A target positioning type egg gender infrared spectrum nondestructive identification system and method

Abstract

This invention discloses a targeted, non-destructive infrared spectral identification system and method for the sex of hatching eggs, belonging to the field of intelligent spectral sensing technology. The system includes a spectral acquisition module, a sample carrying platform, a data processing module, and an intelligent discrimination module. The method first uses a clustered fiber optic probe to acquire spectra and construct a three-dimensional data matrix. After local submatrix partitioning and clustering discrimination, initial screening of the embryonic region is completed. Then, a reverse search strategy is used to accurately locate key biological target areas. Based on this, a support vector machine recognition model is constructed by combining multi-algorithm fusion preprocessing and feature band optimization, thereby achieving rapid determination of the sex of hatching eggs. This invention can target and locate feature regions, effectively eliminate interfering information, and the detection process is fast and non-destructive. It avoids the ethical controversies and resource waste associated with culling male chicks in the later stages of incubation and is suitable for egg-laying hen breeding and automated incubation and sorting.

Description

Technical Field

[0001] This invention belongs to the field of spectral detection and intelligent sensing technology, specifically relating to a targeted positioning infrared spectral non-destructive identification system and method for the sex of hatching eggs. Background Technology

[0002] Eggs are a vital global food and agricultural resource. However, in egg-laying hen farming, roosters are often culled in large numbers after hatching due to their lack of egg-laying ability and poor growth performance. This not only wastes resources but also raises animal welfare controversies. Research shows that embryos develop pain perception around day 7 of incubation. Therefore, accurately determining the sex of fertilized eggs in the early stages of incubation (days 0 to 6) can significantly reduce ethical risks, effectively improve hatching efficiency, and promote the sustainable development of the industry.

[0003] Existing methods for sex determination, such as hormone testing, DNA molecular markers, and histological analysis, while achieving high accuracy, generally suffer from drawbacks such as high invasiveness, long processing times, high costs, and difficulty in meeting the requirements of industrial-scale batch testing. In contrast, infrared spectroscopy, which achieves non-destructive analysis through molecular vibrational characteristics, has shown promising application prospects in fields such as food quality testing and biological tissue identification. However, its application to early embryo detection in fertilized eggs still faces significant challenges: the acquired spectral data is often mixed with a large amount of background noise and baseline drift, resulting in extremely weak embryo-specific sex characteristic signals that are difficult to extract due to low signal-to-noise ratios. Traditional full-spectrum analysis or conventional signal processing methods are insufficient to "remove false signals," failing to accurately eliminate background interference from the eggshell and non-embryo areas, severely limiting the accuracy and stability of identification.

[0004] In view of this, there is an urgent need to develop an intelligent identification technology that can target and locate key biological regions and effectively suppress background interference, hence this application is hereby submitted. Summary of the Invention

[0005] The purpose of this invention is to provide a targeted, non-destructive infrared spectroscopy system and method for sex identification of hatching eggs, aiming to solve the problems of high invasiveness, significant background interference, and difficulty in extracting weak signals in existing hatching egg detection technologies. By introducing a targeted positioning mechanism, this invention can automatically locate key embryonic characteristic regions within fertilized eggs in the early stages of incubation, achieving early, non-destructive, and non-contact high-precision sex detection. This provides the poultry hatching industry with an efficient, reliable, and animal welfare-compliant technical solution.

[0006] A targeted, non-destructive infrared spectroscopy system for sex identification of hatching eggs includes a spectral acquisition module, a sample support platform, a data processing module, and an intelligent discrimination module. The spectral acquisition module uses an infrared spectrometer with a clustered fiber optic probe to collect raw infrared spectral data from the equatorial ring of the hatching eggs at multiple distributed points. The sample support platform is a customized bracket used to stably fix the equatorial plane of the hatching eggs to the focal plane of the infrared spectrometer, avoiding spectral drift caused by egg posture deviations. The data processing module integrates functional units such as 3D matrix construction, local submatrix partitioning and clustering discrimination, reverse search and target area localization, multi-algorithm fusion spectral preprocessing, and feature band selection, used to process and extract features from the acquired spectral data. The intelligent discrimination module incorporates a support vector machine recognition model, using the feature information output by the data processing module to quickly and non-destructively determine the sex of the hatching eggs. This system, through the collaborative design of multiple modules, achieves full automation from high-fidelity data acquisition to intelligent discrimination. On the one hand, customized hardware ensures the stability and consistency of spectral acquisition; on the other hand, the intelligent algorithm module achieves precise targeting of the biological target area and background removal. The combination of these two methods effectively avoids the interference of differences in egg posture and environmental fluctuations on the test results, significantly improving the overall robustness and accuracy of the system.

[0007] Furthermore, the spectral acquisition module employs an infrared spectrometer equipped with a clustered fiber optic probe. The spectral acquisition range is set to 4000 cm⁻¹. -1 Up to 650 cm -1 Spectral resolution of 4 cm -1 The system employs a cumulative scan count of 32 to significantly optimize the spectral signal-to-noise ratio. On the hardware side, it incorporates a high-energy silicon carbide (Globar) light source and utilizes an MCT (mercury cadmium telluride) detector to compensate for fiber optic transmission loss and ensure signal strength. Furthermore, by optimizing the probe's incident angle relative to the sample, it is designed to be incident at a non-perpendicular angle, effectively avoiding specular reflection interference. These specific spectral ranges and resolutions comprehensively cover key information related to the sex characteristics of hatching eggs and embryos, effectively improving the quality of the raw spectral data and providing a reliable data foundation for subsequent feature extraction and sex determination.

[0008] Furthermore, the support vector machine (SVM) recognition model in the intelligent discrimination module employs a radial basis function (RBF) for nonlinear mapping. Given the high-dimensional and nonlinear distribution characteristics of spectral data, the SVM model using the RBF exhibits powerful nonlinear processing capabilities. This method can map gender features that are difficult to distinguish in the original low-dimensional space to a high-dimensional feature space, thereby achieving linear separability of samples and effectively solving the problem of class overlap. This not only significantly improves the model's classification performance for weak gender signals but also greatly reduces the false positive rate and enhances the system's generalization ability.

[0009] A non-destructive infrared spectroscopy method for targeted sex identification of hatching eggs includes the following steps:

[0010] Step 1: Spectral Data Acquisition and 3D Matrix Construction: Using an infrared spectroscopy system in conjunction with a clustered probe, distributed multi-point acquisition of data was performed on the equatorial ring of the hatching eggs. The acquired raw data was reconstructed into a 3D data matrix according to the location coordinates and fiber optic channels, ensuring that each sampling unit contains complete location-spectral information, laying a solid data foundation for subsequent local micro-area analysis and feature recognition.

[0011] Step 2: Local Submatrix Partitioning and Clustering Identification: Based on the principle of spatial proximity, the 3D data matrix is ​​divided into a series of spatially overlapping local submatrices using a sliding window technique. After calculating the regional representative spectrum of each submatrix, unsupervised classification is performed using the K-means clustering algorithm. By identifying the similarity of spectral morphology, preliminary screening of potential embryonic tissue regions is achieved. This step effectively narrows the analysis scope and improves the signal-to-noise ratio and accuracy of subsequent sex feature signal extraction.

[0012] Step 3: Reverse Search and Target Region Localization: For the candidate region set selected in Step 2, a spectral difference scoring mechanism is established to quantify the degree of difference between each candidate region and the background region. Based on the scoring results, the regions are sorted, and a reverse search and region growth expansion are performed starting from the region with the greatest difference. Combining spatial connectivity and spectral consistency criteria, the most significant sex-specific target regions are identified, achieving precise localization of key biological target regions.

[0013] Step 4: Multi-algorithm fusion spectral preprocessing: Extract all spectral information from the located target area, and perform a weighted average of the spectra at each point based on the signal-to-noise ratio to obtain a representative spectrum for the region. Apply multi-algorithm fusion preprocessing technology to this representative spectrum to simultaneously achieve noise suppression, scattering effect correction, and gender feature enhancement. This step significantly improves the stability of the spectral data and the discriminative ability of the classification model.

[0014] Step 5: Feature Band Selection and Classification Modeling: Simulated annealing is introduced to optimize the spectrum across the entire spectrum, selecting feature band combinations with high discriminative power and low redundancy for gender discrimination. The optimized feature wavelength data is input into a Support Vector Machine (SVM) classification model, and a radial basis function (RBF) kernel is used for nonlinear mapping to improve the model's classification performance. A joint strategy of grid search and cross-validation is employed to optimize the model parameters, ensuring that the classification model has good generalization ability and high recognition accuracy.

[0015] Step 6: Gender determination and result output: The optimized support vector machine recognition model is used to determine the gender of the hatching eggs. This model has the characteristics of high recognition accuracy, stable F1 value and low false positive rate, and can reliably complete the non-destructive identification of the gender of hatching eggs in the early stage of incubation. This method has the advantages of fast detection speed and high accuracy, which can meet the application requirements of industrial high-throughput automatic sorting.

[0016] Compared with the prior art, the beneficial effects of the present invention are:

[0017] 1. Targeted Identification: Through intelligent algorithms, the system automatically identifies and accurately locates key developmental areas of the early embryo within the egg, effectively solving the problem of high background noise in traditional detection. It achieves targeted enhancement of weak sex characteristic signals and effective removal of background interference, significantly improving the signal-to-noise ratio and effectiveness of the original data.

[0018] 2. High accuracy and speed: By integrating multi-algorithm spectral preprocessing and feature band optimization strategies, redundant information is effectively removed, significantly improving the model's accuracy and generalization ability in identifying embryo sex.

[0019] 3. Non-contact, non-destructive testing: The testing process is entirely based on infrared optical detection principles, requiring no breaking, puncturing, or sampling of the eggs, and involves no physical contact, thus avoiding the risks of cross-infection, mechanical damage, and stress. This system can complete the testing without altering the internal microenvironment of the eggs or the normal developmental process of the embryo, effectively ensuring hatchability and biosafety.

[0020] 4. Ethical and economic value: This invention moves the time window for sex determination to the early incubation period before the embryo develops pain sensation, thereby reducing the ethical controversy caused by the large-scale culling of male chicks in the later stages of incubation. Attached Figure Description

[0021] Figure 1 A flowchart illustrating the use of a targeted, positioning-based infrared spectroscopy non-destructive identification system for the sex of hatching eggs.

[0022] Figure 2 This is a schematic diagram of infrared spectral data acquisition and egg placement, where: Figure 2 a is a schematic diagram of the overall structure of the infrared spectral data acquisition device; Figure 2 b is a schematic diagram showing the relative positions of the hatching egg and the bundled optical fiber;

[0023] Figure 3 This is the confusion matrix for gender classification in the SVM model of this embodiment;

[0024] Figure 4 This is a diagram showing the training and cross-validation results of the SVM model in this embodiment.

[0025] Among them, 1—infrared spectral data acquisition device; 3—bundled fiber optic probe; 4—fiber optic connection assembly; 5—egg support device; 206—egg shell. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figures 1 to 4 The present invention provides a technical solution: a targeted positioning infrared spectroscopy non-destructive identification system and method for the sex of hatching eggs. The present invention will be further described in detail below with reference to embodiments.

[0028] A specific embodiment of the present invention provides a targeted positioning method for non-destructive infrared spectroscopy identification of the sex of hatching eggs, the specific technical solution of which is as follows:

[0029] 1. System hardware configuration and experimental environment setup:

[0030] Spectral acquisition module: Employs an infrared spectrometer equipped with a bundled fiber optic probe, with a spectral range of 4000 cm⁻¹. -1 Up to 650cm -1 Spectral resolution of 4 cm -1 The cumulative scan count was set to 32 times to optimize the signal-to-noise ratio. The system incorporates a high-energy silicon carbide (Globar) light source and employs an MCT (mercury cadmium telluride) detector to ensure signal strength after fiber optic transmission. By optimizing the incident angle of the probe relative to the sample, specular reflection interference is effectively avoided.

[0031] Sample support platform: A customized bracket is used to ensure that the equatorial plane of the hatching eggs is stably located at the focal plane, thereby avoiding spectral drift caused by attitude deviation.

[0032] 2. Distributed spectral acquisition and 3D matrix construction:

[0033] The hatching eggs were fixed on a high-precision two-dimensional rotating / translating stage. An infrared spectrometer equipped with an L-channel fiber optic probe was used. The control platform moved the eggs along a preset path, allowing the probe array to perform sequential, fixed-point spectral acquisition on the eggshell surface. Each sampling point acquired a full-spectrum data point containing Λ wavenumber points. By associating the position index of the fiber optic probe with the acquisition sequence number, all the acquired spectral data were reconstructed into a three-dimensional data matrix D(M, N, Λ).

[0034] Where M(1, 2, ..., L) is the fiber channel index, N(1, 2, ...) is the scan position sequence, and Λ is the spectral wavenumber dimension. This three-dimensional matrix is ​​equivalent to a spectral data cube with a spatial resolution of L×|N|.

[0035] In the data cube, the spectral vector corresponding to each sampling point (i, j) is:

[0036] .

[0037] in, This represents the spectral value of the k-th fiber channel at the acquisition point (i, j).

[0038] 3. Neighborhood matrix partitioning:

[0039] On the spatial plane (i.e., M×N dimensions) of the reconstructed 3D data matrix D(M, N, Λ), a sliding window of size a×b is defined. This window moves across the spatial grid with a fixed step size, thereby generating a series of spatially overlapping sub-matrices. Each submatrix is ​​defined as:

[0040] .

[0041] Where u and v are the starting index coordinates of the sub-matrix block. Feature continuity is improved by using overlapping sliding windows.

[0042] For each submatrix, calculate its average spectrum:

[0043] (1)

[0044] 4. Cluster analysis:

[0045] Pair matrix {S U,V After calculating the corresponding average spectrum, K-means clustering was used on {S}. U,V Unsupervised classification is performed. Its clustering optimization objective function is:

[0046] (2)

[0047] Among them, C k Let μ represent the centroid of the k-th cluster. p Let be the average spectrum of the p-th submatrix in the set.

[0048] The distance metric used is Euclidean distance:

[0049] (3)

[0050] After clustering is completed, the centroids of each cluster are compared with the pre-defined standard spectral characteristics of early embryonic development regions, and the closest category is selected as the preliminary candidate region set C={S}. k}

[0051] 5. Reverse search and localization of key biological target areas:

[0052] For each candidate region S obtained through filtering k Construct a target saliency scoring function This is used to quantify the spectral difference between the candidate region and the background region. The scoring function is expressed as:

[0053] (4)

[0054] in, Candidate region S k In wavenumber The average spectral intensity at that location.

[0055] , These represent the average spectral intensity and standard deviation of the background region at wavenumber λ, respectively.

[0056] To prevent tiny constants with a denominator of zero.

[0057] This is the spectral weighting coefficient. This coefficient is set based on prior knowledge, assigning higher weight to bands with significant sex characteristics of the hatching egg, thereby guiding the algorithm to focus on more biologically significant spectral regions and achieving precise positioning.

[0058] After calculating the score for each candidate region, the candidate region S with the highest score is selected. k As the initial seed, the reverse search and region growing algorithm is executed, with the following specific steps:

[0059] (1) Initialization: Initialize the current target region R as the high-scoring candidate region.

[0060] (2) Neighborhood expansion and score calculation: Examine the neighborhood sampling points of region R, and calculate the aggregate score F(R') of the expanded new region R':

[0061] (5)

[0062] (3) Growth determination: If the expanded score F(R') increases, the expansion is accepted and the current target region R=R' is updated.

[0063] (4) Iteration Termination: Repeat steps (2)-(3) until the aggregate score F(R) no longer increases significantly or the region area reaches the preset upper limit threshold. The region R obtained at this point... * This is the key biological target area for final location.

[0064] 6. Target region spectral fusion and preprocessing:

[0065] The target area R obtained by positioning * Spectral data processing is performed to eliminate noise interference and enhance effective information, specifically including the following sub-steps:

[0066] (1) Spectral fusion: from target region R * Extract T spectra {x} t The representative spectrum is obtained by weighting and averaging the signals and noise ratios of each spectrum.

[0067] (6)

[0068] in, Let be the signal-to-noise ratio weighting coefficient corresponding to the t-th spectrum.

[0069] (2) Absorbance conversion:

[0070] If the instrument outputs transmittance I(ν), it needs to be converted into absorbance according to Beer-Lambert's law:

[0071] (7)

[0072] Where I0(v) is the background reference spectrum and v is the wavenumber.

[0073] (3) SG smoothing and derivative:

[0074] The spectrum was smoothed using the Savitzky–Golay (SG) filtering algorithm. A p-order polynomial was used for fitting within a window of width (2m+1). The smoothed spectrum is as follows:

[0075] (8)

[0076] Among them, c j These are the fitting coefficients determined using the least squares method. Further, the smoothed spectrum is processed using the first derivative to enhance peak edge features and eliminate baseline drift.

[0077] (9)

[0078] (4) Standard Normal Variable (SNV) Transformation:

[0079] SNV transformation is used to eliminate scattering differences and scale effects between samples. For a single spectral vector a = [a1, ..., a...], ... Nv Standardize:

[0080] , , (10)

[0081] in, σ is the spectral mean. a The standard deviation is denoted as .

[0082] 7. Selection of characteristic bands:

[0083] The simulated annealing (SA) algorithm was used to select the subset of characteristic wavelengths that contribute most to gender differentiation from the full-band spectral data. The specific process is as follows:

[0084] Set an initial temperature T0, and update the temperature by decreasing according to the following formula:

[0085] (11)

[0086] Among them, T k Let α be the temperature at the k-th iteration, and α be the cooling coefficient.

[0087] During the state transition process, the state acceptance probability P is defined based on the Metropolis criterion:

[0088] (12)

[0089] in, This represents the difference between the objective function values ​​corresponding to the old and new states. The algorithm terminates when the objective function value approaches minimum convergence or the temperature falls below a preset threshold, outputting the optimal set of feature wavelengths, S. * .

[0090] 8. SVM Model Establishment and Optimization:

[0091] Support Vector Machine (SVM) is a typical statistical learning-based classification algorithm suitable for both linear and nonlinear classification problems. Given the characteristics of the sex spectral data of hatching eggs, this invention employs SVM to establish a classification model.

[0092] 8.1 Basic Model of Linear SVM:

[0093] Let the training dataset be... , where the input vector ;

[0094] Category Tags , indicating the sample category (male or female).

[0095] The goal of linear SVM is to find a separating hyperplane that maximizes the geometric margin while correctly splitting the dataset. This problem is transformed into solving for the weight vector. Minimization problem:

[0096] (13)

[0097] Introducing Lagrange multipliers >0, construct the Lagrange function:

[0098] (14)

[0099] Based on the principle of duality, we can solve the dual problem of the above problem, that is, find the maximum value of α:

[0100] (15)

[0101] The optimal solution α is obtained. * Afterwards, the optimal linear classification decision function is:

[0102] (16)

[0103] 8.2 Nonlinear kernel function mapping (RBF kernelization):

[0104] To address the potential nonlinear distribution characteristics of spectral data, a radial basis function (RBF) is introduced as the kernel function:

[0105] (17)

[0106] The RBF kernel function is used to map the input vector from the original space to a high-dimensional feature space to achieve linear separability. The kernelized decision function is:

[0107] (18)

[0108] 8.3 Search and optimization of hyperparameters (C, γ):

[0109] The classification performance of the model is mainly affected by the penalty parameter C and the kernel function parameter γ. To obtain the optimal parameter combination, this invention adopts a joint optimization strategy of grid search and k-fold cross-validation. The accuracy calculation formula is as follows:

[0110] (19)

[0111] Where N is the number of samples in the validation set. For predicting categories, For actual categories; As an indicator function, when the predicted value Compared with actual value If the values ​​are the same, the value is 1; otherwise, it is 0.

[0112] 9. Performance Evaluation and Mixed Matrix:

[0113] To comprehensively evaluate the performance of the classification model, a confusion matrix is ​​constructed and accuracy, precision, recall, and F1 score are calculated.

[0114] The structure of the mixed-ingredient matrix is ​​shown in the table below:

[0115] Predicted positive Predicted Negative Actual positive TF FN Actual Negative FP TN

[0116] The calculation formulas for each evaluation indicator are as follows:

[0117] (1) Precision: .

[0118] (2) Accuracy: .

[0119] (3) Recall rate: .

[0120] (4) F1 score: .

[0121] 10. Results Analysis:

[0122] This invention achieves rapid and non-destructive early sex determination of fertilized egg embryos by combining infrared spectroscopy, 3D data modeling, targeted localization algorithms, and intelligent classification models. It innovatively proposes a reverse search localization mechanism based on spectral dependence scoring, which can accurately identify key developmental regions of the embryo in complex signal environments, thereby effectively enhancing sex characteristic signals and suppressing background interference. Furthermore, this invention employs multispectral fusion preprocessing and simulated annealing algorithm optimization, significantly improving feature quality and enhancing the discriminative ability of the classification model. The optimized support vector machine (SVM) classification model constructed in this way exhibits high accuracy and strong generalization ability in experimental tests, and can operate stably under different conditions. This technical solution not only improves the accuracy of sex identification but also has good engineering application prospects. The designed system can seamlessly integrate with automated incubation and sorting equipment, meeting the needs of large-scale production, and has significant economic benefits and application potential. This research provides the poultry breeding industry with an efficient, low-cost, and animal welfare-compliant early sex identification technology solution, promoting the integration of intelligence and automation, and has important industrial application value.

[0123] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A targeted positioning infrared spectroscopy non-destructive identification system for the sex of hatching eggs, characterized in that: include: The system includes a spectral acquisition module, a sample carrying platform, a data processing module, and an intelligent discrimination module. The spectral acquisition module includes an infrared spectrometer and a bundled fiber optic probe, which are used to perform distributed multi-point scanning of the equatorial ring of hatching eggs to acquire raw infrared spectral data. The sample carrier platform is a customized bracket used to fix the hatching eggs and stabilize the equatorial plane of the hatching eggs on the focal plane of the infrared spectrometer to avoid spectral drift caused by attitude deviation; The data processing module integrates a 3D matrix construction unit, a local submatrix partitioning and clustering unit, a reverse search and target area localization unit, and a multi-algorithm fusion spectral preprocessing and feature band selection unit, which are used to process and extract features from the acquired spectral data. The intelligent discrimination module has a built-in support vector machine model, which is used to quickly and non-destructively determine the sex of hatching eggs based on the feature information output by the data processing module.

2. The targeted positioning infrared spectroscopy non-destructive identification system for the sex of hatching eggs as described in claim 1, characterized in that: The spectral acquisition module includes an infrared spectrometer and a clustered fiber optic probe; the spectral acquisition range of the infrared spectrometer is 4000 cm⁻¹. -1 Up to 650 cm -1 Spectral resolution of 4 cm -1 The number of scans is configured to be 32; the infrared spectrometer is equipped with a silicon carbide light source and a mercury cadmium telluride detector; the incident end of the bundled fiber optic probe is tilted relative to the surface of the sample to be tested, and is configured to avoid the specular reflection light path.

3. The targeted positioning infrared spectroscopy non-destructive identification system for the sex of hatching eggs as described in claim 1, characterized in that: It also includes a classification modeling unit, which uses a support vector machine classification model and employs a radial basis kernel function for nonlinear mapping.

4. A non-destructive infrared spectroscopy method for targeted positioning of fertilized eggs to identify sex, characterized in that: Includes the following steps: Step 1: Spectral data acquisition and 3D matrix construction. An infrared spectral system is used in conjunction with a clustered fiber optic probe to collect data from multiple points distributed along the equatorial ring of the hatching eggs. The collected raw data is then reassembled into a 3D data matrix according to the position coordinates and fiber optic channels. Step 2: Local submatrix partitioning and clustering discrimination. Based on the principle of spatial proximity, the three-dimensional matrix is ​​divided into overlapping local submatrixes using the sliding window technique. After calculating the representative spectrum of each submatrix, the K-means clustering algorithm is used for unsupervised classification to screen out potential embryonic tissue regions. Step 3: Reverse search and target localization. Construct a spectral difference scoring mechanism to quantify the difference between each candidate region and the background region. Perform reverse search and region growth expansion starting from the region with the greatest difference according to the score ranking. Combine spatial connectivity and spectral consistency criteria to determine key biological target regions. Step 4: Spectral data fusion and preprocessing. Extract the spectra of all fiber optic acquisition points from the located target area, and perform a weighted average according to the signal-to-noise ratio of the spectra to obtain the regional representative spectrum. Apply multi-algorithm fusion preprocessing technology to the regional representative spectrum. Step 5: Feature band selection and classification modeling. The simulated annealing algorithm is used to screen feature bands with the ability to distinguish gender from the full spectrum. The optimized feature wavelength data is input into the support vector machine classification model, and the model hyperparameters are optimized through cross-validation and grid search. Step 6: Sex determination. The optimized support vector machine classification model is used to output the sex determination result of the hatching eggs.

5. The targeted positioning method for non-destructive infrared spectroscopy identification of the sex of hatching eggs as described in claim 4, characterized in that: In step 2, the local submatrix is ​​divided as follows: the sliding window size is set, and the local submatrix is ​​divided in the spatial dimension of the three-dimensional matrix to generate a series of overlapping local submatrixes.

6. The targeted positioning method for non-destructive infrared spectroscopy identification of the sex of hatching eggs as described in claim 4, characterized in that: In step 4, the multi-algorithm fusion spectral preprocessing includes the combined use of absorbance conversion, Savitzky-Golay smoothing, first derivative solving, and standard normal variable transformation.

7. The targeted positioning method for non-destructive infrared spectroscopy identification of the sex of hatching eggs as described in claim 4, characterized in that: In step 5, the simulated annealing algorithm achieves global screening of feature bands by setting temperature reduction rules and state acceptance probability rules; wherein, the temperature is gradually reduced according to a set coefficient, and the state acceptance probability is determined together with the change in classification error and the current temperature.

8. The targeted positioning method for non-destructive infrared spectroscopy identification of the sex of hatching eggs as described in claim 4, characterized in that: In step 5, the hyperparameters include the penalty parameters and kernel function parameters of the support vector machine classification model. The hyperparameters are optimized using a joint strategy that combines grid search and k-fold cross-validation.

9. The targeted positioning method for non-destructive infrared spectroscopy identification of the sex of hatching eggs as described in claim 4, characterized in that: In step 4, the specific method of weighted averaging is as follows: calculate the signal-to-noise ratio of the spectrum of each acquisition point in the target area, determine the weight of the site based on the signal-to-noise ratio value, and perform weighted calculation on all spectra in the target area based on the weight to obtain the regional representative spectrum of the target area.

Images