J. Dairy Sci. 2008. 91:918-927. doi:10.3168/jds.2007-0661
© 2008 American Dairy Science Association ®
Detection of Internal Cracks in Manchego Cheese Using the Acoustic Impulse-Response Technique and Ultrasounds
T. Conde,
A. Mulet,
G. Clemente and
J. Benedito1
Department of Food Technology, Polytechnic University of Valencia, Camino de Vera, 46022 Valencia, Spain
1 Corresponding author: jjbenedi{at}tal.upv.es
 |
ABSTRACT
|
|---|
Nowadays, due to the more global nature of markets, the commercialization of cheese relies on the high quality of the product. Internal defects such as cracks or flaws may affect quality. Two different nondestructive inspection techniques (ultrasonic and acoustic experiments) were used to detect cracks in Manchego cheese. The existence of small eyes in this type of cheese limited the use of ultrasonic pulse-echo experiments due to high scattering, and only cracks close to the surface of the cheese could be detected. The acoustic impulse-response technique, however, allowed us to study wheel pieces with cracks located elsewhere in the cheese. Two different impact probes (A and B) were assayed. The energy content of the acoustic spectrum was higher for cracked wheel pieces (7,116 and 17,520 V Hz1/2 for probes A and B, respectively) than for normal ones (6,841 and 16,821 V Hz1/2). The differences were mainly found for frequencies higher than 150 Hz, which made the centroid for cracked pieces higher (162 and 170 Hz for probes A and B, respectively) than that for normal cheeses (132 and 148 Hz for probes A and B, respectively). Discriminant functions were developed to classify wheel pieces, and the input variables used were the acoustic parameters from the spectrum and the principal components extracted from the whole spectrum. The best classification procedure used the principal components from the principal components analysis of the spectrum for probe B. In this case, the 50 wheel pieces used in this study were correctly classified. These results showed that a simple and low-cost acoustic impulse-response technique could be used to detect cheese cracks, formed at different moments of Manchego cheese maturation.
Key Words: Manchego cheese • acoustic • ultrasonic • crack
|
INTRODUCTION
|
|---|
Quality could be defined as the ability of a set of inherent characteristics of a product, system, or process to fulfill the requirements of customers. An important requirement of any quality control system in the food industry is the definition of standards with sensory-based tolerance limits for the product. These standards and tolerance limits should be based on product expectations of consumers, and therefore, it is important for the industry to quantify and understand them. During the last decade, the great concern for food safety and quality has resulted in a great deal of research into the exploration of different nondestructive techniques for detecting internal defects (cracks, flaws, and foreign bodies) in food products.
In the food industry, expert assessors of considerable experience or automated visual inspection systems are often used to detect surface defects that appear in local inspections. In many cases, the defects are located internally and can neither be detected by trained human operators nor by automated visual systems. As a consequence, the use of nondestructive techniques such as X-rays, nuclear magnetic resonance, or acoustic systems should be considered.
X-ray images of bags of frozen sweetcorn kernels were used to detect a variety of contaminants such as metal, glass, stone, rubber, or plastic that were embedded in the product (Patel et al., 1996). The method used was based on convolution filtering, in which the convolution masks act as matched filters for certain types of textural variation found in the images. Furthermore, current 3-dimensional image registration techniques used x-ray microtomography to measure, visualize, and analyze cracks and air eyes in food materials like aerated chocolate, mousse, marshmallow, and muffin (Lim and Barigou, 2004).
One important application of nuclear magnetic resonance is the detection of internal defects such as bruising, watercore, and internal browning in fruits. Keener et al. (1999) used this technique for the prediction of defects in apple tissue, using a low-frequency proton magnetic resonance sensor. High-resolution images of these internal characteristics can contribute to the evaluation of maturity and quality parameters, but they can also help the understanding of physiological processes (Butz et al., 2005).
One alternative way of detecting cracks, flaws, or foreign bodies in different materials is by using the acoustic methods. A quality-control inspection technique using the acoustic impulse-response technique was developed for the detection of surface cracks in eggshells (Cho et al., 2000). An experimental system was built to generate the impact force, measure the response signal, analyze the frequency spectrum, and classify the eggs as cracked or normal. Different variables, such as the average area of the power spectrum, the centroid, and the average peak resonance frequency, were found to be significant for classifying eggs. Another piece of research into crack detection in eggshells used an experimental modal analysis to explore the optimal configuration of the egg support and the location and design of the impactor and the response sensor (De Ketelaere et al., 2000). Diezma-Iglesias et al. (2004) showed that a nondestructive acoustic impulse-response device could be used to classify watermelons into quality classes according to the presence of creases and hollows in the flesh. In this research, waveband magnitude parameters were calculated by adding up the normalized spectrum magnitude between 2 different frequencies. Diezma-Iglesias et al. (2005) used experimental modal analysis to investigate the vibrational performance of watermelons to determine the best positions for the impact point and response measurement microphone. An acoustic sorting system has also been developed to separate pistachio nuts with closed shells from those with open ones. In that study, the discriminant parameters were obtained from the time domain signal instead of analyzing the frequency spectrum (Pearson, 2001).
The most popular milk cheese of ewes produced (7,000 tons/yr) and consumed in Spain is Manchego, which matures unpacked in temperature and relative humidity-controlled chambers (Freitas and Malcata., 2000). It is a cured, semihard, uncooked, pressed, high-fat, and highly porous cheese with small-size eyes formed during the fermentation process. Characterization of dairy products that undergo a ripening or maturation process like Manchego cheese is particularly difficult due to the physicochemical and microbiological phenomena that take place within the food. During the maturation process, abnormal fermentations can take place, ending up in a swelling of the pieces. If the cheese stays in the chamber, the swelling decreases and turns into a crack that is impossible to detect either by pressing or visually. In many cheese varieties, the presence of regular round eyes of different sizes or a uniform distribution of irregularly shaped openings is a desirable and typical feature. The existence of slits or cracks in a cheese is not always considered as a defect. In some types of Gruyère cheese, slits may be present together with small round eyes without affecting the visual quality of the cheese. However, in Manchego cheese, the presence of cracks or slits is not acceptable and can be considered as a defect. The acceptability of these cheese cracks has been influenced not only by consumer expectation but also by the needs of commercial operations which cut and prepack cheese into consumer portions. Cracks and voids in the parent cheese block can result in broken portions and lead to unacceptable losses.
At present, the only available defect detection method is cheese inspection by human experts: the pieces are tapped by hand or using a small hammer, and the piece is classified according to the emitted sound. However, this method is subjective, time-consuming, and expensive. Furthermore, it is not feasible to test each sample of a batch, and therefore, cheeses with internal cracks are always found in the market. Thus, the application of nondestructive techniques to detect cheese cracks is of great interest for the cheese-manufacturing industry. In this regard, the application of x-rays in Grana cheese was used to detect defects produced by fermentative bacteria only a few days after their production (Zapparoli, 1997).
The application of magnetic resonance imaging in the detection and evaluation of eye features in cheese was studied by Rosenberg et al. (1991, 1992). Results indicated that magnetic resonance imaging is a nondestructive method that can provide high-resolution images of the inner structure of cheeses, particulary the presence of air and whey pockets. It was used to study and evaluate eye structure, size, and distribution in Swiss-type cheeses as well as to detect structural defects associated with Swiss-type and Cheddar cheeses.
Ultrasonic techniques have been used to detect internal cracks in Mahon cheese, in through-transmission and pulse-echo modes (Benedito et al., 2001). Using through-transmission mode, it was not possible to distinguish cheeses with small eyes from those with cracks, because the presence of many small holes scattered the ultrasonic wave, producing similar energy absorption to that of cheese with cracks. Nevertheless, using the pulse-echo technique, it was possible not only to detect the cracked pieces but also to assess the magnitude and position of cracks (Benedito et al., 2001).
Giangiacomo et al. (1989) compared the use of the impact impulse-response technique with the classical human percussion, using only the resonance frequency parameter for the detection of cracks in Grana cheese. The results showed a close correlation between the human evaluation and the classification carried out through instrumental analysis. However, if these authors had studied a higher number of parameters of the frequency spectrum, their results could have provided a better description of the cheese characteristics and therefore improved classification. The aim of this work was to study the feasibility of detecting cracks within Manchego cheese pieces using ultrasound and acoustic impulse-response tests.
|
MATERIALS AND METHODS
|
|---|
Cheese Samples: Selection and Expert Evaluation
The 50 wheel pieces of the Certified Origin Manchego cheese used in this study were manufactured by 2 companies located in Albacete, Spain. A trained expert in cheese crack detection selected the samples to be examined to have a similar number of defective and normal cheeses. Once the acoustic experiments were carried out, the cheese wheels were opened. It was observed that 28 wheel pieces were defective, whereas 22 had no internal cracks. A discriminant analysis was carried out to classify the pieces of cheese into 2 groups (normal or defective). To develop the discriminant models, 40 wheel pieces were used. The remaining 10 pieces were used to validate the models. When considering normal and defective batches, it was observed that both groups of pieces ranged in maturity from 60 to 650 d. No significant difference as to maturing times was found between both groups. The diameter of the cylindrical pieces was 19 ± 2 cm, the height 10 ± 4 cm, and the weight 3 ± 0.1 kg. The cheeses were transported, refrigerated, to the laboratory where the pieces remained for 24 h in a temperature-controlled chamber at 12°C to reach a uniform temperature before proceeding with the instrumental analysis.
Ultrasonic Analysis
The experimental configuration for carrying out ultrasonic pulse-echo experiments consisted of an ultrasonic transducer to convert electrical pulses into ultrasonic pulses (1 MHz, 0.75' crystal diameter, A314S-SU model, Panametrics, Waltham, MA) and receive the energy reflected in the cracks and inhomogeneities of the cheese, a pulser-receiver (model 5058PR, Panametrics), and a digital storage oscilloscope to display, digitize, and record the received echoes (model TDS 5034, Tektronix Inc., Beaverton, OR). Twenty measurements (10 on each flat surface of the wheel piece) were carried out following 2 crossed straight lines marked on each cheese surface. The surface of Manchego cheese is corrugated due to the use of special molds into which the cheese curd is pressed. To improve the acoustic matching, olive oil was used as a couplant. To identify where the echoes observed on the oscilloscope were coming from and to assess the internal defects (cracks and flaws) in Manchego cheeses, the wheel pieces were opened along the lines where the ultrasonic measurements were carried out.
Acoustic Impulse-Response Techniques
The system specially built for nondestructive cheese measurements (Figure 1
) mainly consisted of a microphone (
', 40AF, G.R.A.S. Sound and Vibration, Holte, Denmark) with preamplifier ( ', 26AF, G.R.A.S. Sound and Vibration) and an impact probe. A signal conditioner (OPUS, 01 db-stell, MVI Technologies Group, Paris, France) was used to feed the microphone as well as to amplify and filter the signal. The signal was then digitized in a PCMCIA card (NI5102, National Instruments, Austin, TX).
Two different types of experiments were carried out. In the first type, an impact probe (probe A), which was previously leveled on the surface of the piece, was released from its rest position (I, Figure 1
) until it struck the cheese surface (II, Figure 1). The signal conditioner worked for this probe at 0-dB gain and 0.3-Hz high pass filter. On the other hand, a second type of experiment was carried out by using a hand-operated probe (probe B), which also struck the cheese surface. The signal conditioner worked for probe B at 20-dB gain and 10-Hz high pass filter. The sampling frequency was set to 20 kHz for both probes. This frequency was chosen from preliminary experiments to avoid aliasing according to the Nyquist-Shannon sampling theorem. Five impacts were carried out with each probe at different points of each flat side of the cheese wheels following a template.
The aim of both types of experiments was to collect the sound emitted by the wheel pieces due to their own vibration after being struck by the probe. The position of the microphone was determined in preliminary experiments to maximize the differences in the spectrum between normal and defective wheel pieces. The sensor for both probes was placed 2 cm above the flat cheese surface and 5 cm from the center of the same flat round surface. The fast-Fourier transform was calculated on a temporal window of the digitized signal (8,192 and 4,096 points for probe A and B, respectively). This allowed a frequency resolution of 2.44 and 4.88 Hz for probes A and B, respectively. Only the frequency band from 10 to 500 Hz for probe A and from 20 to 400 Hz for probe B was used in the classification procedure, because the energy content for the remaining frequencies was negligible. It would be useful to consider intervals in the spectra for comparison purposes (Benedito et al., 2007; Conde et al., 2007). According to preliminary experiments, the spectrum was divided into different intervals to find frequency ranges for cheeses with and without cracks or flaws, containing peaks whose amplitude and frequency distribution could be clearly identified and compared. For probe A, the first interval ranged from 10 to 66 Hz, the second one from 66 to 139 Hz, and the third one from 139 to 500 Hz. For probe B, the spectrum was divided from 20 to 54 (interval I), from 54 to 137 (interval II), and from 137 to 400 Hz (interval III).
For each spectrum and for both probes, the momentum (M1; equation [1], the amount of energy or momentum (M0; equation [2], the central frequency (CF; equation [3], and the variance (Va; equation [4] were calculated for the whole of the studied spectrum. Additionally, the momentum M0 was calculated for each interval (M01, M02, and M03):
 | [1] |
 | [2] |
 | [3] |
 | [4] |
where f = the frequency and X(f) = the spectrum amplitude of the fast-Fourier transform.
Statistical Analysis for Sample Discrimination
A 2-sample comparison analysis was performed on the input parameters (maturation time and acoustic parameters) using Statgraphics Plus 5.1 (Statistical Graphics Corp., Rockville, MD). The software used calculated the average and the standard deviation and, using the t-test, indicated if the parameters considered were significantly different for normal and defective cheeses.
A discriminant analysis was also carried out using the Statgraphics Plus 5.1 statistics software, to study the suitability of the acoustic parameters (Va, M1, M0, M01, M02, M03, and CF) for both probes, to classify the pieces of cheese into the 2 different groups (normal or defective). This software allows us to generate a discriminant function for each group. The forward selection procedure was used to include the variables (acoustic parameters) in the discriminant functions, which account for most of the differences between the 2 groups. From the input variables for each observation, a value for both discriminant functions is obtained. The function that yields the highest value for an observation represents the predicted group.
In addition to the discriminant analysis carried out using the acoustic variables, another discriminant analysis was performed using principal components from a principal component analysis (PCA) for each probe. The PCA was used to analyze the data set and reduce the information of the large number of variables (the amplitude of each frequency in the spectrum) considered by reconstructing them into uncorrelated combinations. The PCA was carried out on the set of spectra from the wheel pieces of cheese by considering the frequency bands of interest (10 to 500 Hz and 20 to 400 Hz for probe A and B, respectively). For each cheese (spectrum), a new set of scores was obtained that accounted for the projection on the common principal components extracted. The Soft Independent Modeling of Class Analogy statistics software (SIMCA-P, Unimetrics, Kinnelon, NJ) was used for this purpose. Later, these components were used in the discriminant analysis using the Statgraphics Plus 5.1.
|
RESULTS AND DISCUSSION
|
|---|
Ultrasonic Analysis
Figure 2 shows a picture of an opened piece of cheese with no cracks (a) and the ultrasonic pulse-echo signal obtained from point T. The ultrasonic echo that reaches the transducer corresponds to the signal reflection on the small eyes located in its pathway around 1 cm from the cheese surface. On the other hand, Figure 3a shows an opened piece of cheese with several cracks inside. In this case, there is also a major echo in the ultrasonic signal (Figure 3b) due to the reflection of the signal on some eyes located in the outermost layers of the cheese. According to the ultrasonic velocity in Manchego cheese at 12°C (1,607 to 1,660 m s–1; Benedito et al., 2006), the echo from the crack (Figure 3a) should be located from 38 to 40 µs. However, no detectable signal is found in that time interval (Figure 3b), probably because all the energy departing from the ultrasonic transducer had been scattered in the eyes of the outermost layers. Similar results were found when analyzing the set of pieces. As shown by Benedito et al. (2006), the existence of small eyes in Manchego cheese could limit the use of ultrasonics, a problem which is not found for other types of cheese such as Cheddar or Mahon cheese (Benedito et al., 2000a,b). The ultrasonic measurements carried out on the whole pieces of Manchego cheeses showed that porosity scatters the ultrasonic waves to an undetectable level when considering large ultrasonic pathways. In fact, the scattering of the ultrasonic signal in porous foodstuffs has been recognized as one of the most important drawbacks for the application of ultrasonic techniques (Povey and McClements, 1988; McClements, 1997). Because the amplitude of the ultrasonic echoes (Figure 2b, 3b) could be linked to the size and number of the air pockets in food (McClements, 1997), ultrasounds could be used to characterize the internal structure of Manchego cheese.
View larger version (52K):
[in this window]
[in a new window]
|
Figure 2. (a) Small eyes in defect-free wheel pieces of Manchego cheese. (b) Typical signal found in ultrasonic pulse-echo testing for a defect-free cheese.
|
|
View larger version (51K):
[in this window]
[in a new window]
|
Figure 3. (a) Cracks in defective wheel pieces of Manchego cheese. (b) Typical signal found in ultrasonic pulse-echo testing for a defective cheese.
|
|
These results show that the internal structure of Manchego cheese (porosity) reflects back the ultrasonic waves, when they reach the eyes situated in the outermost layers. The reflected energy completely disappears at penetration distances ranging from 1 to 2 cm. Therefore, it is not feasible to detect ultrasonically cracks deeper than 1 cm, which is a common case in Manchego cheese. However, for eyeless cheeses like Mahon cheese, it is possible to detect cracked wheel pieces and also to assess the distance from the surface using pulse-echo mode (Benedito et al., 2001). For Manchego cheese, other nondestructive techniques like the acoustic impulse-response one should be tried to detect internal cracks.
Acoustic Response of Cheese Wheels
Figures 4a and 4b show the typical spectra of normal and defective cheeses for probes A and B, respectively. In both of them, it can be observed that there is a major peak (resonance frequency) located in the lower frequency interval (I). Furthermore, it is important to notice that, at first glance, no significant differences in frequency or amplitude are found in intervals I and II between normal and defective wheel pieces. However, the area under the third interval for probes A and B was, at first glance (Figure 4a, 4b), larger for defective cheeses than normal ones.
View larger version (24K):
[in this window]
[in a new window]
|
Figure 4. Typical energy spectra of normal and defective Manchego cheeses for the acoustic impulse-response technique; (a) probes A and (b) B. Spectra were divided into 3 intervals (I, II, and III) to calculate M0 in each interval.
|
|
De Ketelaere et al. (2000) found that the frequency spectrum of normal eggs showed 1 major peak, whereas more than 3 major peaks were observed in the spectrum of cracked eggs. As happens with Manchego cheese, the frequency of the main peak in eggs was not dependent on the existence of cracks. In the present study, no difference in the number of peaks is observed between normal and defective cheeses. Another acoustic sorting system was developed by Pearson (2001), to separate pistachio nuts with closed shells from those with open shells. This study found that the frequency spectrum of sounds emitted from open-shell pistachios exhibited a well-defined peak, whereas closed-shell nuts had a flatter spectrum.
Based on the spectra, acoustic parameters were obtained for the whole spectrum and for each interval. Table 1 shows, for probes A and B, the average and standard deviation for the different acoustic parameters and also indicates if those averages were significantly different.
For probe A, the variance (Va), the momentum of first order (M1), and the amount of energy in the whole spectrum or momentum of zero order (M0) were significantly (P < 0.01) higher for defective cheeses than for normal ones (Table 1). In the case of normal wheel pieces, the energy was absorbed into the whole body of the cheese, and not too much energy reached the microphone situated near the uppermost surface of the cheese (Figure 1). The higher energy content of the spectrum from cracked cheeses could be due to the reflected energy in the internal cracks that reached the microphone.
In the study carried out by Cho et al. (2000) regarding the detection of surface cracks in eggshells, the energy content of the frequency spectrum was also higher for cracked eggs than for normal ones. Another similar research study developed a nondestructive procedure for detecting internal disorders in seedless watermelons (Diezma-Iglesias et al., 2004). The waveband magnitude parameter BM1 (energy content from 85 to 160 kHz) was the best predictor for the internal hollows, BM1 being higher for defective watermelons.
For probe A, no significant differences were found for the momentums in the first (M01) and second intervals (M02) for normal and defective wheel pieces. However, the momentum in the third interval (M03) was significantly higher for defective cheeses (3,058 V Hz1/2) than for normal ones (2,296 V Hz1/2), due to the contribution of the internal cracks to the interval III frequency band. This was the origin of the differences in the energy content for the whole spectrum (M0). Consequently, the higher contribution of frequencies from 139 to 500 Hz for cheeses with cracks makes the energy content of the whole spectra move toward higher frequencies, and therefore, the CF was significantly higher (Table 1) for wheel pieces with cracks (162 Hz) than for normal pieces (132 Hz). Figure 5 shows the dispersion diagram between the centroid for the power spectrum curve (CF) and the momentum in the second interval (M02) for probe A. The centroid for the normal cheeses was clearly smaller than the one for the defective pieces; thus, it is easily possible to discriminate between both types of cheese. Therefore, this parameter could be used as criteria for classification.
View larger version (10K):
[in this window]
[in a new window]
|
Figure 5. Central frequency (CF) vs. momentum in the second interval (M02) for defective and normal Manchego cheeses using the acoustic impulse-response technique; probe A.
|
|
For probe B, the results (Table 1) behaved in the same way as probe A to distinguish between normal and defective wheel pieces through the acoustic parameters. In this case, the acoustic parameters Va, M1, and M0 were also significantly higher for defective pieces. However, no significant differences were found between normal and defective pieces, either for the M01 or for the M02. An important difference in the energy content was also found for the highest frequencies, interval III, from 137 to 400 Hz.
Criteria for Classification
To classify cheese wheel pieces according to the existence of internal cracks, a discriminant analysis was considered. Two different approaches were used when carrying out the discriminant analysis. In the first one, the spectrum descriptors (Va, M1, M0, M01, M02, M03, and CF) were considered as the input variables for the discriminant analysis. Because part of the information contained in the frequency spectrum could be missed by these variables, the principal components extracted from a PCA were also used to carry out the classification procedure.
When the acoustic parameters were used to distinguish between normal and defective pieces, 3 parameters (CF, M01, and M02) for probe B were selected by the discriminant analysis algorithm as significant variables for classification, whereas only the CF was selected for probe A. In Figure 5, the importance of CF for probe A assays for classification purposes has already been shown. Table 2 shows the classification results and the error rates in estimating the normal cheeses as defective (type X error) and estimating the defective cheeses as normal (type Y error). For probe A, type X and Y classification error for the 40 pieces of the model was found to be 6 and 9%, respectively, because 1 normal cheese was classified as defective and 2 defective pieces as normal. On the other hand, the 10 wheel pieces used to validate the model were correctly classified. For probe B, type X error was 6%, and type Y error was 13% for the model, because 1 normal cheese was classified as defective and 3 defective pieces as normal. In the validation test for probe B, 1 piece of each group (normal and defective) was wrongly classified (10% for type X and Y errors). As a consequence, using the acoustic parameters to obtain the discriminant functions, the best predictor of defective cheeses was found to be probe A.
View this table:
[in this window]
[in a new window]
|
Table 2. Classification results from the discriminant analysis using the acoustic parameters and the principal components from the frequency spectrum, for probes A (articulated) and B (hand-operated)
|
|
On the other hand, a PCA was used to analyze the data set and reduce the information considered into a smaller set. In this analysis, 19 components were extracted for both acoustic impulse-response probes accounting for 98 and 99% (Table 3) of the variability for probes A and B, respectively. The number of components in the model is chosen according to the accuracy of prediction parameter Q2, this being 80.5% for probe A and 86.8% for probe B. Figure 6 shows the dispersion diagram between the first 2 components (C1 and C2), which accounts for the largest amount of explained variance (Table 3) for probe B. No differences can be found between normal and defective wheel pieces for the first component (C1), whereas C2 is lower for defective cheeses. Although C1 is the component that explains the highest amount of variability (54%; Table 3), it seems not to be related to the existence of internal defects. In this regard, the variability in the spectrum depicted by C1 could also be related to the textural properties, composition, or porosity of the cheese pieces. When the components obtained from the PCA were used in the discriminant analysis, 5 variables (C1, C2, C4, C7, C9) for probe A and 9 variables (C2, C3, C4, C7, C8, C9, C12, C14, C16) for probe B were selected. Therefore, it seems that the spectra variability caused by the cracks in the pieces is distributed in more variables in the case of probe B than in probe A. Using the principal components, the error rate in estimating the defective cheeses as normal (type Y error) was found to be 4% for the model and 10% for the validation for probe A, because 1 defective cheese in both batches was classified as normal. However, all normal cheeses were correctly classified (error type X = 0) for the samples used in model building and validation. For probe B, the 40 cheeses used to fit the model and the 10 samples used for validation were correctly classified. Thus, it seems that when using the principal components for calculating the discriminant functions, probe B provides better results than probe A for predicting cracks in Manchego cheese.
View this table:
[in this window]
[in a new window]
|
Table 3. Explained variance (%) of the principal components extracted from the principal component analysis of the spectra obtained with the acoustic probes A (articulated) and B (hand-operated)
|
|
View larger version (10K):
[in this window]
[in a new window]
|
Figure 6. Principal component C1 vs. C2 obtained from the principal component analysis carried out on the whole spectra for all defective and normal Manchego cheeses; probe B.
|
|
On average, type Y errors were higher than type X errors in all the experiments. This could be due to the variability that cracks of different sizes and locations can introduce into the frequency distribution of the sound emitted by cracked wheel pieces. The eggshell classification methods using a discriminant procedure also found that the error rates linked to identifying cracked eggs as normal were greater than those when normal eggs were identified as cracked (Cho et al., 2000).
Consequently, by comparing the 2 different procedures (acoustic variables and principal components) considered when carrying out the discriminant analysis for both probes, it can be concluded that the best discrimination procedure involves the use of the principal components extracted from the frequency spectrum for probe B.
|
CONCLUSIONS
|
|---|
Using ultrasonic pulse-echo experiments, it was not possible to detect internal cracks in Manchego cheese due to the typical porosity in this type of cheese that scatters the ultrasonic waves in the outermost layers. An alternative way to detect cracks in Manchego cheese was the use of the acoustic impulse-response technique. Cheese pieces with internal cracks or slits emitted a higher amount of energy, mainly in the range of frequencies higher than 150 Hz.
Discriminant functions including the acoustic parameters allowed for a good classification of cheese wheels, according to the existence or not of cracks. Nevertheless, when a PCA analysis was carried out on the set of spectra and the principal components used in the discrimination procedure, all the pieces used in this study could be correctly classified. This technique could be used online to detect internal cracks in a fast and inexpensive way.
|
ACKNOWLEDGEMENTS
|
|---|
Financial support for this research has been provided by the Spanish Ministerio de Ciencia y Tecnología, INIA (CAL01-077-C3-1).
Received for publication September 3, 2007.
Accepted for publication October 31, 2007.
|
REFERENCES
|
|---|
Benedito, J., J. Carcel, G. Clemente, and A. Mulet. 2000a. Cheese maturity assessment using ultrasonics. J. Dairy Sci. 83:248–254.[Abstract]
Benedito, J., J. Carcel, M. Gisbert, and A. Mulet. 2001. Quality control of cheese maturation and defects using ultrasonics. J. Food Sci. 66:100–104.
Benedito, J., J. Carcel, N. Sanjuan, and A. Mulet. 2000b. Use of ultrasound to assess Cheddar cheese characteristics. Ultrasonics 38:727–730.[CrossRef][Medline]
Benedito, J., T. Conde, G. Clemente, and A. Mulet. 2007. Use of the acoustic impulse-response technique for the non-destructive assessment of Manchego cheese texture. J. Dairy Sci. 89:4490–4502.
Benedito, J., S. Simal, G. Clemente, and A. Mulet. 2006. Manchego cheese texture evaluation by ultrasonics and surface probes. Int. Dairy J. 16:431–438.[CrossRef]
Butz, P., C. Hofmann, and B. Tauscher. 2005. Recent developments in noninvasive techniques for fresh fruit and vegetable internal quality analysis. J. Food Sci. 70:131–141.
Cho, H. K., W. K. Choi, and J. H. Paek. 2000. Detection of surface cracks in shell eggs by acoustic impulse method. Trans. ASAE 43:1921–1926.
Conde, T., J. A. Cárcel, J. V. García-Pérez, and J. Benedito. 2007. Non-destructive analysis of Manchego cheese texture using impact force-deformation and acoustic impulse response techniques. J. Food Eng. 82:238–245.[CrossRef]
De Ketelaere, B., P. Coucke, and J. De Baerdemaeker. 2000. Eggshell crack detection based on acoustic resonance frequency analysis. J. Agric. Eng. Res. 76:157–163.[CrossRef]
Diezma-Iglesias, B., M. Ruiz-Altisent, and P. Barreiro. 2004. Detection of internal quality in seedless watermelon by acoustic impulse response. Biosyst. Eng. 88:221–230.[CrossRef]
Diezma-Iglesias, B., M. Ruiz-Altisent, and P. Jancsók. 2005. Vibrational analysis of seedless watermelons: Use in the detection of internal hollows. Span. J. Agric. Res. 3:52–60.
Freitas, C., and F. X. Malcata. 2000. Microbiology and biochemistry of cheeses with Appélation d’Origine Protégée and manufactured in the Iberian Peninsula from ovine and caprine milks. J. Dairy Sci. 83:584–602.[Abstract]
Giangiacomo, R., G. Messina, and F. Abbiati. 1989. Confronto tra la risonanza acustica strumentale e la battitura per la valutazione commerciale di formaggio Grana. Latte 14:126–129.
Keener, K., R. Stroshine, and J. Nyenhuis. 1999. Evaluation of low field (5.40 MHz) proton magnetic resonance measurements of Dw and T2 as methods of non-destructive quality evaluation of apples. J. Am. Soc. Hortic. Sci. 124:289–295.
Lim, K. S., and M. Barigou. 2004. X-ray micro-computed tomography of cellular food products. Food Res. Intern. 37:1001–1012.[CrossRef]
McClements, D. J. 1997. Ultrasonic characterization of foods and drinks: Principles, methods, and applications. Crit. Rev. Food Sci. Nutr. 37:1–46.[Medline]
Patel, D., E. R. Davies, and I. Hannah. 1996. The use of convolution operators for detecting contaminants in food images. Pattern Recognit. 29:1019–1029.[CrossRef]
Pearson, T. C. 2001. Detection of pistachio nuts with closed shells using impact acoustics. Appl. Eng. Agric. 17:249–253.
Povey, M. J. W., and D. J. McClements. 1988. Ultrasonics in food engineering. Part I: Introduction and experimental methods. J. Food Eng. 8:217–245.
Rosenberg, M., M. McCarthy, and R. Kauten. 1991. Magnetic resonance imaging of cheese structure. Food Struct. 10:185–192.
Rosenberg, M., M. McCarthy, and R. Kauten. 1992. Evaluation of eye formation and structural quality of Swiss-type cheese by magnetic resonance imaging. J. Dairy Sci. 75:2083–2091.[Abstract]
Zapparoli, G. A. 1997. Raggi X per l formaggi. Latte 22:62–63.
TOP
ABSTRACT
INTRODUCTION
