Chicken embryo activity identification method based on pulse waves

Abstract

The invention discloses a chick embryo activity identification method based on pulse waves. The chick embryo activity identification method comprises the following steps: S1, acquiring chick embryo pulse wave signals by adopting a near-infrared photoelectric detection technology; s2, performing high-frequency noise removal and signal baseline drift processing on the data; s3, identifying a peak value and a valley value in the signal, and extracting a peak-valley amplitude value and a pulse period as time domain characteristic parameters; s4, performing FFT conversion on the signal, and extracting main frequency band energy, main frequency and spectrum kurtosis as frequency domain characteristic parameters; s5, fusing the time-domain features and the frequency-domain features, and inputting the fused features into a cascade model: judging whether the time-domain features reach a judgment threshold value or not by the first model, inputting the time-domain features into a second model if the time-domain features reach the judgment threshold value, and otherwise, judging that the frequency-domain features are dead embryos; the second model judges whether the frequency domain feature reaches a set threshold value, if the frequency domain feature reaches the set threshold value, the embryo is judged to be a live embryo, and otherwise, the embryo is judged to be a dead embryo. According to the method, the viability and death of the egg embryo can be judged more quickly and accurately.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of egg embryo activity detection, and particularly relates to a chicken embryo activity recognition method based on a pulse wave. BACKGROUND

[0002] Hatching egg embryos play a core role in multiple key links such as species breeding, poultry breeding and vaccine production. For poultry breeding enterprises, it is necessary to effectively distinguish active embryos from dead or weak embryos in the hatching process to avoid pollution or release of harmful substances caused by dead embryos in a high-temperature hatching environment, thereby affecting the normal development of adjacent live embryos and the survival rate of chicks.

[0003] In the field of vaccine production, active chicken embryos are widely used as biological carriers for virus amplification in the preparation of influenza vaccines. In order to ensure the quality and safety of vaccines, the activity of chicken embryos used before vaccination needs to be pre-screened, and regular monitoring needs to be carried out after vaccination to evaluate their sustained activity and production adaptability.

[0004] Near-infrared biological detection technology as a non-invasive optical detection method has been widely used in the field of embryo activity monitoring. This method uses specific waveband near-infrared light to irradiate and collect the response signal after its propagation in biological tissues, achieving preliminary evaluation of embryo activity. In typical applications, a patch-type LED light source with a center wavelength ranging from 600nm to 900nm and a patch-type photodiode detector are used to form a transmitting-receiving system to continuously irradiate and receive signals from the measured embryo. The obtained optical signal is converted into an electrical signal and transmitted to an upper processing module, and the embryo activity state is determined by a pre-set signal processing and classification algorithm to realize automatic differentiation between dead and live embryos.

[0005] Photoplethysmography (PPG) is a non-contact optical detection method that mainly uses near-infrared or visible light sources to irradiate tissues and measure the change in light intensity caused by the change in blood volume over time, thereby obtaining a pulse wave signal. This signal can dynamically reflect the flow characteristics of blood in the vascular system during the periodic contraction and relaxation of the heart, and contains multi-dimensional physiological parameter information closely related to cardiovascular activity.

[0006] In the prior art, the expression of the selected judgment features is insufficient, the judgment time is long, and a long time is needed for the input signal to achieve high-accuracy judgment of activity. SUMMARY

[0007] The present application provides a chicken embryo activity recognition method based on a pulse wave, which can quickly and accurately judge the activity of an egg embryo.

[0008] The application achieves the above technical effects through the following technical solutions:

[0009] A chicken embryo activity recognition method based on pulse waves, comprising the following steps:

[0010] S1: acquiring a chicken embryo pulse wave signal by using a near-infrared photoelectric detection technology;

[0011] S2: removing high-frequency noise and signal baseline drift from the data;

[0012] S3: identifying peak values and valley values in the signal, and extracting peak-valley amplitudes as time-domain characteristic parameters;

[0013] S4: performing FFT conversion on the signal, and extracting a main frequency band energy, a main frequency, and a spectral kurtosis as frequency-domain characteristic parameters;

[0014] S5: fusing the time-domain characteristics and the frequency-domain characteristics, and inputting them into a cascade model: a first model determines whether the time-domain characteristics reach a determination threshold value, if the time-domain characteristics reach the determination threshold value, the time-domain characteristics are input into a second model, otherwise, the chicken embryo is determined as dead; the second model determines whether the frequency-domain characteristics reach a set threshold value, if the frequency-domain characteristics reach the threshold value, the chicken embryo is determined as alive, otherwise, the chicken embryo is determined as dead.

[0015] Further, the signal acquisition time length in step S1 is 2s, and the signal sequence is 1000.

[0016] Further, in step S2, a 0.7hz Butterworth high-pass filter is used to remove noise.

[0017] Further, in step S3, the peak value and valley value identification calculation method is as follows:

[0018] The sampling signal u of the pulse wave is input as u=[u1, u2, u3,..., un], wherein n is the signal length; n

[0019] A local determination matrix is constructed, and the matrix formula for calculating the peak value is as follows:

[0020]

[0021] The matrix formula for calculating the valley value is as follows:

[0022]

[0023] In the above matrix calculation formula, the scale k is a sliding window scale, and the range is wherein represents rounding down, that is, taking the maximum integer not greater than n / 2, i is a data point index, and the range is k+1,..., n-k;

[0024] ​For each scale k, the rows of the matrix Q(k, i) are computed and calculated as follows:

[0025]

[0026] where V(k) represents the total number of local peaks or valleys at the kth scale;

[0027] The window scale k is selected that minimizes V(k) min , calculated as follows:

[0028]

[0029] The number of times each point is labeled as a peak or valley for all k < k min is counted and denoted by O(i):

[0030]

[0031] The points that satisfy O(i) = k min are finally retained, i.e., the points that are identified as peaks or valleys in all scales, and the peak set with a significant feature O(i) is extracted by the cumulative counting strategy. The formed peak set is defined as the target peak set h f , and the formed valley set is defined as the target valley set g f .

[0032] Furthermore, the main peak set h f is selected from the peak set h final , as follows:

[0033]

[0034] where a is a scaling factor for controlling the sensitivity of retaining the main peak, the value range of the scaling factor a is [0.5, 0.8], f(h) represents the amplitude of the hth candidate peak in the set, |h f | represents the number of peak points in the set, and represents the average amplitude of all peaks in the candidate set.

[0035] Furthermore, the main peak set g f is selected from the peak set g final , and the formula for identifying valleys is as follows:

[0036]

[0037] Wherein, alpha is a proportional factor, used to control the sensitivity of retaining the main valley value, represents the main valley value retained under the proportional alpha coefficient, the value range of the proportional factor alpha is [0.3, 0.5], |f(g)| represents the absolute value of the amplitude of the jth candidate valley value in the set, |g f | represents the number of valley points in the set, | represents the absolute value of the average amplitude of all valley values in the candidate set.

[0038] Further, the time domain feature determination threshold of the first model in step S5 is greater than 1000, which is the average value of all peak-valley amplitudes in the signal.

[0039] Further, the frequency domain feature determination threshold in step S5 is that the main frequency is located in the range of 2Hz to 6Hz, and the spectral kurtosis is not less than 50.

[0040] Compared with the prior art, the present application has the beneficial effects of:

[0041] The present application can more quickly and accurately judge the live and dead classification of the embryo by the multi-feature threshold judgment of the time domain feature + frequency domain feature. Compared with the single-feature threshold judgment of the same kind of embryo activity judgment method, the multi-feature threshold judgment can more accurately reflect the embryo activity. The technical scheme of the present application improves the peak and valley recognition accuracy by the improved peak and valley searching algorithm, further improves the amplitude judgment accuracy, so that a higher activity judgment accuracy can be obtained in a shorter time. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 It is a schematic diagram of the cascade model of the present application. DETAILED DESCRIPTION

[0043] Exemplary embodiments of the present application will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present application can be more thoroughly understood and so that the scope of the present application can be completely conveyed to those skilled in the art. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0044] The application provides an improved egg embryo activity evaluation method, which combines time domain characteristic parameters and frequency domain analysis indexes to jointly determine the activity state of the egg embryo. Figure 1

[0045] The specific steps are as follows:

[0046] Step 1: Obtain the chicken embryo pulse wave signal by using near-infrared photoelectric detection technology.

[0047] The sample sampling rate is 500 Hz, the sampling time length is 2 seconds, and the length of the obtained signal sequence is 1000.

[0048] Step 2: Remove high-frequency noise and signal baseline drift from the data. Based on high-pass filtering, the signal can fully adapt to the noise problem caused by the complex environment.

[0049] The application selects a classic digital filtering method, designs and applies a Butterworth high-pass filter with a cutoff frequency of 0.7 Hz according to the frequency distribution characteristics of the egg embryo PPG signal and the interference characteristics of the egg embryo turning signal. The interference signals generated by most embryo turning actions are mainly concentrated in the low-frequency region below 0.7 Hz. In order to effectively filter out these low-frequency interference signals and at the same time preserve as much as possible the effective pulse components related to the heart beat activity, the patent uses a Butterworth high-pass filter to process the original signal. The filter has good frequency response characteristics and can effectively suppress low-frequency noise without excessively weakening the target signal.

[0050] Step 3: Identify the peak and valley values in the signal and extract the peak-valley amplitude as a time domain characteristic parameter. By jointly extracting the time domain and frequency domain characteristics, such as spectral energy density and main frequency component, the activity of the egg embryo can be more accurately determined.

[0051] During the development of the chicken embryo, the pulse wave signal (PPG) may exhibit obvious double pulse wave interference characteristics with the change of the physiological state of the embryo. Specifically, in the early stage of incubation, the double pulse wave amplitude is weak, the waveform characteristics are not significant, and the interference with the main wave is limited; in the middle stage of incubation, the double pulse wave gradually strengthens, and part of it appears in the main wave, which is not easy to appear, and is easy to cause recognition error; in the late stage of incubation, the double pulse wave frequently occurs, and multiple alternating peaks often appear, which seriously interferes with the positioning of the main peak.

[0052] ​Actual observations have found that as the embryonic development cycle changes, pulse signals at different stages exhibit varying degrees of dicrotic wave superposition. Traditional methods, such as the derivative-based extreme value method, struggle to accurately extract the main wave peaks and valleys in this situation, which in turn affects the accuracy of subsequent calculation and classification of characteristic variables. To address this issue, the present invention improves an extreme value point extraction method suitable for embryonic pulse waves based on the multi-scale adaptive peak location (AMPD) algorithm. This method can stably identify the main peak and valley structure of the pulse wave under complex waveform interference, effectively avoiding the risk of misidentification caused by dicrotic wave interference, thereby improving the accuracy of signal interpretation during embryonic activity assessment.

[0053] The calculation method for peak and valley identification is as follows:

[0054] Input pulse wave sampling signal u=[u1,u2,u3,...,u n ], where n is the signal length;

[0055] Construct a local decision matrix and calculate the peak value using the following matrix formula:

[0056]

[0057] The matrix formula for calculating valley values ​​is:

[0058]

[0059] In the above matrix calculation formula, scale k is the sliding window scale, and the range is in Indicates rounding down, that is, taking the largest integer not greater than n / 2, i is the data point index, ranging from k+1,…,nk;

[0060] For each scale k, the rows and sums of the matrix Q(k,i) are calculated as follows:

[0061]

[0062] Where V(k) represents the total number of local peaks or valleys at the kth scale;

[0063] Select the window size k that minimizes V(k) min , the calculation formula is as follows:

[0064]

[0065] Count each point in all k≤k min The number of times it is marked as a peak or valley is represented by O(i):

[0066]

[0067] The points that satisfy O(i) = k are finally reserved, i.e. the points that are identified as peak or valley in all scales, the peak value set with significant characteristics of O(i) is extracted by the cumulative statistical strategy, and the formed peak value set is defined as the target peak value set h min . f The valley value set formed is defined as the target valley value set g f .

[0068] For the distinguishing condition of the primary wave and the secondary re-pulse wave, the peak value amplitude corresponding to the secondary re-pulse wave is smaller than the primary peak value, and the set h f is used to select the peak value point with the maximum amplitude characteristics, and the selection standard can be defined by the following formula:

[0069]

[0070] Wherein, a is a proportional factor, used to control the sensitivity of reserving the primary peak, represents the primary wave peak value under the proportional a coefficient, the value range of the proportional factor a is [0.5, 0.8], and the optimal value is obtained by adjusting the parameter set through the verification set in actual application, and the initial setting is recommended to be 0.65. f(h) represents the amplitude of the hth candidate peak value in the set, |h f | represents the number of peak value points in the set, represents the average amplitude of all peak values in the candidate set. The mean value of the candidate peak value set is calculated, and a proportional factor is introduced for weighting, and a dynamic threshold is set to eliminate the secondary peak with a smaller amplitude. After the screening by the strategy, only the peak value points with an amplitude not lower than the weighted mean value are reserved to constitute the final primary wave peak value set h final , so as to effectively weaken the influence of the re-pulse wave on the identification accuracy of the primary wave.

[0071] The valley value identification formula is as follows, and the valley value identification is similar to the peak value identification, and the amplitude is calculated from the peak-valley value difference.

[0072]

[0073] Wherein, β is a proportional factor, used to control the sensitivity of reserving the primary valley value, represents the primary wave valley value under the proportional β coefficient, the value range of the proportional factor β is [0.3, 0.5], and the optimal value is obtained by adjusting the parameter set through the verification set in actual application, and the initial setting is recommended to be 0.35. |f(g)| represents the amplitude absolute value of the jth candidate valley value in the set, |g f | represents the number of valley value points in the set, represents the absolute value of the average amplitude of all valley values in the candidate set. The formula calculates the average amplitude of all candidate valley values, and takes the average value as the threshold coefficient with the β coefficient, only the valley value points g∈g f that satisfy the condition are reserved to constitute the final primary valley set g final, so as to effectively exclude the interference of shallow valleys or false valleys.

[0074] In the valley identification process, the whole DC component and low-frequency fluctuation are removed after the signal is filtered by a high-pass filter, and the pulse wave signal fluctuates around zero. Therefore, the valley falls in the negative value area. To adapt to this change, the application uses the absolute value processing method for the valley, that is, the amplitude size is used as the criterion to screen the main valley point.

[0075] Feature extraction:

[0076] In the time domain, a series of feature parameters calculated by the peak and valley values are used, and the selected feature in the time domain is the peak-valley amplitude, which is calculated from the valley value and the peak value.

[0077] Peak-valley amplitude:

[0078] A = h final -g final

[0079] A is the amplitude difference between the main peak and the adjacent main valley, which defines the measure of the intensity of a single pulse wave.

[0080] Step 4: Perform FFT conversion on the signal, and extract the main frequency band energy, main frequency, and spectral kurtosis as frequency domain feature parameters.

[0081] In the frequency domain, the main frequency band energy, main frequency, and spectral kurtosis are used to reflect the frequency distribution of the embryo pulse wave.

[0082] Generally, the FFT (Fast Fourier Transform) algorithm is called to obtain the above summation result more efficiently:

[0083]

[0084] By performing FFT transform on the input signal u = [u1, u2, u3, …, u n ], where n = 0, 1, …, N-1, X(k) is obtained, and the absolute value is generally used as the amplitude spectrum. In the above formula, j represents the imaginary operation.

[0085] Main frequency:

[0086]

[0087] f peak The main frequency of the embryo pulse wave is described as a range [f1, f2], where X(f) is the complex spectrum obtained by performing fast Fourier transform on the input signal. If the frequency range deviates from the normal value, it may be a weak embryo or a dead embryo.

[0088] Spectral kurtosis:

[0089]

[0090] wherein μ is the mean value, σ is the standard deviation in frequency domain, and the greater the spectral kurtosis SK indicates the more peaked the spectral distribution.

[0091] Step 5: The time domain features and the frequency domain features are fused and input into the cascade model. In this embodiment, a cascade discriminant model based on time domain and frequency domain features is used to identify the activity state of the chicken embryo. The model includes two levels of decision logic:

[0092] The first level is time domain preliminary screening, and the threshold is set according to the mean value of the peak-valley amplitude of the pulse signal. When the mean value of the peak-valley amplitude in the signal is greater than 1000, it is determined to have active characteristics and enters the next level of frequency domain analysis; otherwise, it is determined to be a non-active state.

[0093] The second level is frequency domain determination, and the main frequency and spectral kurtosis features are further extracted. The main frequency is required to be within the range of 2Hz to 6Hz, and the spectral kurtosis should be no less than 50. Only when both frequency domain features meet the set conditions, the final determination is a live embryo; otherwise, it is determined to be a dead embryo or a non-active state.

[0094] Through multi-feature threshold judgment (time domain + frequency domain), the live and dead classification of the egg embryo can be more quickly and accurately determined. Compared with the single-feature threshold judgment of the same type of egg embryo activity judgment method, the multi-feature threshold judgment can more accurately reflect the activity of the egg embryo.

[0095] The model after setting the threshold is verified in practice, and the test sample set is 2000 live and dead white shell chicken eggs, of which 1000 are live eggs and 1000 are dead eggs. This variety of eggs is commonly used for vaccine injection culture eggs. After pulse wave measurement, the mean value of the live embryo pulse wave is above 1500, and the frequency is concentrated around 4-5Hz, which meets the threshold setting. After detection, 984 live embryos are correctly identified, 16 are misjudged, 990 dead embryos are correctly identified, and 10 are misjudged. The detection accuracy rate of live and dead chicken eggs reaches 98.7%.

[0096] The above describes the present application in detail through embodiments, but the content described is only exemplary embodiments of the present application and cannot be considered to limit the implementation scope of the present application. The protection scope of the present application is defined by the claims. Any similar technical solution that utilizes the technical solutions described in the present application or is inspired by the technical solutions of the present application within the essence and protection scope of the present application, or any equivalent changes and improvements to the application scope, shall still belong to the patent protection scope of the present application. It should be noted that, in order to clearly express, the description of the present application omits the description of some components and processes that are not directly and obviously related to the protection scope of the present application but are known to those skilled in the art.

Claims

1. A method for identifying chicken embryo activity based on pulse wave, characterized in that: The following steps are involved: S1: Use near-infrared photoelectric detection technology to obtain the chick embryo pulse wave signal; S2: remove high-frequency noise and signal baseline drift from the data; S3: Identify the peaks and valleys in the signal and extract the peak-valley amplitude as the time domain feature parameter; S4: Perform FFT transformation on the signal to extract the main frequency band energy, main frequency, and spectrum kurtosis as frequency domain feature parameters; S5: The time domain features and frequency domain features are fused and input into the cascade model: the first model determines whether the time domain features reach the judgment threshold. If so, it is input into the second model; otherwise, it is judged as a dead embryo; the second model determines whether the frequency domain features reach the set threshold. If so, it is judged as a live embryo; otherwise, it is judged as a dead embryo.

2. The method for identifying chicken embryo activity based on pulse wave according to claim 1, wherein: In step S1, the signal acquisition time length is 2 seconds, and the signal sequence is 1000.

3. The method for identifying chicken embryo activity based on pulse wave according to claim 1, wherein: In step S2, a 0.7 Hz Butterworth high-pass filter is used to remove noise.

4. The method for identifying chicken embryo activity based on pulse wave according to claim 1, wherein: The peak and valley value identification calculation method in step S3 is as follows: Input pulse wave sampling signal u=[u1,u2,u3,...,u n ], where n is the signal length; Construct a local decision matrix and calculate the peak value using the matrix formula: The matrix formula for calculating valley values ​​is: In the above matrix calculation formula, scale k is the sliding window scale, and the range is in Indicates rounding down, that is, taking the largest integer not greater than n / 2, i is the data point index, ranging from k+1, ..., nk; For each scale k, the rows and sums of the matrix Q(k, i) are calculated as follows: Where V(k) represents the total number of local peaks or valleys at the kth scale; Select the window size k that minimizes V(k) min , the calculation formula is as follows: Count each point in all k≤k min The number of times it is marked as a peak or valley is represented by O(i): Finally, the remaining contention is O(i)=k min The points that are finally retained as peaks or valleys in all scales are extracted by cumulative statistical strategy. The peak or valley positions with significant features of O(i) are extracted, and the peak set formed is defined as the target peak set h f , the formed valley value set is defined as the target valley value set g f .

5. A method for classifying dying egg embryos and active egg embryos according to claim 4, characterized in that: Peak Collection h f Filter out the main wave peak set h final , as shown below: Among them, α is the scaling factor used to control the sensitivity of retaining the main peak. The value range of the scaling factor α is [0.5, 0.8]. f(h) represents the amplitude of the h-th candidate peak in the set. f | represents the number of peak points in the set, It is expressed as the average amplitude of all peaks in the candidate set.

6. A method for classifying moribund egg embryos and active egg embryos according to claim 4, characterized in that: Peak Collection g f Filter out the main wave peak set g final , the valley value identification formula is as follows: Among them, α is the scale factor, which is used to control the sensitivity of retaining the main valley value, indicating that the main valley value is retained under the scale α coefficient. The value range of the scale factor α is [0.3, 0.5]. |f(g)| represents the absolute value of the amplitude of the jth candidate valley value in the set, |g f | represents the number of valley points in the set, It is expressed as the absolute value of the average amplitude of all valley values ​​in the candidate set.

7. The method for identifying chicken embryo activity based on pulse wave according to claim 1, characterized in that: In step S5, the time domain feature determination threshold of the first model is that the average of all peak-to-valley amplitudes in the signal is greater than 1000.

8. The method for identifying chicken embryo activity based on pulse wave according to claim 1, characterized in that: In step S5 , the frequency domain feature determination threshold of the second model is that the main frequency is within the range of 2 Hz to 6 Hz, and the spectrum kurtosis is not less than 50.

Images