. A non-destructive and non-invasive method for classifying a natural cork stopper comprising:applying a non-destructive testing technique to obtained electromagnetic information from a natural cork stopper,implementing the method of any of the claims 1-13,classifying the natural cork stopper and automatically sorting the natural cork stopper based on the attributed classification.

19. A system for the non-destructive and non-invasive classification of a natural cork stopper comprising:imaging means for applying a non-destructive testing imaging technique to a natural cork stopper,computational means configured to implement the method of any of the claims 1-13,automatic sorting means configured to sort the according to the attributed classification.

20. A computer program product comprising instructions which, when the program is 5 executed by a computer, cause the computer to carry out the method of any of the claims 1-13.

21. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any of the 10 claims 14-17.

Images