Meal replacement powder grinding quality detection method

Abstract

The present application relates to the technical field of image processing, and particularly relates to a meal replacement powder grinding quality detection method; the method comprises the following steps: acquiring an image of produced meal replacement powder, and uniformly dividing the image to obtain a plurality of sub-blocks; acquiring a spectrum diagram of each sub-block, and extracting a low-frequency region where a low-frequency component is located from the spectrum diagram; marking a frequency point exceeding a set range in the low-frequency region as an out-of-limit frequency point; acquiring an abnormal sub-block; acquiring a discrete degree of a medium-frequency region where a medium-frequency component of a spectrum diagram corresponding to the abnormal sub-block is located, and determining that the abnormal sub-block is a mildew area when the discrete degree is greater than a set value; determining that the abnormal sub-block is a caking area when the discrete degree is less than or equal to the set value; and determining the quality of the meal replacement powder according to all mildew areas or caking areas. That is, the scheme of the present application can efficiently and accurately detect the quality of the meal replacement powder.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method for detecting the grinding quality of meal replacement powder. Background Technology

[0002] Meal replacement powder grinding quality testing mainly serves the food processing and nutritional supplement industry, covering the production of functional foods such as protein powder, dietary fiber powder, and vitamin powder.

[0003] Since the quality of meal replacement powders not only affects the product's safety, nutritional value, and market competitiveness, but also directly impacts consumers' health and user experience, quality testing of meal replacement powders is crucial.

[0004] Meal replacement powder quality issues typically manifest as mold and clumping. Mold is usually related to raw material contamination, substandard production hygiene, contamination during packaging, or problems during storage and transportation. Clumping is generally related to raw material characteristics, humidity levels (during production, storage, or transportation), grinding processes, and packaging sealing. Therefore, the wide range of raw material sources, complex processing techniques, and uncontrollable factors such as grinding and storage environments all affect the quality of meal replacement powder.

[0005] For the quality testing of meal replacement powders during production, existing technologies typically rely on a combination of multiple testing techniques and manual sampling. This approach suffers from low efficiency, high false positive rates, and insufficient sensitivity to defects such as clumping or mold, making it difficult to meet the demands of large-scale production and high-quality requirements for refined quality control. Consequently, it directly impacts the quality and safety of meal replacement powders and the consumer experience, increasing food safety risks.

[0006] Therefore, accurately detecting the quality of meal replacement powder, effectively identifying problems such as clumping and mold, improving the accuracy of detection algorithms, and reducing the risk of missed detections are crucial issues to avoid frequent production line shutdowns for verification, which could lead to raw material waste, batch-specific quality fluctuations in finished products, and increased quality control cost pressures on enterprises. Summary of the Invention

[0007] The purpose of this invention is to provide a method for testing the grinding quality of meal replacement powder, in order to solve the problem that the existing technology cannot efficiently and accurately test the quality of meal replacement powder; to this end, this invention provides a solution in one aspect.

[0008] This invention provides a method for testing the grinding quality of meal replacement powder, comprising:

[0009] Acquire an image of the produced meal replacement powder and divide the image into multiple sub-blocks evenly;

[0010] Obtain the spectrum diagram of each sub-block, and extract the low-frequency region where the low-frequency component is located from the spectrum diagram; mark the frequency points in the low-frequency region that exceed the set range as over-limit frequency points;

[0011] Calculate the first ratio of the number of all out-of-limit frequency points in the low-frequency region to the low-frequency components in the low-frequency region; obtain the second ratio of the number of low-frequency components in the low-frequency region to all components of the spectrum of the corresponding sub-block; if the first ratio is greater than or equal to the first threshold and the second ratio is greater than or equal to the second threshold, then mark the corresponding sub-block as an abnormal sub-block.

[0012] Obtain the dispersion of the intermediate frequency region where the intermediate frequency component of the spectrum corresponding to the abnormal sub-block is located; determine the abnormal sub-block with the dispersion greater than a set value as mold; determine the abnormal sub-block with the dispersion less than or equal to the set value as agglomeration.

[0013] The quality of the meal replacement powder is determined based on all moldy or clumped areas.

[0014] The above scheme reduces computation by dividing the image into blocks and analyzing each block separately. During analysis, the low-frequency region and the out-of-limit frequency points in the low-frequency region are extracted from each block to obtain two judgment conditions (first, the proportion of out-of-limit frequency points in the low-frequency region is higher than a first threshold; second, the proportion of low-frequency energy in the low-frequency region is higher than a second threshold). Based on the two judgment conditions, an abnormal sub-block is initially screened out. Then, the dispersion of the mid-frequency region is used for secondary discrimination to further determine the defect type of the abnormal sub-block, thus achieving efficient and accurate detection of the quality of the meal replacement powder.

[0015] Optionally, the dispersion for:

[0016] ;in, Let be the mean of all intermediate frequency components in the intermediate frequency region corresponding to the i-th sub-block. , These represent the mean and standard deviation of all mid-frequency components in the spectrum corresponding to the historical normal image, respectively.

[0017] The above scheme provides a method for accurately calculating the dispersion.

[0018] Optionally, the set range is a confidence interval, wherein the upper limit of the confidence interval is the sum of the mean of the amplitudes of all frequency points in the low-frequency region and the standard deviation of the set multiple, and the lower limit of the confidence interval is the difference between the mean of the amplitudes of all frequency points in the low-frequency region and the standard deviation of the set multiple.

[0019] The above method of setting confidence intervals can accurately obtain the frequency points exceeding the limit.

[0020] Optionally, the low-frequency region is defined as follows: the distance from each frequency point in the spectrum to the center of the spectrum is calculated, and the frequency points whose distance is less than a first distance threshold are recorded as low-frequency points, and the region formed by all low-frequency points is defined as the low-frequency region; the spectrum is obtained by performing a Fourier transform on each sub-block.

[0021] Optionally, the process of obtaining the intermediate frequency region is as follows:

[0022] In the spectrum diagram, the frequency points whose distance from each frequency point to the center of the spectrum is less than the second distance threshold and greater than or equal to the first distance threshold are recorded as intermediate frequency components, and the region formed by all intermediate frequency components is recorded as the intermediate frequency region.

[0023] The above scheme can accurately extract the intermediate frequency component.

[0024] Optionally, the first distance threshold is The second distance threshold is ;in, As the first coefficient, is the second coefficient, which is greater than the first coefficient; P is the length of the spectrum corresponding to the sub-block; Q is the width of the spectrum corresponding to the sub-block; and min() is the minimum value function.

[0025] Optionally, the method further includes a preprocessing step for the image, specifically:

[0026] The image is enhanced to obtain an enhanced image; and the enhanced image is smoothed using a fourth-order flat-top window method to obtain a smoothed image.

[0027] Optionally, the historical normal images are multiple historically acquired normal images of unadhesive blocks and mold-free areas.

[0028] Optionally, the step of determining the quality of the meal replacement powder based on all moldy or clumped areas includes: extracting the total number of pixels belonging to mold or clumping in all moldy or clumped areas; determining that the clumping content in the meal replacement powder exceeds the standard when the ratio of the total number of pixels corresponding to clumping to the total number of pixels in the image is greater than a first ratio threshold; and determining that the mold content in the meal replacement powder exceeds the standard when the ratio of the total number of pixels corresponding to mold to the total number of pixels in the image is greater than a second ratio threshold.

[0029] The above method can accurately determine whether the quality of meal replacement powder is up to standard by analyzing the proportion of clumped or moldy areas.

[0030] Optionally, extracting the total number of pixels belonging to mold or caking in all moldy or caking areas includes: performing edge detection on each moldy or caking area using the Canny edge detection algorithm to obtain the total number of pixels with abnormal caking or mold in the moldy or caking areas.

[0031] The beneficial effects of this invention are as follows:

[0032] In this invention, the image is analyzed in the frequency domain and then judged in two steps. The first judgment is made by using a "dual-dimensional joint judgment strategy" (the proportion of energy exceeding the limit frequency point in the low-frequency region is higher than the first threshold and the proportion of low-frequency energy in the low-frequency region is higher than the second threshold). The second judgment is made by using the dispersion of the mid-frequency component. In other words, the combined judgment of the two steps achieves efficient and accurate detection of the quality of the meal replacement powder. Attached Figure Description

[0033] Figure 1 The flowchart illustrating the steps of a method for testing the grinding quality of meal replacement powder in this embodiment is shown in the illustration. Detailed Implementation

[0034] Taking a certain meal replacement powder as an example, this invention introduces a method for testing the grinding quality of meal replacement powder.

[0035] Specifically, such as Figure 1 As shown, a method for detecting the grinding quality of meal replacement powder in this embodiment includes the following steps:

[0036] Step S1: Obtain an image of the meal replacement powder and divide the image into multiple sub-blocks evenly.

[0037] In this embodiment, an industrial camera is used to capture images of the meal replacement powder produced on the production line.

[0038] The image can be divided into multiple sub-blocks by uniformly dividing it.

[0039] In this embodiment, before image segmentation, a preprocessing step is included, specifically: the image is enhanced using the CLAHE algorithm to obtain an enhanced image; and the enhanced image is smoothed using a fourth-order flat-top window method to obtain a smoothed image.

[0040] It should be noted that applying a flat-top window to optimize the image can more accurately preserve the true amplitude distribution of low-frequency energy while suppressing the interference of high-frequency noise on edge signals. Specifically, a fourth-order flat-top window is applied to each pixel position in the image, and then the image preprocessed by CLAHE is multiplied with the fourth-order flat-top window, making the gray-level gradient of the edge region of the windowed image smooth.

[0041] Furthermore, before image segmentation, in order to improve the resolution of the spectrum in subsequent analysis, a zero-padding operation is performed on the image, that is, the image size is expanded from L×W to M×N, and then zero values ​​are filled in the rest of the image to obtain the zero-padding image.

[0042] Step S2: Filter out abnormal sub-blocks from the sub-blocks.

[0043] Specifically, the steps for obtaining the abnormal sub-block include the following:

[0044] Step S21: Obtain the spectrum of each sub-block and extract the low-frequency region where the low-frequency component is located from the spectrum.

[0045] In this embodiment, a two-dimensional fast Fourier transform is performed on each sub-block to obtain the corresponding spectrum.

[0046] The process of obtaining the low-frequency region in this embodiment is as follows: calculate the distance from each frequency point in the spectrum to the center of the spectrum, record the frequency points whose distance is less than the first distance threshold as low-frequency points, and take the region formed by all low-frequency points as the low-frequency region.

[0047] Specifically, taking the spectrum center (0, 0) of the spectrum graph as the origin, the distance from each frequency point to the spectrum center is calculated. When the distance is less than the first distance threshold, the frequency point belongs to the low frequency region. All frequency points are combined to form the low frequency region, where the value of the low frequency component of the frequency point in the low frequency region remains unchanged, and the value of other frequency points is set to 0.

[0048] The aforementioned low-frequency region is a circular area. It should be noted that the spectrum diagrams obtained in this embodiment are all centered spectrum diagrams (positions on the image represent frequencies, and brightness represents energy). A spectrum diagram can be viewed as a frequency "map," with the center point (0, 0) representing zero frequency, also called the DC component, which is the lowest frequency. Moving outwards from the center, the frequency amplitude continuously increases. Therefore, a preliminary low-frequency region can be delineated by distance.

[0049] Wherein, the first distance threshold is P is the length of the spectrogram corresponding to the sub-block, and Q is the width of the spectrogram corresponding to the sub-block. This is the first coefficient, which can be adjusted according to the actual size of the foreign object; the empirical value is 0.3.

[0050] Step S22: Mark frequency points in the low-frequency region that exceed the set range as over-limit frequency points.

[0051] In this embodiment, it is analyzed whether a low-frequency point in the low-frequency region is an out-of-limit frequency point. That is, when the amplitude of a low-frequency point in the low-frequency region exceeds a set range, it is marked as an out-of-limit frequency point in the low-frequency region. The set range is a confidence interval, where the upper limit of the confidence interval is the sum of the standard deviation of the mean amplitude of all frequency points in the low-frequency region and a set multiple, and the lower limit of the confidence interval is the difference between the mean amplitude of all frequency points in the low-frequency region and the standard deviation of the set multiple.

[0052] Specifically, the range is set as follows Where k is a set multiple, which is also a coefficient adjusted according to the false positive rate tolerance (e.g., when k=3, it corresponds to a 99.7% confidence interval).

[0053] Step S23: Calculate the first ratio of the number of all out-of-limit frequency points in the low-frequency region to the low-frequency components in the low-frequency region; obtain the second ratio of the number of low-frequency components in the low-frequency region to all components of the spectrum of the corresponding sub-block; if the first ratio is greater than or equal to the first threshold and the second ratio is greater than or equal to the second threshold, then mark the corresponding sub-block as an abnormal sub-block.

[0054] In this embodiment, the ratio of the number of frequency points exceeding the limit in each low-frequency region to the total number of frequency points in the low-frequency region is calculated, and this ratio is used as the first ratio; the ratio of the total number of frequency points in the low-frequency region to the total number of frequency points in the spectrum diagram of the corresponding sub-block is used as the second ratio.

[0055] The first threshold is 0.8; the second threshold is 0.6; of course, other implementation methods can also be determined according to the actual situation.

[0056] Step S3: Classify the abnormal sub-blocks to obtain sub-blocks belonging to the clumping category and sub-blocks belonging to the moldy category, and finally determine the quality of the meal replacement powder.

[0057] It should be noted that in this embodiment, it is considered that when the energy concentration of local abnormal frequency points in the low frequency region is significantly higher, if there are local defects such as adhesive blocks in the image, the corresponding low frequency region will show energy peak characteristics, and the proportion of low frequency energy is relatively high, reflecting the lack of high frequency components. This may be because the continuity of the material structure is destroyed, which is a typical signal characteristic of adhesive block type defects.

[0058] For moldy areas, local density anomalies caused by microbial activity or the accumulation of metabolites can also lead to a significant increase in the proportion of low-frequency energy. (It is worth noting that due to the synergistic effect of local structural damage caused by microbial metabolic activity and the accumulation of metabolites, some moldy features may also exhibit a high proportion of high-frequency energy. This case is not considered in this embodiment; only the case where moldy areas accompanied by local density anomalies may lead to an increase in the proportion of low-frequency energy is considered.) However, the disruption of texture continuity by hyphal networks or local cracks will manifest as a significant increase in the standard deviation of the mid-frequency amplitude spectrum; while the edge smoothing characteristics formed by the overall density change of agglomeration defects usually have a lower mid-frequency standard deviation, reflecting the uniformity of texture distribution.

[0059] Based on the above-mentioned differences, for abnormal sub-blocks, it is necessary to further distinguish the defect type by determining the threshold of the dispersion of intermediate frequency energy: when the dispersion exceeds the threshold, it is determined to be moldy; otherwise, it is determined to be clumping.

[0060] Specifically, the dispersion of the intermediate frequency region where the intermediate frequency component of the spectrum corresponding to the abnormal sub-block is located is obtained, and the abnormal sub-block with the dispersion greater than a set value is determined to be moldy; the abnormal sub-block with the dispersion less than or equal to the set value is determined to be agglomerate.

[0061] In one embodiment, frequency points that are less than a second distance threshold and greater than or equal to a first distance threshold are designated as intermediate frequency points, and all intermediate frequency points constitute an intermediate frequency region.

[0062] Wherein, the second distance threshold is ;in The second coefficient can be determined based on experience. In this embodiment, the value can be 0.7. P is the length of the spectrum corresponding to the sub-block, and Q is the width of the spectrum corresponding to the sub-block.

[0063] In another embodiment, the method for extracting the intermediate frequency region can retain the intermediate frequency components by using a set bandpass filter.

[0064] After obtaining the intermediate frequency region, the amplitude spectrum of the intermediate frequency region is statistically analyzed, that is, the mean of the amplitude values ​​of all intermediate frequency points in the region is calculated, and this mean is used as the key feature quantity of the uniformity of the intermediate frequency signal distribution of the sub-block.

[0065] To avoid interference from false detection sources such as equipment vibration frequency bands, multiple normal images of the unbonded block and moldy block are pre-acquired. Based on these images, the normal standard deviation and mean of the mid-frequency region are calculated. The mean of the mid-frequency component in the mid-frequency region of the current sub-block is then Z-score standardized using the normal mean and amplitude spectrum standard deviation to obtain the mid-frequency dispersion.

[0066] Among them, dispersion for: ;in, Let be the mean of all intermediate frequency components in the intermediate frequency region corresponding to the i-th sub-block. , These represent the mean and standard deviation of all mid-frequency components in the spectrum corresponding to the historical normal image, respectively.

[0067] The above-mentioned historical normal images are multiple normal images collected in the past, without any adhesive blocks or mold.

[0068] In this embodiment, a set value is set. When the dispersion is greater than the set value, it is determined that mold exists. Otherwise, the abnormal sub-block has clumping.

[0069] The above setting value can be obtained through historical data statistics; specifically, multiple historical normal samples (defect-free) and known moldy / caking samples can be collected, and the average of the historical dispersion of the mid-frequency components of all samples can be used as the setting value.

[0070] Furthermore, the quality of the meal replacement powder is determined based on all moldy or clumped areas, including: extracting the total number of pixels belonging to mold or clumping in all moldy or clumped areas; when the ratio of the total number of pixels corresponding to clumping to the total number of pixels in the image is greater than a first ratio threshold, it is determined that the clumping content in the meal replacement powder exceeds the standard; when the ratio of the total number of pixels corresponding to mold to the total number of pixels in the image is greater than a second ratio threshold, it is determined that the mold content in the meal replacement powder exceeds the standard.

[0071] The first and second proportional thresholds mentioned above can both be obtained based on experience; for example, both the first and second proportional thresholds can be 0.03. Of course, the setting of the first and second proportional thresholds can also be determined according to the actual situation.

[0072] In one embodiment, the Canny edge detection algorithm is used to perform edge detection on each moldy or clumped region to obtain the total number of pixels with abnormal clumping or mold growth in the moldy or clumped region.

[0073] In another embodiment, each moldy or clumped region can be segmented to obtain a target region for each moldy or clumped region, and the total number of pixels in the target region can be obtained.

[0074] In this embodiment, the characteristics of the agglomerated block are a high proportion of low-frequency energy and low dispersion of mid-frequency energy. The former reveals the continuity of the overall structure, while the latter reflects the uniformity of the mid-frequency distribution. For the moldy region, the aggregation of metabolic products results in a high proportion of low-frequency energy, but the dispersion of mid-frequency energy is also relatively high. Therefore, the agglomerated region or the moldy region is distinguished by calculating the dispersion of the mid-frequency region.

[0075] The quality inspection method of this invention employs a "dual-dimensional joint judgment strategy": First, preliminary screening is performed using two dimensions: "the proportion of energy exceeding the limit frequency point in the low-frequency region is higher than a first threshold" and "the proportion of low-frequency energy in the low-frequency region is higher than a second threshold." A high proportion of low-frequency energy indicates a continuous change in the overall structure of the detection area (such as large-scale adhesive lumps), while a high proportion of abnormal frequency point energy in the low-frequency region further identifies abnormal areas with concentrated local energy (such as small particle lumps or early-stage mold lesions), thus accurately delineating regions of interest suspected of adhesive lumps or mold. Subsequently, a second judgment is made based on the dispersion of the mid-frequency region: if the dispersion is lower than the threshold corresponding to mold characteristics, the area is confirmed as an adhesive lump; otherwise, it points to texture-damaging defects such as mold, achieving efficient and accurate detection of meal replacement powder quality.

[0076] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

[0077] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for testing the grinding quality of meal replacement powder, characterized in that, include: Acquire an image of the produced meal replacement powder and divide the image into multiple sub-blocks evenly; The spectrum diagrams of each sub-block are obtained, and the low-frequency regions containing low-frequency components are extracted from the spectrum diagrams. Frequency points in the low-frequency regions that exceed a set range are marked as over-limit frequency points. The set range is a confidence interval, where the upper limit of the confidence interval is the sum of the standard deviation of the mean amplitude of all frequency points in the low-frequency region and the set multiple, and the lower limit of the confidence interval is the difference between the standard deviation of the mean amplitude of all frequency points in the low-frequency region and the set multiple. The spectrum diagrams are all centered spectrum diagrams. The position on the image represents frequency, and the brightness represents energy. The spectrum diagram can be regarded as a frequency "map". The center point of the map (0, 0) is the zero frequency, which is the lowest frequency. Moving outward from the center, the frequency amplitude continuously increases. Calculate the first ratio of the number of all out-of-limit frequency points in the low-frequency region to the low-frequency components in the low-frequency region; Obtain the second ratio of the number of low-frequency components in the low-frequency region to the total number of components in the spectrum of the corresponding sub-block; If the first ratio is greater than or equal to the first threshold and the second ratio is greater than or equal to the second threshold, then the corresponding sub-block is marked as an abnormal sub-block; Obtain the dispersion of the intermediate frequency region where the intermediate frequency component of the spectrum corresponding to the abnormal sub-block is located; determine the abnormal sub-block with the dispersion greater than a set value as a moldy region; determine the abnormal sub-block with the dispersion less than or equal to the set value as a clumping region. The quality of the meal replacement powder is determined based on all moldy or clumped areas.

2. The method for detecting the grinding quality of meal replacement powder according to claim 1, characterized in that, The degree of dispersion for: ;in, Let be the mean of all intermediate frequency components in the intermediate frequency region corresponding to the i-th sub-block. , These represent the mean and standard deviation of all mid-frequency components in the spectrum corresponding to the historical normal image, respectively.

3. The method for detecting the grinding quality of meal replacement powder according to claim 2, characterized in that, The low-frequency region is defined as follows: calculate the distance from each frequency point in the spectrum to the center of the spectrum, record the frequency points whose distance is less than the first distance threshold as low-frequency points, and define the region formed by all low-frequency points as the low-frequency region. The spectrum is obtained by performing a Fourier transform on each sub-block.

4. The method for detecting the grinding quality of meal replacement powder according to claim 3, characterized in that, The process of obtaining the intermediate frequency region is as follows: In the spectrum diagram, the frequency points whose distance from each frequency point to the center of the spectrum is less than the second distance threshold and greater than or equal to the first distance threshold are recorded as intermediate frequency components, and the region formed by all intermediate frequency components is recorded as the intermediate frequency region.

5. The method for detecting the grinding quality of meal replacement powder according to claim 4, characterized in that, The first distance threshold is The second distance threshold is ;in, As the first coefficient, is the second coefficient, which is greater than the first coefficient; P is the length of the spectrum corresponding to the sub-block; Q is the width of the spectrum corresponding to the sub-block; and min() is the minimum value function.

6. The method for detecting the grinding quality of meal replacement powder according to claim 1, characterized in that, It also includes a preprocessing step for the image, specifically: The image is enhanced to obtain an enhanced image; and the enhanced image is smoothed using a fourth-order flat-top window method to obtain a smoothed image.

7. The method for detecting the grinding quality of meal replacement powder according to claim 2, characterized in that, The historical normal images are multiple historically acquired normal images of unbonded blocks and mold-free areas.

8. The method for detecting the grinding quality of meal replacement powder according to claim 1, characterized in that, The step of determining the quality of the meal replacement powder based on all moldy or clumped areas includes: extracting the total number of pixels belonging to mold or clumping in all moldy or clumped areas; determining that the clumping content in the meal replacement powder exceeds the standard when the ratio of the total number of pixels corresponding to clumping to the total number of pixels in the image is greater than a first ratio threshold; and determining that the mold content in the meal replacement powder exceeds the standard when the ratio of the total number of pixels corresponding to mold to the total number of pixels in the image is greater than a second ratio threshold.

9. The method for detecting the grinding quality of meal replacement powder according to claim 8, characterized in that, The step of extracting the total number of pixels belonging to mold or caking in all moldy or caking areas includes: using the Canny edge detection algorithm to perform edge detection on each moldy or caking area to obtain the total number of pixels with abnormal caking or abnormal mold in the moldy or caking areas.

Images