A method of assessing the extent of lipid oxidation in an infant formula

Abstract

The present application relates to the technical field of milk powder evaluation, and discloses a method for evaluating the lipid oxidation degree of infant formula milk powder, which comprises the following steps: obtaining reference oxidation values of infant formula milk powder under different oxidation degrees; performing color space conversion and region standardization processing on the digital image of the infant formula milk powder under controlled light conditions to obtain a standardized sample image of the infant formula milk powder; extracting a multi-dimensional color feature vector of the infant formula milk powder from the standardized sample image; taking the reference oxidation values as target variables and the corresponding multi-dimensional color feature vectors as input features to establish a color-oxidation quantitative prediction model of the lipid oxidation degree and color of the infant formula milk powder; inputting the multi-dimensional color feature vector of the infant formula milk powder to be evaluated into the color-oxidation quantitative prediction model to output an oxidation index value of the infant formula milk powder to be evaluated; and the present application can reduce the damage degree of detecting the lipid oxidation degree of milk powder.

Description

Technical Field

[0001] This invention relates to the field of milk powder evaluation technology, and in particular to a method for assessing the degree of lipid oxidation in infant formula milk powder. Background Technology

[0002] In the field of quality control for infant formula, existing technologies typically rely on laboratory chemical analysis methods to determine the degree of lipid oxidation. These methods require professional operation and involve reagent use, sample pretreatment, and complex instrument analysis. The overall process is time-consuming and cannot meet the needs of rapid screening of large numbers of samples during production or distribution.

[0003] Existing detection methods are typically destructive, require sample consumption, and are difficult to implement in-situ or online monitoring. Their periodic sampling model also results in insufficient real-time control over the oxidation state of products, making it difficult to achieve early warning and timely intervention for oxidation deterioration, and has significant limitations in terms of timeliness and application flexibility. Summary of the Invention

[0004] This invention provides a method for assessing the degree of lipid oxidation in infant formula milk powder, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for assessing the degree of lipid oxidation in infant formula milk powder, comprising: S1. Obtain the baseline oxidation value of infant formula milk powder under different oxidation levels; S2. Perform color space conversion and region standardization processing on the digital image of the infant formula milk powder under controlled lighting conditions to obtain a standardized sample image of the infant formula milk powder; S3. Extract the multidimensional color feature vector of the infant formula milk powder from the standardized sample image; S4. Using the baseline oxidation value as the target variable and the corresponding multidimensional color feature vector as the input feature, establish a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in the infant formula milk powder. S5. Input the multidimensional color feature vector of the infant formula milk powder to be evaluated into the color-oxidation quantitative prediction model, and output the oxidation index value of the infant formula milk powder to be evaluated.

[0006] In a preferred embodiment, obtaining the baseline oxidation value of infant formula milk powder under different oxidation levels includes: Multiple basic samples of infant formula milk powder with the same raw materials and processes were prepared. The basic samples were stored under different preset environmental conditions to obtain induced oxidation sample sets with different degrees of oxidation. The peroxide value of each sample in the induced oxidation sample set was determined, and the peroxide value was used as the baseline oxidation value of the sample.

[0007] In a preferred embodiment, the color space conversion of the digital image of the infant formula milk powder under controlled lighting conditions includes: Acquire the original image of the infant formula milk powder under standard light source illumination conditions; The original image is subjected to white balance correction to obtain a color-constant image; The color constant image is transformed into a color space to obtain a preliminary transformed image.

[0008] In a preferred embodiment, after performing color space conversion on the color constant image to obtain a preliminary converted image, the method further includes: The preliminary converted image is segmented by color thresholding to obtain an image of the milk powder region in the preliminary converted image; The image of the milk powder region is scaled to a preset uniform pixel size to obtain a size-normalized image; The size-normalized image is subjected to edge smoothing processing to obtain a standardized sample image of the infant formula milk powder.

[0009] In a preferred embodiment, extracting the multidimensional color feature vector of the infant formula milk powder from the standardized sample image includes: The average values ​​of the luminance component, red-green component, and yellow-blue component in the milk powder region of the standardized sample image are calculated respectively to obtain the basic color mean characteristics of the infant formula milk powder. Calculate the standard deviation of the luminance component, the standard deviation of the red-green component, and the standard deviation of the yellow-blue component within the milk powder region to obtain the discrete color distribution characteristics of the infant formula milk powder. The basic color mean feature and the color distribution discrete feature are combined to obtain the overall statistical feature components of the infant formula milk powder.

[0010] In a preferred embodiment, extracting the multidimensional color feature vector of the infant formula milk powder from the standardized sample image further includes: Based on the numerical distribution of red and green components, areas with abnormal color within the milk powder region are identified, thus obtaining suspected oxidation spot areas of the infant formula milk powder. The mean value of the suspected oxidation spot region on the red-green component is denoted as the spot region mean value. The ratio of the mean value of the spot area to the mean value of the overall red-green component in the milk powder area is calculated to obtain the spot contrast characteristics of the infant formula milk powder; The spot contrast features are used as local abnormal feature components of the infant formula milk powder; The local abnormal feature components are combined with the overall statistical feature components to form the multidimensional color feature vector of the infant formula milk powder.

[0011] In a preferred embodiment, the step of establishing a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in infant formula milk powder, using the baseline oxidation value as the target variable and the corresponding multidimensional color feature vector as the input feature, includes: The baseline oxidation value is paired with the corresponding multidimensional color feature vector to obtain the model training sample set; The training sample set of the model is divided into a training subset and a validation subset; Using the training subset, with the multidimensional color feature vector as input and the baseline oxidation value as output target, a regression model is trained to obtain an initial prediction model. The initial prediction model is validated using the validation subset, and the model parameters are adjusted according to the validation results until the model performance meets the preset requirements, thereby obtaining a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in the infant formula milk powder.

[0012] In a preferred embodiment, the step of validating the initial prediction model using the validation subset and adjusting the model parameters based on the validation results until the model performance meets preset requirements, thereby obtaining a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in the infant formula milk powder, includes: The multidimensional color feature vectors in the validation subset are input into the initial prediction model to obtain the corresponding model-predicted oxidation values. Calculate the prediction error index between the oxidation value predicted by the model and the corresponding baseline oxidation value in the validation subset; The generalization performance of the initial prediction model is evaluated based on the prediction error index between the oxidation value predicted by the model and the corresponding benchmark oxidation value in the validation subset. If the generalization performance does not meet the preset requirements, the model parameters of the initial prediction model are adjusted according to the direction indicated by the prediction error index. If the generalization performance meets the preset requirements, then the initial prediction model is determined to be a trained color-oxidation quantitative prediction model.

[0013] In a preferred embodiment, the calculation formula for the color-oxidation quantitative prediction model is as follows: ; In the formula, To predict oxidation values ​​for the model, This is the scaling factor. The total number of color features in the multidimensional color feature vector. Let be the ordinal number of the color feature in the multidimensional color feature vector. In order to be with the first Color features The corresponding feature weight coefficients, The first component constituting the overall statistical feature component A color feature, These are the weighting coefficients corresponding to the local anomaly feature components. For the local anomaly feature components, For bias terms, It is the hyperbolic tangent activation function.

[0014] In a preferred embodiment, the step of inputting the multidimensional color feature vector of the infant formula milk powder to be evaluated into the color-oxidation quantitative prediction model and outputting the oxidation index value of the infant formula milk powder to be evaluated includes: Images of the infant formula milk powder to be evaluated were acquired and processed to obtain standardized images for evaluation. Extract the multidimensional color feature vector of the infant formula milk powder to be evaluated from the standardized image to be evaluated; The multidimensional color feature vector to be tested is input into the color-oxidation quantitative prediction model to obtain the oxidation index value of the degree of lipid oxidation of the infant formula milk powder to be evaluated.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This method achieves rapid, non-destructive detection based on image analysis by establishing a quantitative prediction model between color features and lipid oxidation levels. This technology utilizes standardized image processing and multidimensional color feature extraction to significantly improve the automation level and execution efficiency of oxidation degree assessment, while reducing the time and resource consumption of traditional chemical analysis. It is suitable for rapid screening and quality monitoring scenarios on production lines.

[0016] 2. This method enhances the model's ability to identify subtle color changes and oxidation spots by fusing overall statistical features with local anomalies, thereby improving the accuracy and stability of oxidation degree prediction. This technology exhibits good repeatability and adaptability, providing reliable technical support for the quality control of infant formula milk powder and facilitating real-time monitoring and early warning of oxidation status. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for assessing the degree of lipid oxidation in infant formula milk powder according to an embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a method for assessing the degree of lipid oxidation in infant formula milk powder. The execution entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for assessing the degree of lipid oxidation in infant formula milk powder can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating a method for assessing the degree of lipid oxidation in infant formula milk powder according to an embodiment of the present invention. In this embodiment, the method for assessing the degree of lipid oxidation in infant formula milk powder includes: S1. Obtain the baseline oxidation value of infant formula milk powder under different oxidation levels; In this embodiment of the invention, obtaining the baseline oxidation value of infant formula milk powder under different oxidation levels includes: Multiple basic samples of infant formula milk powder with the same raw materials and processes were prepared. The basic samples were stored under different preset environmental conditions to obtain induced oxidation sample sets with different degrees of oxidation. The peroxide value of each sample in the induced oxidation sample set was determined, and the peroxide value was used as the baseline oxidation value of the sample.

[0021] Infant formula milk powder raw materials from the same batch and place of origin were selected, and multiple basic samples of infant formula milk powder with consistent raw materials and processes, uniform appearance and particle size were prepared by repeating the operation according to a fixed process.

[0022] Take equal amounts of basic samples and put them into clean, sealed containers of the same specifications and materials. Place the containers in different preset single-variable storage environments (the variable is one of temperature, humidity or oxygen concentration, and the other conditions are fixed) and store them stably for a preset time to obtain an induced oxidation sample set containing multiple samples with different oxidation levels.

[0023] The iodometric method was used. Each induced oxidation sample was quantitatively dissolved, potassium iodide was added to react and generate iodine, and then titrated to the endpoint with sodium thiosulfate standard solution. The peroxide value of the sample was calculated and used as the baseline oxidation value for the corresponding induced oxidation sample. This process was repeated for all samples. S2. The digital images of the infant formula milk powder under controlled illumination were subjected to color space conversion and region standardization to obtain standardized sample images of the infant formula milk powder. In this embodiment of the invention, the color space conversion of the digital image of the infant formula milk powder under controlled illumination includes: Acquire the original image of the infant formula milk powder under standard light source illumination conditions; The original image is subjected to white balance correction to obtain a color-constant image; The color constant image is transformed into a color space to obtain a preliminary transformed image.

[0024] After performing color space conversion on the color constant image to obtain a preliminary converted image, the process further includes: The preliminary converted image is segmented by color thresholding to obtain an image of the milk powder region in the preliminary converted image; The image of the milk powder region is scaled to a preset uniform pixel size to obtain a size-normalized image; The size-normalized image is subjected to edge smoothing processing to obtain a standardized sample image of the infant formula milk powder.

[0025] A standard light source illumination platform was set up, using a D65 standard light source (color temperature 6500K, color rendering index ≥95). The light source was fixed at a preset height directly above the sample, ensuring that the light source angle was perpendicular to the surface of the infant formula milk powder sample, and that the illumination range completely covered the sample without shadows, reflections, or uneven lighting areas. The infant formula milk powder sample was laid flat on a clean, colorless, and highly reflective stage, with a uniform thickness and no lumps or gaps. A high-definition camera was fixed directly above the sample, with the camera lens parallel to the sample surface. The camera focus was adjusted to ensure a clear image of the sample, and the shooting parameters were fixed and kept consistent. The camera was then started to capture the original image of the infant formula milk powder. The original image must completely present the entire milk powder sample without omissions, distortion, or interference from external debris.

[0026] The gray-world method is used to perform white balance correction on the original image. A neutral gray area without interference from other colors is selected in the original image, and the pixel values ​​of the red, green, and blue channels of all pixels in this area are extracted. The average value of the pixel values ​​of the three channels is calculated, and the average value of the three channels is used as the standard neutral gray reference value. The pixel values ​​of the red, green, and blue channels are adjusted respectively so that the average value of the pixel value of each channel is equal to the standard neutral gray reference value. During the adjustment process, the color relative relationship between each pixel remains unchanged. After the correction is completed, a color-constant image is obtained. The color-constant image can eliminate the influence of different lighting on the color presentation of milk powder and ensure the stability of the color characteristics of milk powder itself.

[0027] Color space conversion is performed on the color constant image using the RGB to Lab color space conversion method. First, the red, green, and blue channel pixel values ​​of each pixel in the color constant image are normalized, converting each channel pixel value into a value range of 0-1. Then, according to the correspondence between RGB and Lab color spaces, the normalized red, green, and blue values ​​of each pixel are converted into three components of the Lab color space, where the L component represents brightness, the a component represents the red-green difference, and the b component represents the blue-yellow difference. After completing the color space conversion of all pixels, a preliminary converted image is obtained. The preliminary converted image can highlight the color characteristics of the milk powder, which is convenient for subsequent region segmentation operations.

[0028] The Otsu's method was used to perform color thresholding on the initial converted image. The b-component of the Lab color space in the initial converted image was extracted (the difference between the blue and yellow of the milk powder and the background was the most significant). The pixel value distribution of the b-component was statistically analyzed, and the pixel value that maximized the inter-class variance between the milk powder region and the background region was determined as the segmentation threshold. The segmentation rule was set as follows: regions with pixel b-component values ​​greater than the segmentation threshold were identified as background regions, and regions with pixel b-component values ​​less than or equal to the segmentation threshold were identified as milk powder regions. Based on this rule, the background regions in the image were removed, and the pixel regions corresponding to the milk powder were retained, resulting in the milk powder region image in the initial converted image. The milk powder region image only contains the image portion corresponding to infant formula milk powder and has no background interference.

[0029] The preset uniform pixel size is 256×256 pixels (to meet the standard size requirements of subsequent image analysis). The milk powder area image is scaled to this preset size using bilinear interpolation. During the operation, the scaling ratio between the original pixel size of the milk powder area image and the preset size is first determined. Based on the scaling ratio, the corresponding position of each pixel in the original image after scaling is calculated. The pixel values ​​of the four neighboring pixels around the corresponding position are extracted. The pixel value of the scaled pixel is obtained through weighted calculation. The weight is determined by the distance between the corresponding position of the pixel and the neighboring pixels. The closer the distance, the greater the weight. After all pixels are calculated and filled, a size-normalized image is obtained. The size-normalized image can eliminate the analysis error caused by the size difference of different sample images.

[0030] Gaussian filtering was used to smooth the edges of the size-normalized image. A 3×3 filter template was selected and was used to cover each pixel of the size-normalized image in turn. The average pixel value of all pixels within the template coverage area was calculated and used to replace the pixel value at the center of the template. During the replacement process, the overall image outline was kept unchanged. The filtering process was completed for all pixels in the image in turn to eliminate noise and jagged edges of the image, resulting in a standardized sample image of infant formula milk powder. The standardized sample image can be used for subsequent milk powder quality analysis.

[0031] S3. Extract the multidimensional color feature vector of the infant formula milk powder from the standardized sample image; In this embodiment of the invention, extracting the multidimensional color feature vector of the infant formula milk powder from the standardized sample image includes: The average values ​​of the luminance component, red-green component, and yellow-blue component in the milk powder region of the standardized sample image are calculated respectively to obtain the basic color mean characteristics of the infant formula milk powder. Calculate the standard deviation of the luminance component, the standard deviation of the red-green component, and the standard deviation of the yellow-blue component within the milk powder region to obtain the discrete color distribution characteristics of the infant formula milk powder. The basic color mean feature and the color distribution discrete feature are combined to obtain the overall statistical feature components of the infant formula milk powder.

[0032] The step of extracting the multidimensional color feature vector of the infant formula milk powder from the standardized sample image further includes: Based on the numerical distribution of red and green components, areas with abnormal color within the milk powder region are identified, thus obtaining suspected oxidation spot areas of the infant formula milk powder. The mean value of the suspected oxidation spot region on the red-green component is denoted as the spot region mean value. The ratio of the mean value of the spot area to the mean value of the overall red-green component in the milk powder area is calculated to obtain the spot contrast characteristics of the infant formula milk powder; The spot contrast features are used as local abnormal feature components of the infant formula milk powder; The local abnormal feature components are combined with the overall statistical feature components to form the multidimensional color feature vector of the infant formula milk powder.

[0033] Select the milk powder region from the standardized sample image, and extract the Lab color space components of all pixels within this region. The luminance component is the L component, the red-green component is the a component, and the yellow-blue component is the b component. Enumerate each pixel within the milk powder region, recording the L, a, and b component values ​​for each pixel. Count the total number of pixels within the milk powder region. Sum the L component values ​​of all pixels to obtain the total L component value. Divide the total L component value by the total number of pixels to obtain the average luminance component value. Using the same method, sum the a component values ​​of all pixels to obtain the total a component value, and divide by the total number of pixels to obtain the average red-green component value. Sum the b component values ​​of all pixels to obtain the total b component value, and divide by the total number of pixels to obtain the average yellow-blue component value. These three average values ​​together constitute the basic color mean characteristics of infant formula milk powder.

[0034] Based on the extracted L, a, and b component values ​​of all pixels within the milk powder region and their corresponding basic color mean features, the standard deviation of each component is calculated. For the standard deviation of the luminance component, the difference between the L component value of each pixel and the average value of the luminance component is first calculated. Each difference is squared to obtain the squared difference. The sum of the squared differences of all pixels is then obtained. The sum of the squared differences is divided by the total number of pixels to obtain the variance of the luminance component. The square root of this variance is then taken to obtain the standard deviation of the luminance component. Using the same method, the difference between the a component value of each pixel and the average value of the red and green components is calculated and squared. The sum of the squared differences is then divided by the total number of pixels to obtain the variance of the red and green components. The square root of this variance is then taken to obtain the standard deviation of the red and green components. Similarly, the difference between the b component value of each pixel and the average value of the yellow and blue components is calculated and squared. The sum of the squared differences is then divided by the total number of pixels to obtain the variance of the yellow and blue components. The square root of this variance is then taken to obtain the standard deviation of the yellow and blue components. These three standard deviations together constitute the discrete color distribution characteristics of infant formula milk powder.

[0035] The average values ​​of the brightness component, red-green component, and yellow-blue component included in the basic color mean feature are combined with the standard deviations of the brightness component, red-green component, and yellow-blue component included in the color distribution discrete feature. The combination is carried out in a fixed order of "average value of brightness component, average value of red-green component, average value of yellow-blue component, standard deviation of brightness component, standard deviation of red-green component, and standard deviation of yellow-blue component" to form a set of six feature parameters. This set is the overall statistical feature component of infant formula milk powder. The original values ​​of each feature parameter are retained during the integration process, without changing the value magnitude or the meaning of the representation.

[0036] Focusing on the distribution of red-green component (a-component) values ​​within the milk powder area, the numerical range of the a-component of all pixels within the milk powder area is first statistically analyzed. The difference between the average value of the red-green component and twice the standard deviation of the red-green component is calculated as the lower limit for anomaly judgment, and the sum of the average value of the red-green component and twice the standard deviation of the red-green component is calculated as the upper limit for anomaly judgment. The judgment rule is set as follows: areas where the pixel a-component value is less than the lower limit or greater than the upper limit are judged as color abnormality areas. Each pixel within the milk powder area is enumerated one by one, and all color abnormality pixels are selected according to the above rule. Connectivity analysis is performed on these color abnormality pixels, and adjacent color abnormality pixels are merged to form a continuous area. This continuous area is the suspected oxidation spot area of ​​infant formula milk powder. The suspected oxidation spot area must not overlap with the boundary of the milk powder area, and the number of pixels in a single suspected oxidation spot area must not be less than 3 (to avoid misjudgment by a single noise pixel).

[0037] Extract all pixels within the suspected oxidation spot area, enumerate each pixel in the area one by one, record the red-green component (a component) value of each pixel, count the total number of pixels in the suspected oxidation spot area, add up the a component values ​​of all pixels to obtain the sum of the a components of the spot area, divide the sum of the a components of the spot area by the total number of pixels in the suspected oxidation spot area, and the result is the mean of the spot area. This mean of the spot area only represents the average color level of the suspected oxidation spot area in the red-green component.

[0038] The mean value of the spot area and the average value of the overall red and green components in the milk powder area are retrieved. The mean value of the spot area is used as the dividend and the average value of the overall red and green components in the milk powder area is used as the divisor. A division operation is performed, and the result is retained to four decimal places. The quotient is the spot contrast feature of the infant formula milk powder. This feature is used to characterize the degree of color difference between the suspected oxidized spot area and the overall milk powder area in terms of red and green components.

[0039] The calculated spot contrast features are used as a separate set of feature parameters. This set of feature parameters is the local abnormal feature component of infant formula milk powder. The local abnormal feature component focuses only on the difference in red and green components between the suspected oxidation spot area and the whole milk powder area, accurately characterizing the local color abnormality of milk powder.

[0040] The local abnormal feature components are combined with the overall statistical feature components, with the combination order fixed as "overall statistical feature components first, local abnormal feature components second". The resulting vector contains seven feature parameters, which is the multidimensional color feature vector of infant formula milk powder, and can be used for subsequent analysis to determine the degree of oxidation of infant formula milk powder.

[0041] S4. Using the baseline oxidation value as the target variable and the corresponding multidimensional color feature vector as the input feature, establish a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in the infant formula milk powder. In this embodiment of the invention, the step of establishing a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in infant formula milk powder, using the baseline oxidation value as the target variable and the corresponding multidimensional color feature vector as the input feature, includes: The baseline oxidation value is paired with the corresponding multidimensional color feature vector to obtain the model training sample set; The training sample set of the model is divided into a training subset and a validation subset; Using the training subset, with the multidimensional color feature vector as input and the baseline oxidation value as output target, a regression model is trained to obtain an initial prediction model. The initial prediction model is validated using the validation subset, and the model parameters are adjusted according to the validation results until the model performance meets the preset requirements, thereby obtaining a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in the infant formula milk powder.

[0042] The process of validating the initial prediction model using the validation subset and adjusting the model parameters based on the validation results until the model performance meets the preset requirements, thereby obtaining a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in infant formula milk powder, includes: The multidimensional color feature vectors in the validation subset are input into the initial prediction model to obtain the corresponding model-predicted oxidation values. Calculate the prediction error index between the oxidation value predicted by the model and the corresponding baseline oxidation value in the validation subset; The generalization performance of the initial prediction model is evaluated based on the prediction error index between the oxidation value predicted by the model and the corresponding benchmark oxidation value in the validation subset. If the generalization performance does not meet the preset requirements, the model parameters of the initial prediction model are adjusted according to the direction indicated by the prediction error index. If the generalization performance meets the preset requirements, then the initial prediction model is determined to be a trained color-oxidation quantitative prediction model.

[0043] The calculation formula for the color-oxidation quantitative prediction model is as follows: ; In the formula, To predict oxidation values ​​for the model, This is the scaling factor. The total number of color features in the multidimensional color feature vector. Let be the ordinal number of the color feature in the multidimensional color feature vector. In order to be with the first Color features The corresponding feature weight coefficients, The first component constituting the overall statistical feature component A color feature, These are the weighting coefficients corresponding to the local anomaly feature components. For the local anomaly feature components, For bias terms, It is the hyperbolic tangent activation function.

[0044] Retrieve the baseline oxidation value corresponding to each induced oxidation sample of infant formula milk powder, as well as the multidimensional color feature vector obtained after image standardization and feature extraction of the sample. Ensure that each set of baseline oxidation values ​​and multidimensional color feature vectors correspond one-to-one, and the correspondence is determined based on the same induced oxidation sample with no cross-matching errors. Pair each set of corresponding baseline oxidation values ​​with multidimensional color feature vectors to form a single model training sample. Summarize all the paired single model training samples to obtain the model training sample set. Each sample in the model training sample set contains complete input features (multidimensional color feature vector) and output target (baseline oxidation value).

[0045] A fixed-ratio partitioning method was used to divide the model training sample set into a training subset and a validation subset, with a partitioning ratio of 7:3 (adapting to the conventional data distribution ratio for regression model training, balancing training effectiveness and validation reliability). During the partitioning process, sample randomness was maintained to avoid the concentration of samples with the same oxidation level in a single subset. First, all samples in the model training sample set were randomly sorted. Then, the top 70% of the samples were selected according to the sorting results to form the training subset, and the remaining 30% of the samples were selected to form the validation subset. After the partitioning was completed, the number of samples in the two subsets was counted to ensure that the number of samples in the training subset was sufficient to support model training, and that the number of samples in the validation subset met the requirements for evaluating the model's generalization performance.

[0046] A linear regression model is selected as the regression model to be trained. The training subset is input into the linear regression model. The termination condition for model training is set to the number of training iterations reaching a preset number (the preset number is set according to the number of samples to ensure that the model fully learns the correlation between features and the target). During the training process, a multidimensional color feature vector is used as the model input and the baseline oxidation value is used as the model output target. The model gradually learns the quantitative correlation between each feature component in the multidimensional color feature vector and the baseline oxidation value. The internal correlation weights of the model are continuously optimized through iteration until the training termination condition is met. The training stops and the initial prediction model is obtained. The initial prediction model has the ability to receive multidimensional color feature vectors and output predicted oxidation values.

[0047] The multidimensional color feature vector corresponding to each sample in the validation subset is sequentially input into the initial prediction model. The initial prediction model calculates the multidimensional color feature vector of each input based on the quantitative correlation learned during training and outputs the corresponding prediction result. This prediction result is the model predicted oxidation value. The model predicted oxidation value of each sample in the validation subset is recorded one by one to ensure that the model predicted oxidation value corresponds one-to-one with the sample in the validation subset, without omission or mismatch.

[0048] The mean absolute error (MAE) is selected as the prediction error index. The error between the model-predicted oxidation value and the corresponding baseline oxidation value for each sample in the validation subset is calculated. The calculation method is to subtract the baseline oxidation value of the sample from the model-predicted oxidation value of the individual sample, and take the absolute value of the difference to obtain the absolute error of the individual sample. The absolute errors of all samples in the validation subset are summed to obtain the total absolute error. The total absolute error is divided by the total number of samples in the validation subset to obtain the mean absolute error. This index is used to quantify the degree of deviation between the model prediction result and the actual baseline oxidation value.

[0049] The preset mean absolute error threshold is 0.1 (this threshold is set in conjunction with the accuracy of peroxide value determination of infant formula milk powder to ensure that the model prediction accuracy meets the actual detection requirements). The generalization performance of the initial prediction model is evaluated based on the calculated prediction error index (mean absolute error). The evaluation criteria are as follows: if the mean absolute error is less than or equal to the preset threshold, the model generalization performance is determined to meet the preset requirements; if the mean absolute error is greater than the preset threshold, the model generalization performance is determined to not meet the preset requirements. The generalization performance evaluation results directly reflect the prediction reliability of the initial prediction model for samples not involved in the training.

[0050] If the evaluation result indicates that the generalization performance does not meet the preset requirements, the parameters of the initial prediction model are adjusted according to the deviation direction indicated by the prediction error index. The deviation direction is determined by analyzing the deviation between the model's predicted oxidation value and the baseline oxidation value of the samples in the validation subset. If the model's predicted oxidation value is generally higher than the baseline oxidation value, it indicates that the model weights are overestimated, and the correlation weights of the corresponding feature components need to be reduced. If the model's predicted oxidation value is generally lower than the baseline oxidation value, it indicates that the model weights are underestimated, and the correlation weights of the corresponding feature components need to be increased. After the adjustment, the validation subset is re-input into the model, and the error calculation and performance evaluation steps are repeated until the model's generalization performance meets the preset requirements.

[0051] If the evaluation result shows that the generalization performance meets the preset requirements, there is no need to adjust the model parameters. The initial prediction model is directly determined to be the trained model. This model is the color-oxidation quantitative prediction model for the degree of lipid oxidation and color in infant formula milk powder. This model can accurately output the corresponding oxidation degree prediction result (predicted oxidation value) by inputting the multidimensional color feature vector of infant formula milk powder, thereby realizing the quantitative prediction of the degree of lipid oxidation.

[0052] The scaling factor is derived from the training process of the color-oxidation quantitative prediction model. It uses a training subset as the data basis and aims to minimize the average absolute error between the model's predicted oxidation value and the baseline oxidation value. The final value is determined through iterative adjustments until the model's generalization performance meets the preset requirements.

[0053] The total number of color features in the multidimensional color feature vector is determined by the extraction rules of the multidimensional color feature vector and the number of color features that make up the overall statistical feature component. The overall statistical feature component contains 6 color features, so this parameter is fixed at 6.

[0054] The ordinal number of the color feature in the multidimensional color feature vector is derived from the order identification rule of each color feature in the overall statistical feature components. It is only used to distinguish different color features, and the value ranges from 1 to n according to the fixed arrangement order of the features.

[0055] To match the i-th color feature The corresponding feature weight coefficients are derived from the training process of the color-oxidation quantitative prediction model. Weights are assigned based on the correlation between the i-th color feature F_i in the training subset and the baseline oxidation value. The values ​​are iteratively optimized and adjusted until the model's generalization performance meets the preset requirements. Corresponding to the unique .

[0056] The first component constituting the overall statistical feature component The color feature is derived from the feature extraction process of standardized sample images of infant formula milk powder. It is calculated from the milk powder area image and includes the average value of the brightness component, the average value of the red-green component, the average value of the yellow-blue component, the standard deviation of the brightness component, the standard deviation of the red-green component, and the standard deviation of the yellow-blue component, which are used as the first to sixth color features in a fixed order.

[0057] The weight coefficients corresponding to the local abnormal feature components are derived from the training process of the color-oxidation quantitative prediction model. The weights are assigned based on the correlation between the local abnormal feature components and the baseline oxidation value in the training subset. The values ​​are adjusted through iterative optimization until the model's generalization performance meets the preset requirements.

[0058] The local abnormal feature component is derived from the feature extraction process of standardized sample images of infant formula milk powder. It is composed of spot contrast features alone, which are obtained by dividing the mean of the spot region by the average value of the red and green components of the milk powder region.

[0059] This is a bias term. This parameter comes from the training process of the color-oxidation quantitative prediction model and is used to correct the prediction bias of the model. The optimization objective is to minimize the average absolute error between the model's predicted oxidation value and the baseline oxidation value. The final value is determined through iterative adjustment until the model's generalization performance meets the preset requirements.

[0060] S5. Input the multidimensional color feature vector of the infant formula milk powder to be evaluated into the color-oxidation quantitative prediction model, and output the oxidation index value of the infant formula milk powder to be evaluated.

[0061] In this embodiment of the invention, the step of inputting the multidimensional color feature vector of the infant formula milk powder to be evaluated into the color-oxidation quantitative prediction model and outputting the oxidation index value of the infant formula milk powder to be evaluated includes: Images of the infant formula milk powder to be evaluated were acquired and processed to obtain standardized images for evaluation. Extract the multidimensional color feature vector of the infant formula milk powder to be evaluated from the standardized image to be evaluated; The multidimensional color feature vector to be tested is input into the color-oxidation quantitative prediction model to obtain the oxidation index value of the degree of lipid oxidation of the infant formula milk powder to be evaluated.

[0062] The infant formula milk powder to be evaluated was spread evenly on a clean, colorless, and highly reflective stage, ensuring that the milk powder was spread evenly without lumps or gaps. The operation was carried out under D65 standard light source illumination conditions that were completely consistent with the original image of the induced oxidation sample. The light source was perpendicular to the milk powder surface and the illumination range completely covered the sample without shadows or reflections. A high-definition camera was fixed directly above the milk powder and the lens was kept parallel to the sample surface. The focus was adjusted until the image was clear, and the same fixed shooting parameters as before were used to start shooting and acquire the original image of the infant formula milk powder to be evaluated. Then, the same white balance correction, color space conversion, color threshold segmentation, pixel size scaling, and edge smoothing operations as those used in the preparation of the standardized sample image were performed on the original image. The methods and parameters of each operation were kept consistent with those before. Finally, the standardized image to be evaluated was obtained. The pixel size, color space, and image quality standards of this image were completely consistent with the standardized sample image used for model training. The milk powder region in the standardized image to be evaluated is extracted. Following the same method as extracting the multidimensional color feature vector of the model training samples, the average value of the brightness component, the average value of the red-green component, and the average value of the yellow-blue component in the milk powder region are first calculated to obtain the basic color mean feature. Then, the standard deviation of the brightness component, the standard deviation of the red-green component, and the standard deviation of the yellow-blue component are calculated to obtain the color distribution discrete feature. The basic color mean feature and the color distribution discrete feature are combined in a fixed order to form the overall statistical feature component. Then, based on the red-green component numerical distribution, the same anomaly judgment upper and lower limits and connected component analysis methods are used to identify the suspected oxidation spot region. The spot region mean and spot contrast feature of the spot region are calculated and used as the local anomaly feature component. Finally, the local anomaly feature component and the overall statistical feature component are combined in a fixed order to obtain the test multidimensional color feature vector of the infant formula milk powder to be evaluated. The feature dimension and combination order of this feature vector are completely consistent with the multidimensional color feature vector used for model training. The extracted multidimensional color feature vector of the infant formula to be evaluated is completely input into the trained color-oxidation quantitative prediction model. Based on the quantitative correlation between the multidimensional color feature vector and the baseline oxidation value learned during the training process, the model performs full feature component calculation on the input multidimensional color feature vector. The calculation process follows the model's established quantitative correlation rules and has no parameter adjustment. After the model completes the calculation, it outputs a unique numerical result, which is the oxidation index value of the lipid oxidation degree of the infant formula to be evaluated.

[0063] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0064] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for assessing the degree of lipid oxidation in infant formula milk powder, characterized in that, The method includes: S1. Obtain the baseline oxidation value of infant formula milk powder under different oxidation levels; S2. Perform color space conversion and region standardization processing on the digital image of the infant formula milk powder under controlled lighting conditions to obtain a standardized sample image of the infant formula milk powder; S3. Extract the multidimensional color feature vector of the infant formula milk powder from the standardized sample image; S4. Using the baseline oxidation value as the target variable and the corresponding multidimensional color feature vector as the input feature, establish a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in the infant formula milk powder. S5. Input the multidimensional color feature vector of the infant formula milk powder to be evaluated into the color-oxidation quantitative prediction model, and output the oxidation index value of the infant formula milk powder to be evaluated.

2. The method for assessing the degree of lipid oxidation in infant formula milk powder as described in claim 1, characterized in that, The process of obtaining the baseline oxidation value of infant formula milk powder under different oxidation levels includes: Multiple basic samples of infant formula milk powder with the same raw materials and processes were prepared. The basic samples were stored under different preset environmental conditions to obtain induced oxidation sample sets with different degrees of oxidation. The peroxide value of each sample in the induced oxidation sample set was determined, and the peroxide value was used as the baseline oxidation value of the sample.

3. The method for assessing the degree of lipid oxidation in infant formula milk powder as described in claim 1, characterized in that, The process of color space conversion for the digital image of the infant formula milk powder under controlled lighting conditions includes: Acquire the original image of the infant formula milk powder under standard light source illumination conditions; The original image is subjected to white balance correction to obtain a color-constant image; The color constant image is transformed into a color space to obtain a preliminary transformed image.

4. The method for assessing the degree of lipid oxidation in infant formula milk powder as described in claim 3, characterized in that, After performing color space conversion on the color constant image to obtain a preliminary converted image, the process further includes: The preliminary converted image is segmented by color thresholding to obtain an image of the milk powder region in the preliminary converted image; The image of the milk powder region is scaled to a preset uniform pixel size to obtain a size-normalized image; The size-normalized image is subjected to edge smoothing processing to obtain a standardized sample image of the infant formula milk powder.

5. The method for assessing the degree of lipid oxidation in infant formula milk powder as described in claim 1, characterized in that, The step of extracting the multidimensional color feature vector of the infant formula milk powder from the standardized sample image includes: The average values ​​of the luminance component, red-green component, and yellow-blue component in the milk powder region of the standardized sample image are calculated respectively to obtain the basic color mean characteristics of the infant formula milk powder. Calculate the standard deviation of the luminance component, the standard deviation of the red-green component, and the standard deviation of the yellow-blue component within the milk powder region to obtain the discrete color distribution characteristics of the infant formula milk powder. The basic color mean feature and the color distribution discrete feature are combined to obtain the overall statistical feature components of the infant formula milk powder.

6. The method for assessing the degree of lipid oxidation in infant formula milk powder as described in claim 5, characterized in that, The step of extracting the multidimensional color feature vector of the infant formula milk powder from the standardized sample image further includes: Based on the numerical distribution of red and green components, areas with abnormal color within the milk powder region are identified, thus obtaining suspected oxidation spot areas of the infant formula milk powder. The mean value of the suspected oxidation spot region on the red-green component is denoted as the spot region mean value. The ratio of the mean value of the spot area to the mean value of the overall red-green component in the milk powder area is calculated to obtain the spot contrast characteristics of the infant formula milk powder; The spot contrast features are used as local abnormal feature components of the infant formula milk powder; The local abnormal feature components are combined with the overall statistical feature components to form the multidimensional color feature vector of the infant formula milk powder.

7. The method for assessing the degree of lipid oxidation in infant formula milk powder as described in claim 6, characterized in that, The step of establishing a quantitative prediction model for the degree of lipid oxidation and color in infant formula milk powder, using the baseline oxidation value as the target variable and the corresponding multidimensional color feature vector as the input feature, includes: The baseline oxidation value is paired with the corresponding multidimensional color feature vector to obtain the model training sample set; The training sample set of the model is divided into a training subset and a validation subset; Using the training subset, with the multidimensional color feature vector as input and the baseline oxidation value as output target, a regression model is trained to obtain an initial prediction model. The initial prediction model is validated using the validation subset, and the model parameters are adjusted according to the validation results until the model performance meets the preset requirements, thereby obtaining a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in the infant formula milk powder.

8. The method for assessing the degree of lipid oxidation in infant formula milk powder as described in claim 7, characterized in that, The process of validating the initial prediction model using the validation subset and adjusting the model parameters based on the validation results until the model performance meets the preset requirements, thereby obtaining a color-oxidation quantitative prediction model for the degree of lipid oxidation and color in infant formula milk powder, includes: The multidimensional color feature vectors in the validation subset are input into the initial prediction model to obtain the corresponding model-predicted oxidation values. Calculate the prediction error index between the oxidation value predicted by the model and the corresponding baseline oxidation value in the validation subset; The generalization performance of the initial prediction model is evaluated based on the prediction error index between the oxidation value predicted by the model and the corresponding benchmark oxidation value in the validation subset. If the generalization performance does not meet the preset requirements, the model parameters of the initial prediction model are adjusted according to the direction indicated by the prediction error index. If the generalization performance meets the preset requirements, then the initial prediction model is determined to be a trained color-oxidation quantitative prediction model.

9. The method for assessing the degree of lipid oxidation in infant formula milk powder as described in claim 8, characterized in that, The calculation formula for the color-oxidation quantitative prediction model is as follows: ； In the formula, To predict oxidation values ​​for the model, This is the scaling factor. The total number of color features in the multidimensional color feature vector. Let be the ordinal number of the color feature in the multidimensional color feature vector. In order to be with the first Color features The corresponding feature weight coefficients, The first component constituting the overall statistical feature component A color feature, These are the weighting coefficients corresponding to the local anomaly feature components. For the local anomaly feature components, For bias terms, It is the hyperbolic tangent activation function.

10. The method for assessing the degree of lipid oxidation in infant formula milk powder as described in claim 1, characterized in that, The process of inputting the multidimensional color feature vector of the infant formula milk powder to be evaluated into the color-oxidation quantitative prediction model and outputting the oxidation index value of the infant formula milk powder to be evaluated includes: Images of the infant formula milk powder to be evaluated were acquired and processed to obtain standardized images for evaluation. Extract the multidimensional color feature vector of the infant formula milk powder to be evaluated from the standardized image to be evaluated; The multidimensional color feature vector to be tested is input into the color-oxidation quantitative prediction model to obtain the oxidation index value of the degree of lipid oxidation of the infant formula milk powder to be evaluated.

Images