Rapid batch detection method for milk cow mouth and nose methane emission and application

Abstract

The invention belongs to the field of cow performance determination and cow greenhouse gas emission monitoring, and particularly discloses a rapid batch detection method for cow mouth-nose methane emission and application. The applicant uses a method of manual selection and multiple traversal, adds a part of water absorption areas and near-infrared areas abandoned during conventional milk component analysis modeling, improves the performance of the model, and finally selects a characteristic wave band adapted to the methane emission of Holstein cows. And meanwhile, the applicant screens out a preprocessing and algorithm combination corresponding to the optimal model for establishing and predicting the methane emission of the dairy cow, so that optimal parameters are determined, and the accuracy of the model is improved.

Description

Technical Field

[0001] The invention belongs to the field of dairy cow performance measurement and dairy cow greenhouse gas emission monitoring, and specifically relates to a rapid batch detection method and application of dairy cow oral and nasal methane emission. Background Art

[0002] Methane is an important greenhouse gas with a relatively short half-life. It stays in the atmosphere for only about 12 years, which is much lower than carbon dioxide. Reducing methane emissions is an important way to quickly mitigate climate change in the short term.

[0003] According to statistics, the total amount of methane emitted by dairy cows accounts for about 19% of the total methane emissions from the global livestock industry, second only to beef cattle at 53%. It is worth noting that the methane produced by dairy cow metabolism not only exacerbates the accumulation of greenhouse gases in the global atmosphere, but also causes a loss of 5% to 14% of the total energy intake of dairy cows, resulting in feed waste.

[0004] Although there are many methods and technologies for measuring methane emissions from individual dairy cows, such as respiratory chambers (RC), breathing head boxes or masks, sulfur hexafluoride (SF6) gas tracer method, sniffers, GreenFeed emission monitoring (GEM) systems, and portable laser methane detectors (LMD), they are basically limited by high cost and low efficiency, and it is difficult to quickly and conveniently obtain methane emission data from a large number of individual dairy cows for breeding purposes. Using a combination of dairy cow trait indicators to quickly and massively predict the methane emission level of dairy cows is a feasible alternative, but most of the models currently developed by the academic community are mainly based on predictors such as dry matter intake (DMI), gross energy intake (GEI), and the proportion of dietary fiber in the diet. It is also difficult and expensive to collect these indicators in commercial farms, and the large-scale availability of such prediction models is still limited. Therefore, using low-cost and routinely recorded traits as predictive variables may be a more practical option.

[0005] When a substance absorbs infrared light between 2.5μm and 25μm, the chemical bonds or functional groups in the molecules will vibrate or rotate in a specific way, thus producing characteristic absorption peaks. Mid-infrared spectroscopy (MIRS) analysis technology can use the characteristics of the interaction between mid-infrared light and matter to analyze and determine the structure and composition of the substance, and then obtain rich chemical composition information in the sample. It has the characteristics of high sensitivity, high throughput, simple and non-destructive. MIRS of cow milk can be obtained in large quantities from the routine DHI determination of dairy cows at low cost and high efficiency. The technology of predictive analysis through milk MIRS has been routinely used for the quantitative or qualitative analysis of components in milk. In recent years, it has gradually been used to predict phenotypic information such as health diseases and physiological conditions of dairy cows.

[0006] In the past decade, foreign researchers have been exploring the feasibility of predicting methane emissions from dairy cows based on mid-infrared spectroscopy information from milk. It has been confirmed that predicting methane emissions from dairy cows based on mid-infrared spectroscopy from milk is feasible, with strong biological rationality and moderate prediction accuracy. Although models for predicting methane emissions from dairy cows through mid-infrared spectroscopy of milk have been established internationally, the accuracy of these models is not high and there is still a lot of room for improvement. In addition, the selection of bands when establishing these models is limited to the bands used for conventional milk composition analysis, and the near-infrared region and water absorption region of milk MIRS are mostly ignored.

[0007] Academic circles usually use methane production MEP (Methane production, g / d), methane density MEY (Methane yield, g / kg DMI) or methane intensity MEI (Methane intensity, g / kg Fat protein correction milk) indicators to measure the methane emission level of dairy cows. However, considering that dairy cows are mainly raised to obtain the milk they produce, the evaluation of individual methane emission levels of dairy cows should also be combined with indicators such as milk production. MEI may be more suitable for evaluating the methane emission level of dairy cows than MEP and MEY. Therefore, the present invention chooses to establish a rapid batch detection technology for dairy cow methane emissions MEI based on the mid-infrared spectrum of milk, breaking through the technical bottleneck and providing technical support for the low-carbon, healthy and sustainable development of China's dairy industry. Summary of the invention

[0008] The purpose of the present invention is to provide a rapid batch detection method for the methane emission from the mouth and nose of dairy cows. The method is simple, rapid and has a high accuracy rate compared with the true value.

[0009] Another object of the present invention is to provide an application of a method for rapid batch detection of methane emissions from the mouth and nose of dairy cows.

[0010] In order to achieve the above object, the present invention adopts the following technical measures:

[0011] A rapid batch detection method for methane emissions from the mouth and nose of dairy cows includes the following steps:

[0012] 1. The characteristic band of the infrared spectrum of the collected milk sample is: 933.64cm -1 -991.51cm -1 、1315.58cm -1 -1651.22cm -1 、1697.52cm -1 -1917.43cm -1 、2195.20cm -1 -2426.68cm -1 、2430.54cm -1 -2658.16cm -1 、2662.02cm -1 -3001.52cm -1 、3101.83cm -1 -3433.62cm -1 、3572.51cm -1 -3634.24cm -1 、3842.57cm -1 -3993.03cm -1 、4371.11cm -1 -4398.12cm -1 、4421.27cm -1 -4591.02cm -1 、4795.49cm -1 -4907.38cm -1 MIR data in;

[0013] 2. Substitute the measured MIR data into the constructed first-order derivative + trend correction + partial least squares regression model to output the predicted results of methane emissions from the mouth and nose of dairy cows;

[0014] The trend correction method is the least squares fitting method;

[0015] The principal component of the partial least squares regression is 10.

[0016] In the method described above, preferably, a gap of two wave points is allowed before and after each section of the wave band.

[0017] In the above method, preferably, the milk is milk produced by Chinese Holstein cows.

[0018] Application of a rapid batch detection method for oronasal methane emissions from dairy cows in dairy cow performance determination.

[0019] Compared with the prior art, the present invention has the following advantages:

[0020] 1. In terms of the selection of characteristic bands, we broke the commonly used method of using algorithms to screen characteristic bands, and instead used a method of manual selection + multiple traversals. In particular, it was confirmed that adding some water absorption areas and near-infrared areas that were discarded during conventional milk component analysis modeling can improve the performance of the model. Finally, we selected the characteristic bands that are suitable for the methane emissions from the mouth and nose of Chinese Holstein cows.

[0021] 2. The preprocessing and algorithm combination corresponding to the optimal model for predicting oral and nasal methane emissions from dairy cows was screened out, the optimal parameters were determined, and the accuracy of the model was improved.

[0022] 3. It realizes the rapid, accurate, low-cost and high-throughput monitoring of the methane emissions from the mouth and nose of dairy cows. It only takes 10-15 seconds to obtain the MIRS and milk composition information of the cow's milk. After importing the model, the methane emissions of the cow can be directly obtained, which greatly improves the monitoring efficiency. It has strong practicality and can be widely used in dairy cow performance measurement and dairy cow greenhouse gas emission monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is the original mid-infrared spectrum of the milk sample.

[0024] Figure 2 This is the overall spectrum of the 12 selected characteristic bands.

[0025] Figure 3 A magnified group of images for each characteristic band.

[0026] Figure 4 The correlation between the true value and predicted value of the model training set and test set and the fitted straight line graph. DETAILED DESCRIPTION

[0027] The technical solutions described in the present invention, unless otherwise specified, are all conventional solutions in the art; the reagents or materials described, unless otherwise specified, are all from commercial channels.

[0028] 1. Experimental Materials

[0029] Measuring methane in dairy cows is costly, difficult, and time-consuming, which makes it difficult to measure and obtain a large number of samples and measurement data. The GreenFeed system (GF384, C-Lock Inc., USA) was used to measure the methane emissions of dairy cows. The cows were of different parities and lactation stages to ensure that the data were representative and diverse. During the methane monitoring period, milk produced by the cows was collected and sent to the local DHI center to collect mid-infrared spectroscopy (MIRS) data.

[0030] 2. Method for determining the true (reference) value of methane emission levels from the mouth and nose of dairy cows

[0031] The GreenFeed system (GF384, C-Lock Inc., USA) was selected to monitor the methane emissions of dairy cows. The air filter of the equipment was replaced once a week, and a CO2 gas recovery test was conducted once a month to ensure that the measurement time for each cow was no less than 2 minutes each time. The raw data was corrected and processed by the C-Lock data team and exported to https: / / greenfeed.c-lockinc.com.

[0032] At the same time, considering that the methane emission levels of dairy cows vary at different times of the day, a special design is used to ensure that the emission levels of the participating dairy cows at different time points are obtained as much as possible. With 8 days as a cycle, each batch of dairy cows is measured for 3 cycles (except the fourth batch due to climate reasons). Usually, the measurement time on the first day of the cycle is 0:00AM, 8:00AM and 16:00PM, and the measurement time on the second day is 1:00AM, 9:00AM and 17:00PM, and so on. Finally, the measurement time of 8 days is covered with a measurement cycle throughout 24 hours. Ideally, each cow can obtain 24 measurement records at different time points per cycle. Each cow is guaranteed to have more than 20 valid measurement records per cycle, and the arithmetic mean is taken as the final data. Data that does not meet this condition will be eliminated.

[0033] 3. Milk collection and mid-infrared spectroscopy

[0034] The cows were milked three times a day, at around 3:00 AM, 11:00 AM and 19:00 PM, using milking equipment in the milking parlor to record milk production. During the monitoring of methane emissions from the cows to be tested, milk samples were collected from the entire milking process. 40 mL of milk was collected from each cow each time and divided into cylindrical sampling tubes with a diameter of 3.5 cm and a height of 9 cm. They were numbered in sequence, and bronopol preservative was immediately added to each sampling bottle and slowly shaken to fully dissolve it. The milk samples were stored at 4°C before being transported to the DHI center to prevent deterioration, and spectra were collected immediately after the samples arrived at the DHI center.

[0035] After arriving at the DHI testing center, the sample tube containing the milk sample was placed in a 42°C water bath for 15-20 minutes and measured using FOSS's high-throughput milk somatic cell analyzer (COMBIFOSS 7DC). After the sample was shaken well, it was sent into the instrument for scanning to obtain milk composition information and MIRS. Figure 1 This is the original mid-infrared spectrum of the milk sample.

[0036] 4. Data Processing

[0037] All milk samples collected during the monitoring period were excluded due to spoiled samples, empty tubes (not collected), abnormal milk composition determination (only the determination records of 1.5% < milk fat rate < 9%, 1% < milk protein rate < 7%, and somatic cell count < 1000K were retained), and invalid data caused by abnormal mid-infrared spectroscopy determination were removed. Then, the retained milk composition records and MIRS within a single cycle were weighted averaged according to the milk production of each milking, and the daily average fat protein corrected milk production FPCM (Fat protein correction milk, kg / d) and average MIRS of the participating cows in that cycle were calculated. FPCM = milk production × (0.337 + 0.116 × milk fat rate + 0.06 × milk protein rate).

[0038] According to the experimental design, the cattle that withdrew due to illness during the monitoring period were first eliminated; then only the cattle to be tested with a single-cycle measurement record of not less than 20 times were retained to ensure the accuracy of the measurement results; finally, all the measurement records in a single cycle were averaged as the average methane production of the test cattle in that cycle, and then the cycle average methane level MEI of the test cattle was obtained (MEI=MEP / FPCM), and a total of 210 data were obtained for modeling.

[0039] Embodiment 1:

[0040] Establishment of a method for detecting methane emissions from dairy cows using mid-infrared spectroscopy:

[0041] 1. The actual methane emissions of dairy cows monitored by the GreenFeed system and the division of the modeling data set are as follows

[0042]

[0043] In the modeling data set division in this embodiment, about 75% of the data is used as a training set and 25% of the data is used as a test set.

[0044] 2. Screening of MIRS data preprocessing methods for modeling

[0045] The embodiment of the present invention uses the partial least squares regression (PLSR) algorithm to establish the model. Effective feature screening is the basic operation of processing spectral data, the purpose is to eliminate noise, and lay a good foundation for extracting features. Effective feature screening mainly includes feature extraction, feature preprocessing and feature dimension reduction. This embodiment mainly performs feature preprocessing on spectral data. In the modeling process, sometimes only one preprocessing will limit the generalization ability and optimization potential of the model, and integrated preprocessing has been proven to produce better results in the model building process. Therefore, before dividing the data set, this embodiment selects None (no preprocessing), MMS (maximum and minimum normalization), SS (standardization), CT (mean centering), MA (moving average smoothing), SG (convolution smoothing), MSC (multiple scattering correction), DT (trend correction), SNV (standard normal variable transformation), wave (wavelet transform), D1 (first-order derivative) and D2 (second-order derivative) A total of 12 methods are selected to select two preprocessing methods to perform integrated preprocessing on spectral data, and the method of trend correction is least squares fitting.

[0046] The single preprocessing comparison results are as follows:

[0047]

[0048] After comparison, it was found that when the original spectrum was used for modeling, there was a large difference between the performance of the training set and the test set, indicating that the model was overfitted. When the spectrum after DT pretreatment was used for modeling, the overfitting of the model was best improved. Therefore, this embodiment selected DT pretreatment as the basis for integrated pretreatment.

[0049] The comparison results of integrated preprocessing based on DT preprocessing are as follows:

[0050]

[0051]

[0052] After comparison, it was found that when the spectrum modeling was performed after SG+DT integrated preprocessing, the overfitting of the model was improved, but the performance of the model decreased. When the spectrum modeling was performed after D1+DT integrated preprocessing, the overfitting of the model was not improved, but the performance of the model was greatly improved. Considering that the characteristic bands will be selected later, the performance of the model has greater room for improvement and optimization. Therefore, this embodiment selects D1+DT integrated preprocessing.

[0053] 1. Manual selection process and determination of modeling feature bands

[0054] There are many methods for selecting feature bands, mainly including algorithmic feature selection and manual feature selection. The principle of algorithmic feature selection is mainly derived from the correlation between each wave point and the reference value. The advantages are fast speed and high efficiency, but the disadvantages are that the synergy between adjacent wave points is ignored and the idea is relatively simple. The advantage of manual feature selection is that the role of the band (i.e., adjacent wave points) can be strengthened during the selection process, and at the same time, the original information state of the spectrum can be retained more in the process of improving the model. It has stronger inclusiveness and generalization ability, and the band selection is accurate. The disadvantages are slow selection speed and low efficiency.

[0055] This embodiment selects the characteristic band by manual selection, and the steps are as follows:

[0056] (1) Determine the modeling algorithm. In this embodiment, the classic partial least squares regression algorithm is selected as the prediction algorithm for the methane emission from the mouth and nose of dairy cows.

[0057] (2) Determine the best preprocessing. In this embodiment, the best preprocessing combination selected before dividing the data set is the D1+DT preprocessing combination.

[0058] (3) Manually select characteristic bands. The specific process is as follows:

[0059] 1) The 925.92cm -1 -5011.54cm -1 The spectrum area is divided into 11 segments, of which 925.92cm -1 -4780.06cm -1 Divided into 10 sections, each with 100 dots, greater than 4780.06cm -1 This is the last section, containing 60 dots.

[0060] 2) Using the partial least squares regression algorithm, first add or subtract a group of wave points (0 to 100 wave points) at both ends of the critical point of the first band, select the optimal number of wave point additions and subtractions based on the model effect, and perform similar operations on the second band based on this. Finally, the first traversal is completed after completing a round of operations on all eleven bands.

[0061] 3) After the first traversal is completed, perform a second, third or more manual traversals until all the wave points no longer change, which is the optimal characteristic band.

[0062] 4) If adding or subtracting wave points at both ends of a band fails to improve the model effect and the band is long, consider dividing the band in half from the middle and adding or subtracting wave points again from both ends of the dividing point.

[0063] Finally, after multiple rounds of screening, the optimal results were obtained, as shown in the table:

[0064]

[0065] A total of 12 characteristic bands were selected, and the result is: 933.64cm -1 -991.51cm -1 、1315.58cm -1 -1651.22cm - 1 、1697.52cm -1 -1917.43cm -1 、2195.20cm -1 -2426.68cm -1 、2430.54cm -1 -2658.16cm -1 、2662.02cm -1 -3001.52cm -1 、3101.83cm -1 -3433.62cm -1 、3572.51cm -1 -3634.24cm -1 、3842.57cm -1 -3993.03cm -1 、4371.11cm -1 -4398.12cm -1 、4421.27cm -1 -4591.02cm -1 、4795.49cm -1 -4907.38cm -1 , a gap of two wave points is allowed before and after each section. Figure 2 This is the overall spectrum of the 12 selected characteristic bands. Figure 3 A magnified group of images for each characteristic band.

[0066] 2. Screening and determination of model parameters

[0067] The parameters of the model usually include the parameters of the preprocessing method and the parameters of the modeling algorithm. Since this model uses the D1+DT integrated preprocessing method and the PLSR modeling algorithm, it only needs to select the algorithm parameters. The parameters of the partial least squares regression algorithm are the principal components (n_component). The comparison of parameter selection results is as follows:

[0068]

[0069]

[0070] It can be found that when the principal component (n_component) is 12, the result of the model training set is the best, but there are relatively more overfitting cases, so the principal component (n_component) is finally selected as 10.

[0071] In summary, after comparative analysis, the best regression model for methane emissions from the mouth and nose of dairy cows is the D1+DT+PLSR (n_component=10) model. The correlation coefficients of the training set and the test set are 0.9356 and 0.9355, respectively; the root mean square errors of the training set and the test set are 1.0794 and 1.3735, respectively. Figure 4 The correlation between the true value and predicted value of the model training set and test set and the fitted straight line graph.

[0072] Embodiment 2:

[0073] Application of rapid batch detection method for oronasal methane emissions from dairy cows:

[0074] The established optimal regression model for oronasal methane emissions from dairy cows (D1+DT+PLSR (n_component=10)) was used to predict five randomly selected samples (not one of the 210 modeling samples), and the predicted results were compared with the true values.

[0075] How to use the model:

[0076] 1. The characteristic band of the infrared spectrum of the collected milk sample is: 933.64cm -1 -991.51cm -1 、1315.58cm -1 -1651.22cm -1 、1697.52cm -1 -1917.43cm -1 、2195.20cm -1 -2426.68cm -1 、2430.54cm -1 -2658.16cm -1 、2662.02cm -1 -3001.52cm -1 、3101.83cm -1 -3433.62cm -1 、3572.51cm -1 -3634.24cm -1 、3842.57cm -1 -3993.03cm -1 、4371.11cm -1 -4398.12cm -1 、4421.27cm-1 -4591.02cm -1 、4795.49cm -1 -4907.38cm -1 MIR data in;

[0077] 2. Substitute the MIR data obtained by the measurement into the D1+DT+PLSR (n_component=10) model constructed in Example 2 to output the prediction results of the methane emissions from the mouth and nose of dairy cows.

[0078] The results predicted by the model are very close to the actual results of monitoring the methane emissions from the oronasal cavity of dairy cows using the GreenFeed system (GF384, C-Lock Inc., USA) (as shown in the table below). Therefore, the model has a high accuracy and can be used to predict the methane emissions from the oronasal cavity of dairy cows.

[0079]

Claims

1. A rapid batch detection method for methane emissions from the mouth and nose of dairy cows, comprising the following steps:

1. The characteristic band in the infrared spectrum of the collected milk sample is: 933.64 cm -1 -991.51 cm -1 、1315.58 cm -1 -1651.22 cm -1 、1697.52 cm -1 -1917.43 cm -1 、2195.20 cm -1 -2426.68 cm -1 、2430.54 cm -1 -2658.16 cm -1 、2662.02 cm -1 -3001.52 cm -1 、3101.83 cm -1 -3433.62 cm -1 、3572.51 cm -1 -3634.24 cm -1 、3842.57 cm -1 -3993.03 cm -1 、4371.11 cm -1 -4398.12 cm -1 、4421.27 cm -1 -4591.02 cm -1 、4795.49 cm -1 -4907.38 cm -1 MIR data in; 2. Substitute the measured MIR data into the constructed first-order derivative + trend correction + partial least squares regression model to output the predicted results of dairy cow methane emissions; The trend correction method is the least squares fitting method; The principal component of the partial least squares regression is 10.

2. The method according to claim 1, characterized in that: A gap of two wave points is allowed before and after each section of the band.

3. The method according to claim 1, characterized in that: The milk in the method is milk produced by Chinese Holstein cows.

4. Application of the rapid batch detection method for oronasal methane emissions from dairy cows as described in claim 1 in dairy cow performance measurement.

Images