Fish anti-oxidation state evaluation method and system

Abstract

The invention belongs to the technical field of feed analysis and data processing, and provides a fish antioxidant state evaluation method and system, and the method specifically comprises the steps: firstly carrying out the hyperspectral collection of a fish slice sample, obtaining a spectral image, carrying out the white meat region selection of the spectral image, and dividing the spectral image into a plurality of segmentation regions; then, the content characteristics of the vitamin E in the segmentation areas are obtained, finally, the segmentation areas are graded according to the time sequence changes of the content characteristics, and different samples are subjected to antioxidant state evaluation according to the grading result. An anti-oxidation state grading method for evaluating stability by reflecting an anti-oxidation micromechanism is constructed, the problem that in the prior art, transverse comparison is difficult due to the fact that fish anti-oxidation state evaluation results are affected by tissue structure differences and thickness inconsistency is solved, and the suitability of comparison between samples and the fairness of anti-oxidation ability evaluation are improved; and a rapid, stable and comparable evaluation tool is provided for application scenes such as feed regulation and control, transportation stress judgment and health state monitoring in fish culture.

Description

Technical Field

[0001] The invention belongs to the technical field of feed analysis and spectral data processing, and particularly relates to a method and system for evaluating the antioxidant status of fish. Background Art

[0002] In fish farming, adding antioxidants to feed can effectively reduce oxidative stress and improve fish immunity and growth. However, ensuring that the amount of antioxidants added meets the fish's needs requires assessment of the fish's antioxidant status. Furthermore, in fish transportation and disease prevention, assessing the antioxidant status of fish is often necessary. This antioxidant status refers to the balance of the body's antioxidant system, specifically the relationship between oxidative stress and antioxidant defense. Excessive levels of reactive oxygen species (ROS) produced by metabolism in fish can lead to oxidative damage, including lipid peroxidation, protein oxidation, and DNA damage. Traditional methods for analyzing fish antioxidant status involve measuring tissue ROS content and antioxidant enzyme activity. These methods require rigorous and time-consuming laboratory measurements and are therefore extremely time-sensitive. Leading-edge research laboratories typically use hyperspectral equipment to assess the antioxidant capacity of fish. Specifically, hyperspectral data are collected from fish slices, along with data on oxidative stress products, antioxidant substances, and tissue metabolites, and then inverted to obtain their content. This database is then used to assess the overall antioxidant performance of the sample. However, horizontal assessments between samples suffer from severe misalignment. This misalignment stems from the fact that even when the same part of a fish is rigorously sliced ​​from different fish, significant differences in muscle fiber structure, water content, and lipid content can still exist, directly affecting light reflection and absorption. Furthermore, hyperspectral imaging is sensitive to tissue thickness, and differences in sample thickness can alter the absorption path length and mislead the determination of actual content. This misalignment leads to bias in the hyperspectral inversion of sample content, so direct horizontal comparisons of samples in the database can easily yield results that are inconsistent with the true state. In the feed development process, the antioxidant status of fish is often assessed by averaging multiple slices to reduce the risk of data distortion. However, this approach still fails to effectively establish a standard for horizontal comparisons between individual fish samples. Alternatively, one or a group of samples considered to have good antioxidant levels are selected as a benchmark, and the distance or similarity of other samples to it in the hyperspectral feature space is compared. This approach still ignores the misalignment between samples. This method cleverly uses vitamin C, which is easily oxidized in the air to produce dehydroascorbic acid, as a reference bridge to classify the non-aligned areas in the spectral image, and uses the classification results to invert the substance content of the antioxidant status of the fish body, thereby improving the accuracy of comparison between samples. Summary of the Invention

[0003] The purpose of the present invention is to propose a method and system for evaluating the antioxidant status of fish to solve one or more technical problems existing in the prior art and at least provide a beneficial option or create conditions.

[0004] In order to achieve the above object, according to one aspect of the present invention, a method for evaluating the antioxidant status of fish is provided, the method comprising the following steps: S100, performing hyperspectral acquisition on the fish slice sample to obtain a spectral image; S200, selecting a white meat area of ​​the spectral image and dividing it into a plurality of segmentation areas; S300, obtaining the content characteristics of vitamin E in the segmented area; S400, grading each segmented area according to the temporal change of content characteristics; S500, the antioxidant status of different samples was evaluated through grading results.

[0005] Furthermore, in step S100, the method of performing hyperspectral acquisition on the fish slice sample to obtain a spectral image is: performing hyperspectral image acquisition on the fish slice sample to obtain three-dimensional hyperspectral image data containing several bands, that is, a spectral image, wherein the hyperspectral image data covers the visible light to near-infrared band to reflect the spectral characteristics related to the antioxidant status in the fish tissue.

[0006] Preferably, fish slices are selected from the abdominal or dorsal muscle tissue of the fish body to make samples, spread on a non-fluorescent platform with a diffuse reflection background, and collected by line scanning or area array using a hyperspectral imaging device. The thickness of the fish slice samples ranges from 1mm to 3mm. After the thickness is determined, the thickness of the remaining samples should be ensured to be the same to ensure the consistency of the spectral absorption path.

[0007] The imaging device used for spectral image acquisition has at least 100 continuous wavelengths and scans across the wavelength range of 400nm-1000nm, from visible light to the near-infrared. Before spectral image acquisition, the fish slices must be pre-conditioned at a constant temperature of 20°C to 25°C for at least 20 minutes to minimize external environmental interference.

[0008] Furthermore, in step S200, the method of selecting the white meat region of the spectral image and dividing it into a plurality of segmented regions is as follows: constructing a pseudo-color image from a plurality of key bands in the spectral image, and extracting the white meat region of the fish through the pseudo-color image; Preferably, the contour detection operator is the findContours function in OpenCV; The white meat area is divided into several segments based on the similarity of spectral features, so that each segment corresponds to a local area in the fish tissue with similar spectral response characteristics.

[0009] The SLIC superpixel segmentation algorithm usually uses Euclidean distance for similarity calculation, but it should be replaced here with SAM spectral angular distance. This is because Euclidean distance assumes that the spectral space is linearly distributed, is insensitive to the direction of high-dimensional spectral curves, and is easily affected by light intensity or sample thickness. In addition, key differences in hyperspectral images are often reflected in spectral direction rather than amplitude. Using the complete similarity vector instead of the neighborhood average angular distance is to avoid the problem of the mean method ignoring local microstructure, because it may mask the contribution of key pixels to heterogeneous regions, leading to misclustering or weakening of principal component features. The similarity result ignores the principal component layers with less contribution. Retaining the complete similarity vector as the similarity measurement feature can preserve local diversity and microscopic feature differences, improving segmentation sharpness and spectral direction accuracy.

[0010] Furthermore, in step S300, the method for obtaining the content characteristics of vitamin E in the segmented area is: for any segmented area, the pixel spectral curve in the segmented area is averaged or median processed to form the spectral curve of the segmented area, and the vitamin E content of the spectral curve is obtained through an inversion algorithm, which is recorded as the content characteristic.

[0011] The specific steps are to extract the hyperspectral reflectance curves of all pixels in the segmented area, and obtain a representative spectral curve representing the overall spectral characteristics of the segmented area by averaging or median processing the reflectance of each pixel in each band, and input the representative spectral curve into a pre-trained inversion model to obtain the predicted value of the vitamin E content in the segmented area; the pre-trained inversion model is a commonly used technology in hyperspectral content analysis and will not be repeated here.

[0012] The rationale for using vitamin E as a time-series observation substance is as follows: First, vitamin E is a significant fat-soluble non-enzymatic antioxidant in fish and a common additive in fish feed. It can directly reflect the antioxidant defense level of fish, and is therefore often used as a hyperspectral indicator for determining antioxidant activity in fish. Compared to other non-antioxidant indicator substances, vitamin E does not require additional modeling and has the advantage of hyperspectral system integration. Secondly, when vitamin E is exposed to air or a gas environment with a specific oxygen concentration, the content change process has significant time correlation and regional differences, and can serve as a bridge variable for further dynamic classification. Because its distribution uniformity is stable compared to other antioxidants, under this premise, it can also show a controllable rather than drastic slow decline trend. Therefore, it has the advantage of being a response variable for constructing time series modeling.

[0013] Furthermore, in step S400, the method for grading each segmented area according to the temporal variation of the content characteristics is: Assume a time period as the monitoring period Dpet, Dpet∈[5,10] hours, where the monitoring period is terminated when the content characteristic changes tend to be static. During the monitoring period, the content characteristics of the segmented area are obtained every 20 minutes, and the time scale of obtaining the content characteristics is recorded as the measurement point; In one embodiment, salmon samples were prepared, and slices were taken from the white meat area of ​​the fish's back. The slices were 2 mm thick and maintained at a constant temperature of 20 ± 1°C using a thermostat. The monitoring period was set to 6 hours, at which point the characteristic change in vitamin E content tended to be insignificant. During the monitoring period, the content characteristics of each segmented area constitute a characteristic sequence Ft.Ls; each Ft.Ls is smoothed by Savitzky-Golay filtering; the window size is set to 5 and the polynomial order is set to 2; the Savitzky-Golay filter is a digital filter based on local polynomial regression. The main reason for using the Savitzky-Golay filter is that it can reduce noise while maintaining the high-order moments of the signal, which means that the peaks, valleys and other characteristics of the signal can be better maintained.

[0014] At both ends of Ft.Ls, if polynomial fitting cannot be performed because the window cannot completely cover the five content features, the original data at both ends of Ft.Ls are retained as fitting values.

[0015] The regression fitting values ​​of Ft.Ls in the linear regression model and the piecewise linear model were calculated by the BIC Bayesian Information Criterion method: the calculation process of the corresponding regression fitting value in the linear regression model is as follows: for any segment, assuming that the corresponding Ft.Ls conforms to the linear regression model, the least squares method is used to construct the fitting model and calculate the residual sum of squares, which is recorded as ARs; The regression fitting value obtained by inputting the residual sum of squares into the BIC method is recorded as a single estimate Fio.bic; The calculation process of the regression fitting value corresponding to the piecewise linear model is as follows: continuously assume that there is an inflection point in the segmented area, and use the least squares method to fit two linear regression models before and after the inflection point, namely the piecewise linear model; the number of piecewise linear models is the number of inflection points plus 1; In Ft.Ls, the measurement points where the data follows different linear relationships before and after a measurement point are recorded as inflection points; each candidate inflection point HTC may become an inflection point TC; the candidate inflection points are all measurement points except the corresponding measurement points at the two ends of Ft.Ls, that is, the measurement points corresponding to the beginning and end will not become candidate inflection points; Based on the Bayesian information criterion, the regression fitting values of each piecewise linear model are calculated using the BIC method, denoted as the piecewise estimate Sio.bic; The Bayesian information criterion BIC aims to select the optimal model among candidate models. By the different rates of vitamin E consumption, it can reveal whether the antioxidant capacity of fish muscle tissue can maintain a stable physiological state. The obtained oxidation loss rate and oxidation grading sequence can be used to reflect the dynamic changes or differences in the antioxidant defense systems within different tissues of the same sample, providing a standardized and quantitative analysis tool for horizontal comparison between multiple samples. It can not only enhance the comparability of evaluation results between different tissue types but also reduce the misalignment effect caused by natural differences between samples. Therefore, the Bayesian information criterion is used to determine whether there are significant inflection points in the segmentation area; For any segmentation area, if min(Sio.bic(HTC i )) < Fio.bic, then the piecewise linear model is accepted; otherwise, there is no inflection point in this segmentation area, and the linear regression model is accepted; where min() is the minimum value function; By calculating the regression fitting values corresponding to each candidate inflection point and selecting the model corresponding to the minimum value as the optimal model to determine whether there are statistically significant content change nodes in this segmentation area. When the piecewise linear model is better than the overall linear model, it indicates that there may be stagewise oxidation rate changes in this area during the exposure process, such as accelerated oxidation responses caused by local antioxidant depletion, tissue structure exposure, or enhanced microzone ventilation. The appearance of such inflection points can reflect the turning structure of the antioxidant capacity within the tissue over time, thus serving as a signaling marker of antioxidant status imbalance. This method not only improves the fitting effect of regional modeling but also depicts the complexity of the physiological oxidation process through piecewise dynamic structures, which helps to identify potential oxidation hypersensitive regions and support the precise classification of different tissue states.

[0016] Calculate the absolute value of the slope of the corresponding model for each segmentation area and denote it as the oxidation loss rate; among them, for the piecewise linear model, its oxidation loss rate is the maximum value of the absolute values of the slopes of the front and rear segments; Construct an oxidation grading sequence from the oxidation loss rates of each segmentation area. In the oxidation grading sequence, the oxidation loss rates are evenly divided into K grades according to the MinMax method; Where K is the number of levels, and the oxidation loss rate is evenly divided into K equal-width intervals with an interval width of (max.Ox-min.Ox) / K, where max.Ox is the maximum oxidation loss rate in the oxidation graded sequence and min.Ox is the minimum oxidation loss rate in the oxidation graded sequence. Then, each oxidation loss rate is assigned a level according to the interval threshold. The interval is closed on the left and open on the right, that is, it includes the left endpoint but not the right endpoint. However, the right endpoint of the Kth interval includes the maximum oxidation loss rate to cover the entire range. All partitions are sorted according to their slope values ​​and divided into K levels according to the principle of equal quantity. Partitions with smaller absolute slope values ​​are assigned to higher levels, indicating that they have stronger antioxidant capacity, and partitions with larger absolute slope values ​​are assigned to lower levels, indicating that they have weaker antioxidant capacity.

[0017] The absolute value of the slope output by the model in each segment was defined as the oxidative loss rate, and a regional grading sequence was constructed based on this. During the grading process, equal-width intervals were used to rank the antioxidant capacity. A smaller oxidative loss rate indicates slower vitamin E consumption per unit time in the tissue, and the antioxidant system is more likely to be in a steady-state or well-regulated state. A larger oxidative loss rate may indicate that the tissue region rapidly enters a state of oxidative stress under exposure conditions, leading to accelerated antioxidant depletion or limited tissue defense mechanisms. In particular, the co-occurrence of an inflection point and a high-slope segment can be further interpreted as a state of critical antioxidant collapse. This rate-grading approach effectively enhances the interpretability of dynamic trends, further incorporating the regional characteristics of spectral images to construct a temporally and spatially resolved antioxidant function map, and provides a more biologically meaningful reference for tissue activity assessment.

[0018] However, in practical applications, the aforementioned slope-based rate-of-change classification method can quickly reflect the linear attenuation trend of vitamin E content within a set time interval and is suitable for segmented areas where antioxidant status changes more regularly and has a clear trend. However, while ignoring the total amount of cumulative vitamin E loss during the entire exposure process, while it can capture the impact of linear fluctuations, it is often not adaptable to tissue areas with drastic local changes or complex antioxidant release processes. Therefore, the present invention proposes a more preferred solution as follows: Preferably, in step S400, the method for grading each segmented area according to the temporal variation of the content characteristics is as follows: set a time period as a monitoring period TETH, TETH∈[5,10] hours, and obtain the content characteristics every 10-20 minutes during the monitoring period; the content characteristics of any segmented area during the monitoring period constitute a content time series, and the content time series is subjected to Min-Max normalization to obtain a normalized time series, wherein the normalized time series x_norm after Min-Max normalization is obtained by the formula Transformation, x is any value in the content time series, min(x) is the minimum value of the content time series, and max(x) is the maximum value of the content time series. The purpose is to eliminate the asynchronous decay differences of the content time series of different fish body segments caused by different oxidation rates, and then compare and grade the segments of different fish bodies; For any segment, the absolute value of the difference between the content characteristics at each moment in the normalized time series and the content characteristics at the first moment in the reverse time direction is taken as the local sinking amount Losua, the median of all local sinking amounts is recorded as the local sinking median, and the local sinking amounts greater than the local sinking median are accumulated to form the change point sinking cumulative value Dpcot of the segment. This variable is used to measure the total loss level of antioxidants in the segment during the stage of significant content change; The calculation principle of the cumulative change-point sinking value is to effectively eliminate the interference of noise during the stable period through median filtering, so that the rapid consumption period of the oxidation process can be more accurately reflected. The calculation of the local sinking amount is based on the natural law that vitamin E exhibits a staged attenuation during the oxidation process. The transition from the rapid consumption period to the stable period is the core feature of its antioxidant dynamics. This method can accurately track the consumption of antioxidant substances. For any segmented area, the normalized time series is decomposed at multiple scales by wavelet transform, where wavelet transform is called using the wavedec function of the PyWavelets library in Python. The high-frequency detail coefficients and low-frequency approximation coefficients are separated, and the coefficients are further reconstructed to obtain the high-frequency energy and low-frequency energy at each moment, which are respectively recorded as the high-frequency content feature Hfrct and the low-frequency content feature Lfrct. Reconstruction here refers to the inverse transformation process: the high-frequency detail coefficients and the low-frequency approximation coefficients are re-synthesized into a time series by inverse transformation, so as to obtain the high-frequency energy corresponding to each moment in the time series synthesized by the high-frequency detail coefficients, and the low-frequency energy corresponding to each moment in the time series synthesized by the low-frequency approximation coefficients. The cumulative subsidence trend under the contribution of high and low frequency components is calculated based on the high-frequency content features, low-frequency content features and local subsidence, which is recorded as the denoised cumulative subsidence area Decus:

[0019] Where t is the cumulative variable, T is the number of all monitoring moments in the monitoring period, the local subsidence cannot be calculated at the initial monitoring moment, so the denoised cumulative subsidence area is calculated from the second monitoring moment, Hfrct t and Lfrct t They are the high-frequency content features and low-frequency content features corresponding to the t-th moment, Losua t The local subsidence corresponding to the t-th moment, max() is the maximum value function; Hyperspectral noise and sample thickness differences can easily introduce high-frequency interference. Therefore, multi-scale decomposition of the time series using wavelet transform can separate the true antioxidant signal from the noise. The robustness and credibility of the classification can be improved by constructing an interference-resistant denoising cumulative sinking area index. The comprehensive subsidence classification value Cprsu is calculated based on the cumulative subsidence value of the change point and the denoised cumulative subsidence area: ; Calculate the lower quartile and upper quartile of the cumulative value of the change point sinking of all the segmented areas. If the cumulative value of the change point sinking of a segmented area is greater than its upper quartile, the weight wta=0.7 is taken; if it is between the lower quartile and the upper quartile, the weight wta=0.5 is taken; if the cumulative value of the change point sinking is less than its lower quartile, the weight wta=0.3 is taken; Quantiles are insensitive to extreme values ​​and are suitable for the non-normal distribution commonly found in antioxidant biological data. Based on the quantiles of the cumulative change-point sink values ​​in the segmented regions, the system automatically divides the segmented regions into high, medium, and low antioxidant loss regions, avoiding the subjective bias of manually set thresholds. Change-point sink values ​​greater than the upper quartile represent regions of high antioxidant loss, directly reflecting loss during the significant oxidation phase, where there is less noise interference. High-frequency energy in the denoised cumulative sink area has been suppressed. Fish samples are under acute oxidative stress, and loss during the rapid consumption phase is key to assessment. Therefore, the change-point sink value should be prioritized and assigned a higher weight. Values ​​between the lower and upper quartiles reflect normal fluctuations, and the complementarity of the change-point sink value and the denoised cumulative sink area should be considered and assigned equal weight. Change-point sink values ​​less than the lower quartile represent slow antioxidant depletion in healthy fish samples. Change-point sink values ​​are affected by stationary noise, while the denoised cumulative sink area is more reliable through multi-scale denoising and should be given a higher weight. The partition with the smallest comprehensive subsidence grade is selected as the high-level benchmark sample. The difference between the comprehensive subsidence grade of any partition and the high-level benchmark sample is calculated, and the ratio of the comprehensive subsidence grade of the partition is recorded as the subsidence relative difference ratio. The subsidence relative difference ratios of all partitions are clustered using density peak clustering to obtain K grades. The density peak clustering is implemented in Python by calling the dpc function of the pyclustering library. K is a preset number, with a default value of 10. Assigning a higher level to a cluster of partitions with relatively small differences in sinking indicates that they have little difference from the high-level benchmark sample and their antioxidant capacity is well maintained; assigning a lower level to a cluster of partitions with relatively large differences in sinking indicates that they have a large difference from the high-level benchmark sample and their antioxidant capacity is significantly attenuated; The segment corresponding to the minimum comprehensive sinking grade value represents the area with the least antioxidant loss, reflecting that the antioxidant system in this area is in optimal condition. Such areas correspond to healthy tissue that is not subject to oxidative stress and are ideal reference benchmarks. The objective selection based on data avoids human misjudgment. The influence of absolute value dimensions is further eliminated through ratios, making the results of different segments comparable. At the same time, because antioxidant capacity presents continuous or non-uniform grading, density peak clustering adaptively determines the number of clusters, avoiding the limitation of K-means clustering that requires pre-specified K values, thus achieving highly robust antioxidant status classification. Beneficial effects: The slope method focuses on rate identification, reflecting the linear attenuation trend of vitamin E within a set time interval. It is suitable for classifying segmented areas with relatively regular changes in antioxidant status and obvious trends, or for classifying tissue areas that reflect slow changes in oxidation reactions. The area method focuses on the evaluation of the total amount of cumulative loss. Especially in the research process involving the release of antioxidants from feed types, it can better deal with local drastic changes or complex antioxidant consumption processes. It reflects the actual loss effect of vitamin E by capturing the rapid consumption period of the oxidation reaction, thereby classifying the sample tissue areas. Therefore, the slope method is suitable for fish samples with small differences in content obtained by spectral inversion between samples, while the area method is suitable for fish samples with obvious oxidation peaks or areas of irregular consumption.

[0020] Furthermore, in step S500, the method for evaluating the antioxidant status of different samples based on the grading results is as follows: the number of segmented areas corresponding to different grades is recorded as the total grade, and the total grades are arranged in descending order to select the top 5 as representative grades; the intersection of the representative grades of the current sample and the standard sample is used as the comparative grade, and all the segmented areas corresponding to the comparative grades are respectively intercepted from the original spectral images of the current sample and the standard sample as evaluation areas, and the antioxidant status of fish in the two evaluation areas is evaluated by hyperspectral.

[0021] The current sample refers to the sample that needs to be evaluated, or the sample whose antioxidant status is not labeled and is read from the database. The standard sample is a benchmark sample whose antioxidant status is labeled and already exists in the database.

[0022] The usual method for evaluating the antioxidant status of fish in two evaluation areas using hyperspectral analysis is to extract spectral data from each evaluation area and invert antioxidant indices. The core step is to input the spectral data into a previously trained inversion model to output predicted values ​​of the content of antioxidant status-related substances of interest, including MDA, GSH, CoQ10, etc., to form a set of characteristic vectors of the antioxidant status of the area. The various contents in the set are accumulated using the PCA weighted method to form multiple antioxidant status indices. The higher the index value, the better the antioxidant status.

[0023] The grades of different segments in the spectral graph represent different stages of antioxidant loss, which are physiologically progressive. Therefore, performing hyperspectral inversion between segments of the same grade can effectively control the relative constancy of interference variables such as tissue type and oxidation stage, thereby ensuring that the evaluation system has a dynamic spatial classification with physiological consistency to improve the scientificity and accuracy of further comparative evaluation.

[0024] Preferably, all undefined variables in the present invention, if not clearly defined, can be manually set thresholds.

[0025] The present invention also provides a fish antioxidant status assessment system, the fish antioxidant status assessment system comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the fish antioxidant status assessment method when executing the computer program. The fish antioxidant status assessment system can be run on computing devices such as desktop computers, laptop computers, PDAs, and cloud data centers. The executable system may include, but is not limited to, a processor, a memory, and a server cluster. The processor executes the computer program in the following system units: A spectral sample acquisition unit, used for performing hyperspectral acquisition on fish slice samples to obtain spectral images; A region division unit is used to select a white meat region of the spectral image and divide it into a plurality of segmentation regions; Content feature inversion unit, used to obtain the content features of vitamin E in the segmented area; A regional classification unit is used to classify each segmented area according to the temporal changes of content characteristics; The antioxidant assessment unit is used to evaluate the antioxidant status of different samples through grading results.

[0026] The present invention provides a method and system for assessing the antioxidant status of fish. By inverting the content of vitamin E in fish tissue using hyperspectral imagery and combining it with the time-series variation observed during oxidation, the system constructs an antioxidant status grading method that not only reflects the microscopic mechanisms of antioxidant activity but also provides robust assessment stability. Taking advantage of the fact that vitamin E, a typical fat-soluble antioxidant, exhibits good spatial uniformity and spectral response consistency during the initial stages of oxidative stress in fish, and its slow decay process in tissue, it effectively avoids interference from factors such as thickness and tissue density on hyperspectral inversion. Two strategies, the slope method and the area method, are proposed. The slope method reflects the rhythm of antioxidant loss during the stable decay process by modeling the linear rate of change of content and is suitable for identifying samples with continuous and stable tissue oxidation status. The area method, on the other hand, reveals total antioxidant loss through integral analysis of the local rapid decline phase and is suitable for scenarios with significant oxidation reaction fluctuations or complex feed release mechanisms. Both strategies can be flexibly applied based on sample characteristics, forming a multidimensional assessment framework for temporal antioxidant behavior, thereby improving the accuracy and physiological interpretability of antioxidant capacity comparisons between different samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The above and other features of the present invention will become more apparent through a detailed description of the embodiments shown in conjunction with the accompanying drawings. The same reference numerals in the drawings of the present invention represent the same or similar elements. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can derive other drawings based on these drawings without inventive work. In the drawings: Figure 1 Shown is a flow chart of a method for assessing the antioxidant status of fish; Figure 2 Shown is a structural diagram of a fish antioxidant status assessment system. DETAILED DESCRIPTION

[0028] The following will be combined with the embodiments and drawings to clearly and completely describe the concept, specific structure and technical effects of the present invention so as to fully understand the purpose, scheme and effect of the present invention. It should be noted that the embodiments and features in the embodiments of this application can be combined with each other unless there is a conflict.

[0029] like Figure 1 The figure shows a flow chart of a method for evaluating the antioxidant status of fish. Figure 1 A method for evaluating the antioxidant status of fish according to an embodiment of the present invention is described below. The method comprises the following steps: S100, performing hyperspectral acquisition on the fish slice sample to obtain a spectral image; S200, selecting a white meat area of ​​the spectral image and dividing it into a plurality of segmentation areas; S300, obtaining the content characteristics of vitamin E in the segmented area; S400, grading each segmented area according to the temporal change of content characteristics; S500, the antioxidant status of different samples was evaluated through grading results.

[0030] Furthermore, in step S100, the method of performing hyperspectral acquisition on the fish slice sample to obtain a spectral image is: performing hyperspectral image acquisition on the fish slice sample to obtain three-dimensional hyperspectral image data containing several bands, that is, a spectral image, wherein the hyperspectral image data covers the visible light to near-infrared band to reflect the spectral characteristics related to the antioxidant status in the fish tissue.

[0031] Preferably, fish slices are selected from the abdominal or dorsal muscle tissue of the fish body to make samples, spread on a non-fluorescent platform with a diffuse reflection background, and collected by line scanning or area array using a hyperspectral imaging device. The thickness of the fish slice samples ranges from 1mm to 3mm. After the thickness is determined, the thickness of the remaining samples should be ensured to be the same to ensure the consistency of the spectral absorption path.

[0032] The imaging device used for spectral image acquisition has at least 100 continuous wavelengths and scans across the wavelength range of 400nm-1000nm, from visible light to the near-infrared. Before spectral image acquisition, the fish slices must be pre-conditioned at a constant temperature of 20°C to 25°C for at least 20 minutes to minimize external environmental interference.

[0033] Furthermore, in step S200, the method of selecting the white meat region of the spectral image and dividing it into a plurality of segmented regions is as follows: constructing a pseudo-color image from a plurality of key bands in the spectral image, and extracting the white meat region of the fish through the pseudo-color image; The specific steps are to perform principal component analysis on the hyperspectral image and extract the first three principal component layers. The corresponding bands of the principal component layers include 900nm, 650nm and 540nm, representing the water absorption band, fish meat absorption band and myoglobin absorption band respectively; construct a pseudo-color image by selecting bands through each principal component layer; grayscale the pseudo-color image and extract the foreground area with Otsu, identify the connected areas in the foreground area through the contour detection operator, and exclude the connected areas that do not conform to the characteristics of white meat. The excluded areas specifically include noise, blood vessels and edge fat. The obtained foreground area is used as the final white meat area.

[0034] Preferably, the contour detection operator is the findContours function in OpenCV; The white meat area is divided into several segments based on the similarity of spectral features, so that each segment corresponds to a local area in the fish tissue with similar spectral response characteristics.

[0035] The similarity of spectral features is calculated using SAM spectral angular distance. The specific process includes: for each pixel, the principal component score vector is directly obtained through principal component analysis. The 8-neighborhood or 24-neighborhood of any pixel is used as its comparison pixel. The sequence of SAM spectral angular distances between the pixel and each comparison pixel is calculated as the similarity vector. The similarity vector is used as the clustering distance term of the SLIC superpixel segmentation algorithm to perform regional segmentation on the white meat area to obtain each segmentation domain. The clustering distance term of the superpixel segmentation algorithm is expressed as dis_spec(p i1 , p i2 )=1-cos(v i1 ,v i2 ), p i1 and p i2 Represents two pixels numbered i1 and i2, v i1 and v i2 Represents the similarity vector corresponding to the two pixels with serial numbers i1 and i2.

[0036] Furthermore, in step S300, the method for obtaining the content characteristics of vitamin E in the segmented area is: for any segmented area, the pixel spectral curve in the segmented area is averaged or median processed to form the spectral curve of the segmented area, and the vitamin E content of the spectral curve is obtained through an inversion algorithm, which is recorded as the content characteristic.

[0037] The specific steps are to extract the hyperspectral reflectance curves of all pixels in the segmented area, and obtain a representative spectral curve representing the overall spectral characteristics of the segmented area by averaging or median processing the reflectance of each pixel in each band, and input the representative spectral curve into a pre-trained inversion model to obtain the predicted value of the vitamin E content in the segmented area; the pre-trained inversion model is a commonly used technology in hyperspectral content analysis and will not be repeated here.

[0038] Furthermore, in step S400, the method for grading each segmented area according to the temporal variation of the content characteristics is: Assume a time period as the monitoring period Dpet, Dpet∈[5,10] hours, where the monitoring period is terminated when the content characteristic changes tend to be static. During the monitoring period, the content characteristics of the segmented area are obtained every 20 minutes, and the time scale of obtaining the content characteristics is recorded as the measurement point; In one embodiment, salmon samples were prepared, and slices were taken from the white meat area of ​​the fish's back. The slices were 2 mm thick and maintained at a constant temperature of 20 ± 1°C using a thermostat. The monitoring period was set to 6 hours, at which point the characteristic change in vitamin E content tended to be insignificant. During the monitoring period, the content characteristics of each segmented area constitute a characteristic sequence Ft.Ls; each Ft.Ls is smoothed by Savitzky-Golay filtering, where the window size is set to 5 and the polynomial order is set to 2; At both ends of Ft.Ls, if polynomial fitting cannot be performed because the window cannot completely cover the five content features, the original data at both ends of Ft.Ls are retained as fitting values.

[0039] The regression fitting values ​​of Ft.Ls in the linear regression model and the piecewise linear model were calculated by the BIC Bayesian Information Criterion method: the calculation process of the corresponding regression fitting value in the linear regression model is as follows: for any segment, assuming that the corresponding Ft.Ls conforms to the linear regression model, the least squares method is used to construct the fitting model and calculate the residual sum of squares, which is recorded as ARs; The regression fitting value is obtained by inputting the residual sum of squares into the BIC method, which is recorded as a single estimate Fio.bic; The formula for the single estimate is Fio.bic = Snp × ln(ARs / Snp) + 2ln(Snp); where Snp is the number of elements in Ft.Ls and ln() is the logarithmic function with base e. The BIC Bayesian Information Criterion method is abbreviated as the BIC method. The calculation process of the regression fitting value corresponding to the piecewise linear model is as follows: continuously assume that there is an inflection point in the segmented area, and use the least squares method to fit two linear regression models before and after the inflection point, namely the piecewise linear model; the number of piecewise linear models is the number of inflection points plus 1; In Ft.Ls, the measurement points where the data follows different linear relationships before and after a measurement point are recorded as inflection points; each candidate inflection point HTC may become an inflection point TC; the candidate inflection points are all measurement points except the corresponding measurement points at the two ends of Ft.Ls, that is, the measurement points corresponding to the beginning and end will not become candidate inflection points; Based on the Bayesian Information Criterion, the BIC method was used to calculate the regression fitting value of each piecewise linear model, which was recorded as the piecewise estimate Sio.bic; The formula for segmented estimation is Sio.bic(HTC i )=Snp×ln(ARs(HTC i ) / Snp)+4ln(Snp); where i is the number of candidate inflection points in Ft.Ls, Snp is the number of elements in Ft.Ls, Sio.bic(HTCi ) and ARs (HTC i ) are the regression fitting value and the sum of squared residuals of the piecewise linear model when the $i$-th candidate inflection point becomes an inflection point, respectively. $\ln()$ is the natural logarithm function with base $e$; For any partition area, if $\min(S_{io.bic}(HTC i )) < F_{io.bic}$, the piecewise linear model is accepted; otherwise, there is no inflection point in this partition area, and the linear regression model is accepted; where $\min()$ is the minimum value function; Calculate the absolute value of the slope of the corresponding model for each partition area and denote it as the oxidation loss rate. For the piecewise linear model, its oxidation loss rate is the maximum of the absolute values of the slopes of the two adjacent segments; Construct an oxidation classification sequence from the oxidation loss rates of each partition area. In the oxidation classification sequence, the oxidation loss rates are evenly divided into $K$ grades according to the MinMax method; where $K$ is the number of grades. The oxidation loss rates are evenly divided into $K$ equal-width intervals, and the interval width is $(max.Ox - min.Ox) / K$, where $max.Ox$ is the maximum value of the oxidation loss rates in the oxidation classification sequence, and $min.Ox$ is the minimum value of the oxidation loss rates in the oxidation classification sequence. Then, assign grades to each oxidation loss rate according to the interval thresholds. The intervals are left-closed and right-open, that is, the left endpoint is included but the right endpoint is not included, except that the right endpoint of the $K$-th interval includes the maximum value of the oxidation loss rates to cover the entire range; Sort according to the slope values of all partition areas, and divide the partition areas into $K$ grades according to the equal-number principle. Among them, the partition areas with smaller absolute values of the slopes are assigned to higher grades, indicating stronger antioxidant ability, and the partition areas with larger absolute values of the slopes are assigned to lower grades, indicating weaker antioxidant ability.

[0040] Preferably, in step S4**00**, the method for grading each partition area according to the temporal variation of the content characteristics is as follows: Set a time period as the monitoring period $T_{ETH}$, $T_{ETH} \in [5, 10]$ hours, and obtain the content characteristics every 10 - 20 minutes within the monitoring period. The content characteristics of any partition area within the monitoring period form a content time series. Perform Min - Max normalization on the content time series to obtain a normalized time series. The normalized time series $x_{norm}$ after Min - Max normalization is transformed by the formula , where $x$ is any value in the content time series, $\min(x)$ is the minimum value of the content time series, and $\max(x)$ is the maximum value of the content time series. The purpose is to eliminate the asynchronous decay differences caused by different oxidation rates of the content time series of different fish body partition areas, and then compare and grade the partition areas of different fish bodies; For any segment, the absolute value of the difference between the content characteristics at each moment in the normalized time series and the content characteristics at the first moment in the reverse time direction is taken as the local sinking amount Losua, the median of all local sinking amounts is recorded as the local sinking median, and the local sinking amounts greater than the local sinking median are accumulated to form the change point sinking cumulative value Dpcot of the segment. This variable is used to measure the total loss level of antioxidants in the segment during the stage of significant content change; For any segmented area, the normalized time series is decomposed at multiple scales by wavelet transform, where wavelet transform is called using the wavedec function of the PyWavelets library in Python. The high-frequency detail coefficients and low-frequency approximation coefficients are separated, and the coefficients are further reconstructed to obtain the high-frequency energy and low-frequency energy at each moment, which are respectively recorded as the high-frequency content feature Hfrct and the low-frequency content feature Lfrct. Reconstruction here refers to the inverse transformation process: the high-frequency detail coefficients and the low-frequency approximation coefficients are re-synthesized into a time series by inverse transformation, so as to obtain the high-frequency energy corresponding to each moment in the time series synthesized by the high-frequency detail coefficients, and the low-frequency energy corresponding to each moment in the time series synthesized by the low-frequency approximation coefficients. The cumulative subsidence trend under the contribution of high and low frequency components is calculated based on the high-frequency content features, low-frequency content features and local subsidence, which is recorded as the denoised cumulative subsidence area Decus:

[0041] Where t is the cumulative variable, T is the number of all monitoring moments in the monitoring period, the local subsidence cannot be calculated at the initial monitoring moment, so the denoised cumulative subsidence area is calculated from the second monitoring moment, Hfrct t and Lfrct t They are the high-frequency content features and low-frequency content features corresponding to the t-th moment, Losua t The local subsidence corresponding to the t-th moment, max() is the maximum value function; The comprehensive subsidence classification value Cprsu is calculated based on the cumulative subsidence value of the change point and the denoised cumulative subsidence area: ; Calculate the lower quartile and upper quartile of the cumulative value of the change point sinking of all the segmented areas. If the cumulative value of the change point sinking of a segmented area is greater than its upper quartile, the weight wta=0.7 is taken; if it is between the lower quartile and the upper quartile, the weight wta=0.5 is taken; if the cumulative value of the change point sinking is less than its lower quartile, the weight wta=0.3 is taken; The partition with the smallest comprehensive subsidence grade is selected as the high-level benchmark sample. The difference between the comprehensive subsidence grade of any partition and the high-level benchmark sample is calculated, and the ratio of the comprehensive subsidence grade of the partition is recorded as the subsidence relative difference ratio. The subsidence relative difference ratios of all partitions are clustered using density peak clustering to obtain K grades. The density peak clustering is implemented in Python by calling the dpc function of the pyclustering library. K is a preset number, with a default value of 10. Assigning a higher level to a cluster of partitions with relatively small differences in sinking indicates that they have little difference from the high-level benchmark sample and their antioxidant capacity is well maintained; assigning a lower level to a cluster of partitions with relatively large differences in sinking indicates that they have a large difference from the high-level benchmark sample and their antioxidant capacity is significantly attenuated; Furthermore, in step S500, the method for evaluating the antioxidant status of different samples based on the grading results is as follows: the number of segmented areas corresponding to different grades is recorded as the total grade, and the total grades are arranged in descending order to select the top 5 as representative grades; the intersection of the representative grades of the current sample and the standard sample is used as the comparative grade, and all the segmented areas corresponding to the comparative grades are respectively intercepted from the original spectral images of the current sample and the standard sample as evaluation areas, and the antioxidant status of fish in the two evaluation areas is evaluated by hyperspectral.

[0042] The usual method for evaluating the antioxidant status of fish in two evaluation areas using hyperspectral analysis is to extract spectral data from each evaluation area and invert antioxidant indices. The core step is to input the spectral data into a previously trained inversion model to output predicted values ​​of the content of antioxidant status-related substances of interest, including MDA, GSH, CoQ10, etc., to form a set of characteristic vectors of the antioxidant status of the area. The various contents in the set are accumulated using the PCA weighted method to form multiple antioxidant status indices. The higher the index value, the better the antioxidant status.

[0043] The grades of different segments in the spectral graph represent different stages of antioxidant loss, which are physiologically progressive. Therefore, performing hyperspectral inversion between segments of the same grade can effectively control the relative constancy of interference variables such as tissue type and oxidation stage, thereby ensuring that the evaluation system has a dynamic spatial classification with physiological consistency to improve the scientificity and accuracy of further comparative evaluation.

[0044] The embodiment of the present invention provides a fish antioxidant status assessment system, such as Figure 2 Shown is a structural diagram of a fish antioxidant status assessment system of the present invention. A fish antioxidant status assessment system of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned fish antioxidant status assessment method embodiment are implemented.

[0045] The system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to run in the following units of the system: A data acquisition unit, used for performing hyperspectral acquisition on fish slice samples to obtain spectral images; A dynamic merging unit is used to select the white meat area of ​​the spectral image and divide it into several segmentation areas; A buffer pool construction unit is used to obtain the content characteristics of vitamin E in the segmented area; The data transmission unit is used to classify each segmented area according to the temporal change of the content characteristics.

[0046] The fish antioxidant status assessment system can be run on computing devices such as desktop computers, laptops, PDAs, and cloud servers. The system can include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the example is merely an example of a fish antioxidant status assessment system and does not constitute a limitation on the fish antioxidant status assessment system. The system can include more or fewer components than the example, or a combination of certain components, or different components. For example, the system can also include input and output devices, network access devices, buses, and the like.

[0047] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the fish antioxidant status assessment system operating system, and utilizes various interfaces and circuits to connect various parts of the entire fish antioxidant status assessment system operating system.

[0048] The memory can be used to store the computer programs and / or modules. The processor implements the various functions of the fish antioxidant status assessment system by running or executing the computer programs and / or modules stored in the memory and accessing the data stored in the memory. The memory may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function or an image playback function); the data storage area may store data generated based on the use of the mobile phone (such as audio data and a phone book). Furthermore, the memory may include high-speed random access memory (RAM) and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0049] Although the present invention has been described in considerable detail and with particularity with respect to several embodiments, it is not intended to limit the present invention to any of these details or embodiments or any particular embodiment, so as to effectively encompass the intended scope of the present invention. In addition, the present invention has been described above with respect to embodiments foreseen by the inventors for the purpose of providing a useful description, and those insubstantial modifications of the present invention that are not currently foreseen may still represent equivalent modifications of the present invention.

Claims

1. A method for evaluating the antioxidant status of fish, characterized in that: The method comprises the following steps: S100, performing hyperspectral acquisition on the fish slice sample to obtain a spectral image; S200, selecting a white meat area of ​​the spectral image and dividing it into a plurality of segmentation areas; S300, obtaining the content characteristics of vitamin E in the segmented area; S400, grading each segmented area according to the temporal change of content characteristics; S500, the antioxidant status of different samples was evaluated through grading results.

2. A method for evaluating the antioxidant status of fish according to claim 1, characterized in that: In step S100, the method for obtaining a spectral image by performing hyperspectral acquisition on the fish slice sample is as follows: performing hyperspectral image acquisition on the fish slice sample to obtain three-dimensional hyperspectral image data containing a plurality of bands, i.e., a spectral image, wherein the hyperspectral image data covers the visible light to near-infrared band to reflect the spectral characteristics related to the antioxidant status in the fish tissue.

3. The method for evaluating the antioxidant status of fish according to claim 1, wherein: In step S200, the method for selecting the white meat region of the spectral image and dividing it into a plurality of segmented regions is as follows: constructing a pseudo-color image from a plurality of key bands in the spectral image, and extracting the white meat region of the fish through the pseudo-color image; The white meat area is divided into several segments based on the similarity of spectral features.

4. The method for evaluating the antioxidant status of fish according to claim 1, wherein: In step S300, the method for obtaining the content characteristics of vitamin E in the segmented area is: for any segmented area, the pixel spectral curve in the segmented area is averaged or median processed to form the spectral curve of the segmented area, and the vitamin E content of the spectral curve is obtained by an inversion algorithm and recorded as the content characteristic.

5. The method for evaluating the antioxidant status of fish according to claim 1, wherein: In step S400, the method for grading each segmented area according to the temporal variation of the content characteristics is as follows: the time scale of obtaining the content characteristics within the monitoring period is recorded as a measurement point; the content characteristics of each segmented area constitute a feature sequence Ft.Ls; Savitzky-Golay filtering and smoothing are performed on each Ft.Ls; and the regression fitting value of Ft.Ls in the linear regression model and the piecewise linear model is calculated using the Bayesian Information Criterion method (BIC); The calculation process of the regression fitting value corresponding to the linear regression model is as follows. Assume that the corresponding Ft.Ls conforms to the linear regression model. The least squares method is used to construct a fitting model and calculate the sum of squared residuals, denoted as ARs. The regression fitting value is obtained by inputting the sum of squared residuals into the BIC method, denoted as the single estimate Fio.bic. Then, the calculation process of the regression fitting value corresponding to the piecewise linear model is as follows. Continuously assume that there is an inflection point in the segmentation area. The least squares method is used to fit two linear regression models before and after the inflection point respectively, that is, the piecewise linear model. Based on the Bayesian information criterion, the BIC method is used to calculate the regression fitting values of each piecewise linear model, denoted as the piecewise estimate Sio.bic. If the segmentation area satisfies min(Sio.bic(HTC i )) < Fio.bic, then the piecewise linear model is accepted; otherwise, the linear regression model is accepted, where Sio.bic(HTC i ) represents the regression fitting value of the piecewise linear model corresponding to the i-th candidate inflection point; The absolute value of the slope of the corresponding model of each segmented area is calculated and recorded as the oxidation loss rate; the oxidation loss rates of each segmented area are used to form an oxidation graded sequence, and the oxidation loss rates of the oxidation graded sequence are evenly divided into several grades according to the MinMax method.

6. The method for evaluating the antioxidant status of fish according to claim 1, wherein: In step S400, the method for grading each segmented area according to the temporal variation of the content characteristics is as follows: the content characteristics of any segmented area during the monitoring period constitute a content time series, the content time series is subjected to Min-Max normalization processing to obtain a normalized time series; the absolute value of the difference between the content characteristics at each moment in the normalized time series and the content characteristics at the first moment in the reverse time direction is used as the local subsidence amount, the median of all local subsidence amounts is recorded as the local subsidence median, and the local subsidence amounts greater than the local subsidence median are accumulated to form the change point subsidence cumulative value of the segmented area; For any segmented area, the normalized time series is decomposed at multiple scales through wavelet transform to separate high-frequency detail coefficients and low-frequency approximate coefficients. The high-frequency energy and low-frequency energy corresponding to each moment are obtained through reconstruction, which are recorded as high-frequency content features and low-frequency content features, respectively. The weight wta is set based on the distribution of the cumulative value of the strain point subsidence in the numerical set of the segmented area. The cumulative subsidence trend under the contribution of high and low frequency components is calculated based on the weight wta, high-frequency content features, low-frequency content features and local subsidence, and recorded as the subsidence classification value. The segment with the smallest comprehensive subsidence grade value is selected as the high-level benchmark sample. The ratio of the difference between the comprehensive subsidence grade value of any segment and the high-level benchmark sample and the comprehensive subsidence grade value of the segment is recorded as the relative subsidence difference ratio; the relative subsidence difference ratios of all segment areas are clustered using density peak clustering to obtain several grades.

7. The method for evaluating the antioxidant status of fish according to claim 1, wherein: In step S500, the method for evaluating the antioxidant status of different samples based on the grading results is as follows: the number of segmented areas corresponding to different grades is recorded as the total grade, and the total grades are arranged in descending order to select the top 5 as representative grades; the intersection of the representative grades of the current sample and the standard sample is used as the comparative grade, and all the segmented areas corresponding to the comparative grades are intercepted from the original spectra of the current sample and the standard sample as evaluation areas, and the antioxidant status of fish in the two evaluation areas is evaluated by hyperspectral.

8. A fish antioxidant status assessment system, characterized in that: The fish antioxidant status assessment system includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the fish antioxidant status assessment method described in any one of claims 1 to 7 are implemented. The fish antioxidant status assessment system runs on a desktop computer, a laptop computer, a PDA, and a computing device in a cloud data center.

Images