A method and system for recognizing spectral features of intestinal lesions caused by Salmonella in chickens

Abstract

This application proposes a method and system for recognizing spectral feature patterns of intestinal lesions caused by Salmonella in chickens. The method involves acquiring the original spectrum of chicken intestinal tissue; determining the distortion pattern on the surface of the spectral probe; correcting the original spectrum based on the distortion pattern to obtain a corrected spectrum; matching the corrected spectrum with a standard lesion spectrum and calculating the matching degree between the corrected spectrum and the standard lesion spectrum; and determining whether Salmonella lesions are present in the chicken intestinal tissue based on the spectral matching degree. This application can improve the detection accuracy and reliability of the spectral recognition system in the complex environment of large-scale chicken farms.

Description

Technical Field

[0001] This application relates to the field of spectral feature pattern recognition technology for enteropathogenic diseases of chicken Salmonella, and more specifically, to a method and system for spectral feature pattern recognition of enteropathogenic diseases of chicken Salmonella. Background Technology

[0002] In large-scale chicken farms, a detection method and system based on spectral feature pattern recognition is widely used to promptly detect and control intestinal lesions caused by Salmonella in chickens. This system collects spectral information from chicken intestinal tissue and analyzes its unique patterns to determine the presence of lesions. However, due to the unique environment of chicken farms, the focusing lens at the front end of the spectral probe accumulates an uneven chemical film due to the chemical reaction between cleaning solutions and environmental pollutants. This results in non-uniform, wavelength-dependent distortion of the acquired raw spectral signal, significantly reducing the accuracy of existing spectral recognition systems in early lesion detection and increasing the risk of missed detections. Summary of the Invention

[0003] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a method and system for recognizing spectral feature patterns of chicken Salmonella intestinal lesions. The aim is to solve the problem in large-scale chicken farms where the cumulative degradation of the focusing lens surface at the front end of the spectral probe due to chemical incompatibility with cleaning processes and special environments leads to non-uniform distortion in the spectral signal acquisition and preprocessing stages, which is difficult to completely correct. This causes the recognition model to fail in defining the "characteristic wavelength range" for chicken Salmonella intestinal lesions, ultimately resulting in a significant decrease in the system's recognition accuracy when dealing with chickens with early, mild lesions, increasing the risk of "missed detections."

[0004] In a first aspect, this application discloses a method for recognizing spectral features of enteropathogenic factors in chickens caused by Salmonella enterica, the method comprising: Obtain the raw spectrum of chicken intestinal tissue; Determine the distortion mode on the surface of the spectral probe; Based on this distortion mode, the original spectrum is corrected to obtain the corrected spectrum; The corrected spectrum is matched with the standard lesion spectrum, and the matching degree between the corrected spectrum and the standard lesion spectrum is calculated to obtain the spectral matching degree. Based on the spectral matching degree, it can be determined whether there is chicken salmonella lesions in the chicken intestinal tissue.

[0005] Furthermore, in some implementations, the step of determining the distortion mode of the spectral probe surface includes: Identify the endogenous spectral fingerprint in the original spectrum, which refers to the spectral fingerprint that is inherent in chicken intestinal tissue under normal physiological conditions and changes only slightly when early Salmonella lesions occur; Calculate the characteristic parameters of this endogenous spectral fingerprint, including the peak positions; Based on the characteristic parameters of the intrinsic spectral fingerprint and the characteristic parameters of the spectral fingerprint corresponding to the standard non-lesion spectrum, the distortion mode of the spectral probe surface is determined.

[0006] Furthermore, in some embodiments, the step of matching the corrected spectrum with a standard lesion spectrum and calculating the degree of matching between the corrected spectrum and the standard lesion spectrum to obtain the spectral matching degree includes: Multiple spectral features are extracted from the corrected spectrum, including waveform features, relative intensity features, and peak area features. The corrected spectrum is compared with the standard lesion spectrum to obtain the similarity of the waveform features, the relative intensity features, and the peak area features; Based on the distortion pattern, matching weights are assigned to the waveform feature, the relative intensity feature, and the peak area feature; Based on the matching weight and the similarity, the matching degree between the corrected spectrum and the standard lesion spectrum is calculated to obtain the spectral matching degree.

[0007] As an optional approach, the step of extracting multi-dimensional spectral features from the corrected spectrum, wherein the multi-dimensional spectral features include waveform features, relative intensity features, and peak area features, includes: The curve obtained by taking the second derivative of the corrected spectrum characterizes the waveform features. Extract the intensity of multiple preset wavelengths from the corrected spectrum, and calculate the intensity ratio of multiple specific wavelengths to characterize the relative intensity features; The peak area of ​​the corrected spectrum within the preset wavelength range is calculated to characterize the peak area features.

[0008] Through this technical solution, this application can comprehensively and precisely extract the waveform, relative intensity, and peak area characteristics of the spectrum by calculating the second derivative, intensity ratio, and peak area of ​​the corrected spectrum, providing rich and discriminative information for subsequent matching and lesion judgment.

[0009] Based on the above, this application further proposes that, according to the distortion mode, the step of assigning matching weights to the waveform feature, the relative intensity feature, and the peak area feature includes: Determine the extent to which the distortion mode affects the waveform characteristics, the relative intensity characteristics, and the crest area characteristics; The waveform feature, relative intensity feature, and peak area feature are assigned matching weights based on the degree of influence. The matching weights of the waveform feature, relative intensity feature, and peak area feature are negatively correlated with the degree of influence of the distortion mode.

[0010] In one embodiment, the step of correcting the original spectrum according to the distortion mode to obtain the corrected spectrum includes: The original spectrum was divided into multiple wavelength ranges; Based on this distortion mode, the local distortion characteristics of the spectrum within each wavelength range are calculated. Based on this local distortion characteristic, the original spectrum in each wavelength range is independently adjusted for wavelength, scaled for intensity, and shifted for baseline, thereby correcting the original spectrum; The corrected spectra of all wavelength ranges are stitched together to obtain the corrected spectrum.

[0011] In another implementation, the step of calculating the local distortion characteristics of the spectrum within each wavelength range, based on the distortion mode, includes: For wavelength ranges containing intrinsic spectral fingerprints, evaluate the intensity of the intrinsic spectral fingerprint in each wavelength range; The wavelength range where the intensity of the intrinsic spectral fingerprint is below a preset threshold is defined as the low intensity range. Based on the local distortion characteristics in adjacent wavelength ranges, the local distortion characteristics of the low intensity range are calculated by interpolation or trend extrapolation. The wavelength range in which the intensity of the intrinsic spectral fingerprint is not lower than a preset threshold is defined as the high-intensity range. Based on the difference between the characteristic parameters of the intrinsic spectral fingerprint and the characteristic parameters of the spectral fingerprint corresponding to the standard non-lesion spectrum, and in combination with the distortion mode, the local distortion characteristics of the high-intensity range are calculated. For wavelength ranges that do not contain intrinsic spectral fingerprints, based on the distortion mode and in combination with the non-intrinsic spectral fingerprints within the range, the local distortion characteristics of the range are calculated.

[0012] Preferably, the wavelength range where the intensity of the endogenous spectral fingerprint is below a preset threshold is defined as the low-intensity range. The step of calculating the local distortion characteristics of the low-intensity range based on the local distortion characteristics within adjacent wavelength ranges, using interpolation or trend extrapolation fitting formulas, includes: The wavelength range in which the intensity of the endogenous spectral fingerprint is below a preset threshold is defined as the low intensity range; The degree of uncertainty calculated based on the local distortion characteristics of the adjacent wavelength range; Based on this degree of uncertainty, the weights of the waveform characteristics, relative intensity characteristics, and peak area characteristics of the adjacent wavelength range to the low intensity range are dynamically adjusted to obtain the first weight; Based on this first weight, and combined with the collected information on ammonia concentration and organic dust amount, the parameters of the interpolation or trend extrapolation fitting formula are corrected to obtain the local distortion characteristics of this low-intensity range.

[0013] Based on the above, this application further proposes that, for wavelength ranges that do not contain intrinsic spectral fingerprints, after calculating the local distortion characteristics of the range based on the distortion mode and in combination with the non-intrinsic spectral fingerprints within that range, the following steps are included: The signal-to-noise ratio for identifying this non-endogenous spectral fingerprint; Based on the signal-to-noise ratio, the weight of the non-endogenous spectral fingerprint in calculating the local distortion characteristics is dynamically adjusted to obtain the second weight; Based on this second weight, and combined with the collected information on ammonia concentration and organic dust content, the local distortion characteristics of this range are corrected.

[0014] Secondly, this application also discloses a spectral feature pattern recognition system for enteropathogenic lesions in chickens caused by Salmonella enterica, the system comprising: The raw spectrum acquisition module is used to acquire the raw spectrum of chicken intestinal tissue; The distortion mode determination module is used to determine the distortion mode on the surface of the spectral probe. The spectral correction module is used to correct the original spectrum according to the distortion mode to obtain the corrected spectrum; The matching module is used to match the corrected spectrum with the standard lesion spectrum and calculate the matching degree between the corrected spectrum and the standard lesion spectrum to obtain the spectral matching degree. The lesion assessment module determines whether there is Salmonella lesions in the chicken's intestinal tissue based on the spectral matching degree.

[0015] The technical solution according to the embodiments of this application has at least the following beneficial effects: Through the above technical solution, this application overcomes the problem in the prior art where cumulative degradation of the probe surface leads to non-uniform distortion in the spectral signal acquisition and preprocessing stages, which is difficult to completely correct. This distortion causes the recognition model to fail in defining the "characteristic wavelength range" of chicken Salmonella intestinal lesions, ultimately resulting in a significant decrease in the system's recognition accuracy when facing chickens with early, mild lesions, increasing the risk of "missed detection." This application, through the identification and targeted correction of distortion patterns, significantly improves the detection accuracy and reliability of the spectral recognition system in the complex environment of large-scale chicken farms, especially its ability to identify early lesions. This effectively supports early warning and precise control of chicken Salmonella outbreaks, and has significant practical application value.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0018] Figure 1 This is a flowchart illustrating a method for recognizing spectral features of intestinal lesions caused by Salmonella in chickens, provided in one embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical methods, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] It should be noted that the meaning of "multiple" (or "more than") in the description of the embodiments of this application refers to two or more, and "greater than," "less than," "exceeding," etc. are understood to exclude the number itself, while "above," "below," "within," etc. are understood to include the number itself. If "first," "second," etc. are described, they are only for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated.

[0021] In the description of this application, unless otherwise expressly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution.

[0022] Based on the above, this application proposes a method and system for recognizing spectral features of intestinal lesions caused by Salmonella enterica in chickens. See also... Figure 1 , Figure 1 This is a flowchart illustrating a method for recognizing spectral features of enteropathogenic lesions in chickens according to an embodiment of this application. The method for recognizing spectral features of enteropathogenic lesions in chickens according to an embodiment of this application includes, but is not limited to, steps S110 to S150, which are described below.

[0023] Step S110: Obtain the original spectrum of chicken intestinal tissue; Step S120: Determine the distortion mode of the spectral probe surface; Step S130: Correct the original spectrum according to the distortion mode to obtain the corrected spectrum; Step S140: Match the corrected spectrum with the standard lesion spectrum, and calculate the matching degree between the corrected spectrum and the standard lesion spectrum to obtain the spectral matching degree; Step S150: Determine whether there is chicken salmonella lesion in the chicken intestinal tissue based on the spectral matching degree.

[0024] "Raw spectrum" refers to spectral data collected directly from chicken intestinal tissue without any processing. It contains the tissue's own biochemical information as well as various interferences introduced by the collection environment and equipment.

[0025] "Distortion mode" refers to the distortion pattern of spectral signal caused by factors such as contamination or wear on the surface of the spectral probe. This distortion may manifest as an overall decrease in spectral intensity, enhanced absorption or scattering in specific wavelength regions, and shift in peak position.

[0026] "Corrected spectrum" refers to spectral data that has undergone distortion mode correction. It aims to eliminate or reduce the distortion effects in the original spectrum, making it closer to the true tissue spectrum.

[0027] "Standard lesion spectrum" refers to pre-collected and validated spectral data that represents the typical intestinal lesion state of chicken Salmonella, serving as a benchmark for comparison.

[0028] "Spectral matching degree" is a quantitative indicator that measures the similarity between the corrected spectrum and the standard lesion spectrum. The higher the value, the higher the similarity.

[0029] Firstly, various methods can be used to obtain the raw spectra of chicken intestinal tissue. For example, a portable spectrometer can be used, placing the spectral probe directly in contact with or close to the surface of the chicken intestinal tissue, illuminating it with a built-in light source and collecting the reflected or transmitted spectra. As a preferred implementation, the collected chicken intestinal tissue sample is placed on a sample stage and scanned using a laboratory-grade high-precision spectrometer to obtain more detailed spectral data. In practice, to ensure data consistency, fixed acquisition distances, angles, and integration times are typically set.

[0030] Secondly, determining the distortion mode of the spectral probe surface is one of the key steps in this application. One approach is to perform measurements using a standard reference material with known optical properties (e.g., a white plate with stable reflectivity or a calibration solution with specific absorption peaks) before each spectral acquisition. By comparing the difference between the theoretical spectrum of the standard reference material and the actual measured spectrum, the distortion mode of the probe surface can be inferred. For example, if the measured white plate reflectance spectrum shows an abnormal decrease in intensity or a peak shift in a specific wavelength region, then distortion can be considered to exist in that region. As a specific implementation, the probe is periodically physically inspected, for example, by observing the probe surface under a microscope for scratches, stains, or thin film deposits, and a distortion mode database is established by combining this with historical data analysis.

[0031] Next, the original spectrum is corrected according to the distortion mode to obtain the corrected spectrum. One correction method is to perform a mathematical transformation on the original spectrum based on a pre-established distortion mode model. For example, if the distortion mode shows an intensity attenuation in a specific wavelength region, this attenuation can be compensated by multiplying by a wavelength-dependent correction factor. If the distortion mode indicates a peak shift, it can be adjusted using a wavelength calibration algorithm. As a preferred implementation, a machine learning model is used, taking the original spectrum and the corresponding distortion mode as input, to train the model and output the corrected spectrum. For example, a neural network model can be constructed, taking the original spectrum and distortion mode parameters as input, and outputting the corrected spectrum.

[0032] Subsequently, the corrected spectrum is matched with the standard lesion spectrum, and the degree of matching between the corrected spectrum and the standard lesion spectrum is calculated to obtain the spectral matching degree. One matching method is to use spectral similarity algorithms, such as Euclidean distance, correlation coefficient, or spectral angle mapping (SAM). These algorithms calculate the matching degree by quantifying the geometric distance or shape similarity between the two spectral curves. For example, the smaller the Euclidean distance, the more similar the two spectra are, and the higher the matching degree. As a specific implementation method, characteristic parameters of the spectrum (such as the intensity, position, and full width at half maximum of specific peaks) are extracted, and then the similarity of these characteristic parameters is compared. For example, the intensity ratio of the corrected spectrum and the standard lesion spectrum at multiple key wavelengths can be calculated, and these ratios can be combined to evaluate the matching degree.

[0033] Finally, the presence of Salmonella lesions in the chicken's intestinal tissue is determined based on the spectral matching degree. One method is to set a preset matching degree threshold. If the calculated spectral matching degree is higher than this threshold, it is judged that Salmonella lesions are present; if it is lower than this threshold, it is judged that no lesions are present. For example, by testing a large number of known lesion and non-lesion samples, an optimal threshold can be determined to minimize the false positive rate. As a preferred implementation, a classifier model, such as a support vector machine (SVM) or random forest, is used. The spectral matching degree is used as an input feature to train the classifier model to output the lesion judgment result.

[0034] Traditional spectral recognition systems struggle to accurately compensate for non-uniform, wavelength-dependent spectral distortions and stray light caused by chemical films on the focusing lens surface. This leads to the failure of the recognition model to define the "characteristic wavelength range" for chicken Salmonella intestinal lesions, resulting in a significant decrease in recognition accuracy for early-stage, mildly lesions and increasing the risk of missed detections. This application introduces two key steps: "determining the distortion pattern of the spectral probe surface" and "correcting the original spectrum based on the distortion pattern." This proactively identifies and quantifies the optical degradation of the probe surface, and accordingly compensates and corrects the original spectrum. This method effectively eliminates or significantly reduces non-uniform spectral distortions caused by probe surface contamination, allowing the corrected spectrum to more accurately reflect the biochemical information of chicken intestinal tissue. Therefore, this application significantly improves the recognition accuracy of chicken Salmonella intestinal lesions, especially early-stage lesions, reducing the risk of missed detections and enabling more reliable early warning and precise control in large-scale chicken farms.

[0035] In some embodiments, the step of determining the distortion mode of the spectral probe surface includes: Identify the endogenous spectral fingerprint in the original spectrum, wherein the endogenous spectral fingerprint refers to the spectral fingerprint that is inherent in chicken intestinal tissue under normal physiological conditions and changes only slightly when early Salmonella lesions occur; Calculate the characteristic parameters of the endogenous spectral fingerprint, the characteristic parameters including peak positions; The distortion mode of the spectral probe surface is determined based on the characteristic parameters of the intrinsic spectral fingerprint and the characteristic parameters of the spectral fingerprint corresponding to the standard non-lesion spectrum.

[0036] Identifying endogenous spectral fingerprints in raw spectra refers to using specific spectral analysis algorithms (e.g., peak detection, spectral deconvolution, or machine learning models) to identify specific spectral signals in raw spectra of chicken intestinal tissue that are stable under normal physiological conditions and whose spectral characteristics (e.g., peak position, relative intensity) change only slightly during early Salmonella lesions. These endogenous spectral fingerprints can be understood as the tissue's own "internal references," such as the spectral absorption or scattering characteristics of certain proteins, lipids, water, or hemoglobin in specific wavelength regions. The purpose is to provide a relatively stable benchmark for subsequent determination of distortion patterns.

[0037] Furthermore, the characteristic parameters of the intrinsic spectral fingerprint are calculated, including the peak position. Specifically, after identifying the intrinsic spectral fingerprint, its key spectral features need to be quantified. The peak position is one important characteristic parameter, reflecting the energy state of spectral absorption or scattering of a specific molecule or structure. By processing the intrinsic spectral fingerprint with a peak detection algorithm, the wavelength position of its peak can be accurately determined. In addition to the peak position, other characteristic parameters such as peak intensity, full width at half maximum (FWHM), and peak area can also be included to more comprehensively characterize the intrinsic spectral fingerprint. The purpose is to transform the qualitative information of the intrinsic spectral fingerprint into quantifiable data for comparison and analysis.

[0038] Therefore, the distortion mode of the spectral probe surface is determined based on the characteristic parameters of the intrinsic spectral fingerprint and the characteristic parameters of the spectral fingerprint corresponding to the standard non-pathological spectrum. Specifically, after obtaining the characteristic parameters (e.g., peak position) of the intrinsic spectral fingerprint in the original spectrum, it is compared with the characteristic parameters of the corresponding intrinsic spectral fingerprint in the pre-established standard non-pathological chicken intestinal tissue spectrum. The standard non-pathological spectrum is spectral data collected under ideal conditions, i.e., when the spectral probe is distortion-free and the chicken intestinal tissue is healthy. By comparing the differences between the two, such as the shift of the peak position, the change of the peak intensity, or the waveform broadening, the distortion mode present on the spectral probe surface can be quantified. For example, if the peak position of a certain intrinsic spectral fingerprint is red-shifted or blue-shifted relative to the standard value, it indicates that the spectral probe may have a wavelength calibration deviation; if the peak intensity is generally reduced, it may indicate that there is contamination or optical path attenuation on the probe surface. The purpose is to accurately identify and quantify various distortions introduced by the spectral probe through internal reference, providing a precise basis for subsequent spectral correction.

[0039] Through the above technical solution, this application can significantly improve the accuracy and robustness of determining the surface distortion mode of the spectral probe. Since endogenous spectral fingerprints are inherent to the tissue itself and remain relatively stable in early lesions, they provide a reliable internal calibration benchmark, effectively avoiding errors that may be introduced by external calibration standards. This precise determination of distortion modes ensures the effectiveness of subsequent spectral correction, thus laying a solid foundation for the early and accurate diagnosis of chicken Salmonella enteropathy, improving diagnostic reliability and clinical application value.

[0040] In some preferred embodiments, a specific example is illustrated below. Assume that in chicken intestinal tissue, water molecules have a stable absorption peak at a specific wavelength (e.g., 1450 nm), and the position of this absorption peak changes only slightly during early Salmonella lesions; this can be used as an endogenous spectral fingerprint. First, the raw spectrum of the chicken intestinal tissue is acquired. Next, the absorption peak of water molecules near 1450 nm is identified in the raw spectrum. Then, characteristic parameters of this absorption peak are calculated, such as its precise peak position. Simultaneously, a standard non-pathological chicken intestinal tissue spectrum acquired under ideal distortion-free conditions is pre-stored, in which the standard peak position of the water molecule absorption peak is known. By comparing the difference between the actual peak position of the water molecule absorption peak in the raw spectrum and the standard peak position, for example, if the actual peak position is found to be shifted 2 nm towards longer wavelengths relative to the standard position, a 2 nm wavelength redshift distortion mode can be determined for the spectral probe. Based on this distortion mode, the raw spectrum can be precisely corrected.

[0041] In some embodiments, the step of matching the corrected spectrum with the standard lesion spectrum and calculating the matching degree between the corrected spectrum and the standard lesion spectrum to obtain the spectral matching degree includes: Multiple spectral features are extracted from the corrected spectrum, including waveform features, relative intensity features, and peak area features. The corrected spectrum is compared with the standard lesion spectrum to obtain the similarity of the waveform features, the relative intensity features, and the peak area features; Based on the distortion pattern, matching weights are assigned to the waveform feature, the relative intensity feature, and the peak area feature; The spectral matching degree is obtained by calculating the matching degree between the corrected spectrum and the standard lesion spectrum based on the matching weight and the similarity.

[0042] Specifically, extracting multi-dimensional spectral features from calibrated spectra refers to in-depth analysis of calibrated spectral data to obtain multiple independent or related parameters that comprehensively characterize its properties. Among these, waveform features can be understood as the overall shape of the spectral curve, peak and valley distribution, slope changes, and other macroscopic structural information, aiming to capture morphological differences in the spectrum; relative intensity features refer to the intensity ratio of the spectrum at different wavelengths or wavelength ranges, such as the intensity ratio at a specific wavelength, aiming to reflect the relative content changes of different components or states; peak area features refer to the area covered by a specific peak in the spectrum, aiming to quantify the concentration or degree of reaction of a specific substance. The extraction of these features aims to characterize subtle changes in the spectrum from different perspectives, providing a rich data foundation for subsequent matching.

[0043] Comparing the corrected spectrum with the standard lesion spectrum to obtain the similarity of the waveform features, relative intensity features, and peak area features refers to a quantitative comparison between the extracted features of the corrected spectrum and their corresponding features in the pre-established standard lesion spectrum. For example, the similarity of waveform features can be calculated using algorithms such as correlation coefficient, Euclidean distance, or dynamic time warping (DTW); the similarity of relative intensity features can be measured by ratio difference, percentage deviation, etc.; and the similarity of peak area features can be obtained by area ratio or absolute difference, etc. These similarity values ​​reflect the degree of closeness between the corrected spectrum and the standard lesion spectrum in various dimensions.

[0044] Assigning matching weights to the waveform features, relative intensity features, and peak area features based on the distortion mode means taking into account the potential differences in the degree to which the distortion mode on the spectral probe surface affects different spectral features. For example, some distortions may primarily affect the overall intensity of the spectrum (and thus the relative intensity and peak area features), while having a smaller impact on the waveform features; other distortions may cause peak broadening or shifting (and thus affect the waveform and peak area features). Therefore, based on a predetermined distortion mode, features less affected by distortion are assigned higher matching weights, while features more affected by distortion are assigned lower matching weights, in order to reduce the uncertainty caused by distortion and improve the accuracy of matching.

[0045] Therefore, calculating the spectral matching degree between the corrected spectrum and the standard lesion spectrum based on the matching weights and similarities involves weighted summation or other combined operations of the similarities of spectral features in each dimension with their corresponding matching weights. For example, a weighted average method can be used, multiplying the similarity of each feature by its weight, and then summing all weighted similarities to obtain the final spectral matching degree. This weighted calculation method can more accurately reflect the true matching degree between the corrected spectrum and the standard lesion spectrum, effectively reducing the interference of distortion on the matching results.

[0046] Through the above technical solution, this application can significantly improve the accuracy and reliability of spectral feature pattern recognition for chicken Salmonella intestinal lesions. Especially when there is distortion on the surface of the spectral probe, by weighted matching of spectral features in different dimensions, the negative impact of distortion on the matching results is effectively reduced. This allows for more accurate lesion judgment results even under non-ideal acquisition conditions, thus providing strong technical support for the early, rapid, and accurate diagnosis of chicken Salmonella intestinal lesions.

[0047] In some embodiments, the step of extracting multi-dimensional spectral features from the corrected spectrum, wherein the multi-dimensional spectral features include waveform features, relative intensity features, and peak area features, includes: The curve obtained by taking the second derivative of the corrected spectrum characterizes the waveform features. Extract the intensity of multiple preset wavelengths from the corrected spectrum, and calculate the intensity ratio of multiple specific wavelengths to characterize the relative intensity features; The peak area of ​​the corrected spectrum within the preset wavelength range is calculated to characterize the peak area features.

[0048] Specifically, the curve obtained by taking the second derivative of the corrected spectrum is used to characterize the waveform features. The second derivative can effectively highlight the inflection points and peaks / valleys in the spectral curve, thus more sensitively capturing subtle changes in the spectral shape, which are often closely related to changes in the chemical composition or structure within biological tissues. By analyzing the shape, peak position, and intensity of the second derivative curve, a more refined description of the spectral waveform can be obtained.

[0049] The relative intensity characteristic is characterized by extracting the intensity of the corrected spectrum at multiple preset wavelengths and calculating the ratio between the intensities at these specific wavelengths. The preset wavelengths are typically selected where the absorption or scattering characteristics change significantly during the occurrence of Salmonella lesions in chickens. By calculating the intensity ratio, the effects of baseline drift or overall intensity fluctuations that may occur during spectral acquisition can be effectively eliminated, thus more accurately reflecting the relative spectral changes caused by the lesions.

[0050] In practical applications, the peak area characteristic is characterized by calculating the peak area of ​​the corrected spectrum within a preset wavelength range. The peak area comprehensively reflects the concentration information of a specific substance. When Salmonella infection occurs in the chicken's intestinal tissue, the content of certain biomolecules (such as hemoglobin, lipids, and proteins) changes, leading to changes in the area of ​​their characteristic absorption peaks. By calculating the area of ​​these characteristic peaks, the degree of change in biochemical components caused by the disease can be quantified.

[0051] Through the above technical solution, this application enables comprehensive and refined extraction of spectral features from chicken intestinal tissue. Compared with traditional methods that rely solely on a single spectral feature, this application, by combining waveform features, relative intensity features, and peak area features, can more sensitively capture complex spectral changes caused by Salmonella lesions in chickens, including fine-tuning of spectral shape, changes in the relative content of specific components, and alterations in biomolecule concentration. This significantly improves the accuracy and reliability of spectral feature pattern recognition, providing strong technical support for the early and accurate diagnosis of Salmonella intestinal lesions in chickens.

[0052] In some embodiments, the step of assigning matching weights to waveform features, relative intensity features, and crest area features based on the distortion mode includes: Determine the extent to which the distortion mode affects waveform characteristics, relative intensity characteristics, and crest area characteristics; Matching weights are assigned to waveform features, relative intensity features, and peak area features based on the degree of influence. The matching weights of waveform features, relative intensity features, and peak area features are negatively correlated with the degree of influence of the distortion mode.

[0053] Specifically, determining the degree of influence of distortion modes on waveform features, relative intensity features, and peak area features refers to assessing the stability and reliability of these spectral features under a specific distortion mode. For example, some distortion modes may primarily cause a drift in the spectral baseline, thus having a greater impact on peak area features and a relatively smaller impact on waveform features; other distortion modes may lead to a decrease in spectral resolution, thus significantly affecting the fine structure of waveform features. This degree of influence can be obtained through preliminary experiments, simulations, or modeling analysis based on historical data. Matching weights are assigned to waveform features, relative intensity features, and peak area features according to the degree of influence, and the matching weights of waveform features, relative intensity features, and peak area features are negatively correlated with their degree of influence from the distortion mode. This means that features less affected by the distortion mode have a higher weight in the matching degree calculation; conversely, features more affected by the distortion mode have a lower weight. For example, if the distortion mode has an impact of 0.2 on waveform features, 0.5 on relative intensity features, and 0.8 on peak area features, then the waveform features will have the highest matching weight, and the peak area features will have the lowest matching weight. This negative correlation ensures that when calculating spectral matching, more reliance is placed on spectral features that are more reliable and less distorted under the current distortion conditions.

[0054] The proposed solution first quantifies the specific impact of distortion modes on different spectral features (waveform features, relative intensity features, and peak area features). Then, based on this quantification, it dynamically adjusts the matching weights of each feature according to a negative correlation principle. This solves the problem that simple weight allocation might introduce errors due to the varying sensitivities of different spectral features to distortion modes. It is precisely this refined weight allocation mechanism that effectively highlights those spectral features less affected by distortion and with greater diagnostic value, even in the presence of distortion on the spectral probe surface, thereby improving the accuracy and reliability of the matching.

[0055] The above technical solution enables adaptive weight adjustment based on the degree of influence of distortion modes on different spectral features, effectively reducing the interference of distortion on spectral matching results. This dynamic and intelligent weight allocation method allows the system to more accurately identify the spectral characteristics of chicken intestinal tissue lesions even when there are different degrees or types of distortion on the surface of the spectral probe. This significantly improves the accuracy and robustness of diagnosing chicken Salmonella intestinal lesions and avoids misdiagnosis or missed diagnosis due to distortion.

[0056] As a specific implementation method, it is assumed that during a single spectral acquisition process, slight oil contamination exists on the surface of the spectral probe, causing an overall rise in the spectral baseline, accompanied by a slight scattering effect. Through pre-established models or experimental data analysis, it can be determined that this oil contamination distortion pattern has a significant impact of 0.7 on the peak area feature, a moderate impact of 0.4 on the relative intensity feature, and a minor impact of 0.2 on the waveform feature. According to this implementation method, a higher matching weight (e.g., 0.5) is assigned to the waveform feature, a moderate matching weight (e.g., 0.3) to the relative intensity feature, and a lower matching weight (e.g., 0.2) to the peak area feature. In subsequent matching degree calculations, the waveform feature will contribute more to the final matching degree, while the contribution of the peak area feature will be relatively smaller. Therefore, even if oil contamination distortion causes some distortion to the peak area feature, its low weight effectively suppresses its impact on the final lesion diagnosis result, thus ensuring diagnostic accuracy.

[0057] In some embodiments, the step of correcting the original spectrum according to the distortion mode to obtain the corrected spectrum includes: The original spectrum is divided into multiple wavelength ranges; Based on the distortion mode, for each wavelength range, the local distortion characteristics of the spectrum within that range are calculated; Based on the aforementioned local distortion characteristics, the original spectrum in each wavelength range is independently adjusted for wavelength, scaled for intensity, and shifted for baseline, thereby correcting the original spectrum. The corrected spectra of all wavelength ranges are stitched together to obtain the corrected spectrum.

[0058] Specifically, dividing the original spectrum into multiple wavelength intervals aims to decompose the entire spectrum into smaller, relatively uniform segments, facilitating fine-grained local correction for each segment. These wavelength intervals can be divided according to a preset fixed width, such as every 10 or 20 nanometers; alternatively, they can be adaptively divided based on spectral characteristic points, such as absorption peaks, emission peaks, or inflection points, to ensure relatively consistent spectral characteristics within each interval. Calculating the local distortion characteristics of the spectrum within each wavelength interval based on the distortion mode refers to further analyzing the specific manifestation and influence of the distortion mode within a particular wavelength interval, based on the known overall distortion mode. For example, the overall distortion mode might indicate the presence of a thin film on the probe surface, whose absorption characteristics in the ultraviolet band may differ from its scattering characteristics in the visible light band. Therefore, it is necessary to calculate its unique local distortion parameters for different wavelength intervals. In practical applications, independently adjusting the wavelength, scaling the intensity, and shifting the baseline of the original spectrum within each wavelength interval based on local distortion characteristics is crucial for achieving fine-grained correction. Wavelength adjustment aims to correct wavelength drift caused by optical system dispersion or minor changes in probe position; intensity scaling compensates for signal intensity attenuation or enhancement caused by probe contamination, optical path attenuation, or changes in sample scattering characteristics; baseline shifting eliminates baseline shifts caused by background noise or fluorescence interference. These adjustments are performed independently for each wavelength range, ensuring local adaptability of the correction. Therefore, the corrected spectra of all wavelength ranges are stitched together to obtain the corrected spectrum. The purpose is to recombine the locally finely corrected spectral fragments into a complete and continuous corrected spectrum. The stitching process needs to ensure spectral continuity and smoothness between different ranges, avoiding the introduction of new discontinuities or artifacts at the stitching points. For example, methods such as weighted averaging or curve fitting can be used to achieve a smooth transition.

[0059] This application's solution divides the original spectrum into multiple wavelength intervals and calculates the local distortion characteristics for each interval, thereby enabling a more detailed capture and quantification of the specific impact of surface distortion on the spectral probe across different wavelength ranges. Because distortion modes can exhibit differences across different wavelength intervals, traditional global correction methods struggle to address these issues comprehensively. By independently adjusting the wavelength, scaling the intensity, and shifting the baseline for each wavelength interval, this application accurately compensates for local wavelength drift, signal intensity variations, and baseline shifts caused by distortion, avoiding the "overcorrection" or "undercorrection" problems that can occur with global correction. Finally, by seamlessly stitching these locally corrected spectral segments together, a highly accurate overall corrected spectrum is formed, effectively overcoming the limitations of a single global correction strategy in handling complex local distortions.

[0060] In some preferred embodiments, it is assumed that a non-uniform biofilm contamination exists on the surface of the spectral probe when acquiring the raw spectrum of chicken intestinal tissue. This contamination mainly causes signal attenuation and blue shift in the ultraviolet band, and mainly causes baseline rise and enhanced scattering in the visible band. First, the acquired raw spectrum is divided into multiple wavelength ranges, for example, according to the detection range of the spectrometer, it can be divided into an ultraviolet range (200-400 nm), a visible range (400-700 nm), and a near-infrared range (700-1000 nm). Next, based on a predetermined overall distortion pattern (e.g., a model of the effect of biofilm contamination on the spectrum obtained through calibration experiments), the local distortion characteristics are calculated for the ultraviolet range, which may show a large blue shift and intensity attenuation; for the visible range, the local distortion characteristics are calculated, which may show a significant baseline rise and a small intensity change. Then, based on these local distortion characteristics, the original spectrum in the ultraviolet region is independently adjusted in wavelength (corrected towards the redshift direction) and intensity scaled (increased in intensity), while the original spectrum in the visible region is independently adjusted in baseline shift (downward shift) and intensity scaled (fine-tuned in intensity). Finally, these independently corrected wavelength ranges are stitched together to obtain a corrected spectrum that is accurately compensated across the entire spectral range.

[0061] In some embodiments, the step of calculating the local distortion characteristics of the spectrum within each wavelength range based on the distortion mode includes: For wavelength ranges containing intrinsic spectral fingerprints, evaluate the intensity of the intrinsic spectral fingerprint within each wavelength range; The wavelength range where the intensity of the intrinsic spectral fingerprint is lower than a preset threshold is defined as the low intensity range. Based on the local distortion characteristics in adjacent wavelength ranges, the local distortion characteristics of the low intensity range are calculated by interpolation or trend extrapolation. The wavelength range in which the intensity of the intrinsic spectral fingerprint is not lower than a preset threshold is defined as the high intensity range. Based on the difference between the characteristic parameters of the intrinsic spectral fingerprint and the characteristic parameters of the spectral fingerprint corresponding to the standard non-lesion spectrum, and in combination with the distortion mode, the local distortion characteristics of the high intensity range are calculated. For wavelength ranges that do not contain intrinsic spectral fingerprints, the local distortion characteristics of the range are calculated based on the distortion mode and the non-intrinsic spectral fingerprints within that range.

[0062] Specifically, the intrinsic spectral fingerprint refers to the spectral fingerprint inherent in chicken intestinal tissue under normal physiological conditions, which changes only slightly during the early stages of Salmonella infection. Its intensity can be assessed by calculating the integral area, peak height, or average intensity of the fingerprint region. Preset thresholds can be set based on experimental data or expert experience to distinguish whether the intensity of the intrinsic spectral fingerprint is sufficient as a reliable reference. When the intensity of the intrinsic spectral fingerprint is low, it may be significantly affected by noise or background signals, and directly relying on its features to calculate local distortion characteristics may introduce errors. Therefore, by using interpolation or trend extrapolation, more reliable local distortion characteristic information within adjacent wavelength intervals can be utilized to effectively compensate for the uncertainty of low-intensity interval data. Interpolation can employ linear interpolation, polynomial interpolation, or spline interpolation methods, while trend extrapolation can be based on the changing trends of adjacent intervals for prediction. When the intensity of the intrinsic spectral fingerprint is high, its characteristic parameters (such as peak position, full width at half maximum, peak intensity, etc.) are relatively stable and representative, and can serve as an important basis for calculating local distortion characteristics. At this point, by comparing the differences between these characteristic parameters and the characteristic parameters of the spectral fingerprint corresponding to the standard non-lesion spectrum, and combining this with the overall distortion mode, the local distortion in this high-intensity range can be quantified more accurately. For wavelength ranges that do not contain intrinsic spectral fingerprints, due to the lack of inherent references, calculations can be performed by combining non-intrinsic spectral fingerprints (e.g., spectral features caused by other tissue components or environmental factors) and the overall distortion mode. The use of non-intrinsic spectral fingerprints requires caution and usually needs to be evaluated in conjunction with their signal-to-noise ratio and stability.

[0063] This application's solution effectively addresses the potential accuracy limitations of single calculation methods when processing complex spectral data by employing differentiated local distortion characteristic calculation strategies for different wavelength ranges. Specifically, for wavelength ranges containing intrinsic spectral fingerprints, their intensity is first evaluated, enabling the system to identify the reliability of the intrinsic spectral fingerprint. When the intensity of the intrinsic spectral fingerprint is low, directly relying on its features for calculation may lead to errors. Therefore, by utilizing more reliable local distortion characteristics within adjacent wavelength ranges for interpolation or trend extrapolation, the uncertainty of calculation results in low-intensity ranges can be effectively reduced, thereby improving the robustness of local distortion characteristic calculation in this region. When the intensity of the intrinsic spectral fingerprint is high, its characteristic parameters have high reliability. In this case, by comparing the differences between its characteristic parameters and standard non-lesion spectra, and combining this with the overall distortion mode, accurate quantification of local distortion characteristics in high-intensity ranges can be achieved. Furthermore, for wavelength ranges that do not contain intrinsic spectral fingerprints, this application combines the non-intrinsic spectral fingerprints within that range with the overall distortion mode for calculation, ensuring that all wavelength ranges can obtain reasonable local distortion characteristics and avoiding computational blind spots caused by the lack of intrinsic references.

[0064] In some preferred embodiments, it is assumed that the original spectrum of chicken intestinal tissue needs to be calibrated after acquisition. First, the original spectrum is divided into multiple wavelength intervals. When calculating the local distortion characteristics of these wavelength intervals, the system first identifies whether each interval contains an endogenous spectral fingerprint. For example, in wavelength interval A, an endogenous spectral fingerprint is detected, and its intensity is much higher than a preset threshold, thus it is identified as a high-intensity interval. At this time, the system extracts the peak position, full width at half maximum (FWHM), and other characteristic parameters of the endogenous spectral fingerprint in this interval, and compares them with the characteristic parameters of the corresponding fingerprint in the standard non-lesion spectrum. Combined with the previously determined spectral probe distortion mode, the system accurately calculates the local distortion characteristics of interval A. In wavelength interval B, an endogenous spectral fingerprint is also detected, but its intensity is lower than a preset threshold, thus it is identified as a low-intensity interval. Since the fingerprint signal is weak, direct calculation may be inaccurate. Therefore, the system uses the local distortion characteristics of its adjacent wavelength intervals A and C (assuming that reliable local distortion characteristics have also been calculated for wavelength interval C) to calculate the local distortion characteristics of wavelength interval B through linear interpolation. In wavelength range D, no obvious endogenous spectral fingerprint was detected. Therefore, the system calculates the local distortion characteristics of wavelength range D based on the overall distortion pattern and other non-endogenous spectral fingerprints present in this range (such as background absorption peaks caused by intestinal contents). This divide-and-conquer strategy allows for more accurate and reliable calculation of the local distortion characteristics of each wavelength range, laying a solid foundation for subsequent overall spectral correction.

[0065] In some embodiments, the step of determining the wavelength range where the intensity of the intrinsic spectral fingerprint is below a preset threshold as a low-intensity range, and calculating the local distortion characteristics of the low-intensity range by interpolation or trend extrapolation based on the local distortion characteristics within adjacent wavelength ranges, includes: The wavelength range in which the intensity of the intrinsic spectral fingerprint is below a preset threshold is defined as the low intensity range; The degree of uncertainty calculated based on the local distortion characteristics of the adjacent wavelength range; Based on the degree of uncertainty, the weights of the waveform characteristics, relative intensity characteristics, and peak area characteristics of the adjacent wavelength intervals to the low intensity interval are dynamically adjusted to obtain the first weight; Based on the first weight, and combined with the collected ammonia concentration and organic dust information, the parameters of the interpolation or trend extrapolation fitting formula are corrected to obtain the local distortion characteristics of the low-intensity range.

[0066] Specifically, wavelength ranges where the intensity of the intrinsic spectral fingerprint is below a preset threshold are defined as low-intensity ranges. This indicates that the signal of the intrinsic spectral fingerprint is weak within these ranges, and directly extracting distortion information from them may be inaccurate. Therefore, it is necessary to rely on information from adjacent wavelength ranges for inference. The degree of uncertainty calculated based on the local distortion characteristics of adjacent wavelength ranges refers to an assessment of the data quality or reliability of the adjacent wavelength ranges used for interpolation or trend extrapolation. This degree of uncertainty can be quantified based on various indicators, such as the signal-to-noise ratio of the spectral signals in adjacent ranges, data volatility, or the degree of deviation from the standard spectrum. The higher the degree of uncertainty, the less reliable the data in the adjacent range. In practical applications, the weights of the waveform characteristics, relative intensity characteristics, and peak area characteristics of the low-intensity range in adjacent wavelength ranges are dynamically adjusted according to the degree of uncertainty to obtain the first weight. This means that adjacent ranges with lower uncertainty (i.e., more reliable data) will be assigned a higher first weight when calculating the local distortion characteristics of the low-intensity range; conversely, adjacent ranges with higher uncertainty will be assigned a lower first weight. This dynamic adjustment ensures that the interpolation or trend extrapolation process prioritizes the use of more reliable data sources. Furthermore, based on the first weight and combined with the collected ammonia concentration and organic dust amount information, the parameters of the interpolation or trend extrapolation fitting formula are corrected to obtain the local distortion characteristics in the low-intensity range. Ammonia concentration and organic dust amount are important factors affecting the spectral acquisition environment, which may lead to surface contamination of the spectral probe or enhanced scattering effects, thereby introducing additional distortion. By taking these environmental parameters into account, the coefficients or parameters of the mathematical models for interpolation or trend extrapolation (e.g., linear interpolation, polynomial fitting, etc.) can be finely adjusted to more accurately reflect the actual local distortion situation.

[0067] This application's solution introduces an uncertainty assessment of the calculation of local distortion characteristics in adjacent wavelength ranges and dynamically adjusts the weights accordingly. This allows for the prioritization of information from adjacent ranges with higher data quality and greater reliability when inferring distortion characteristics in low-intensity ranges, effectively reducing errors caused by data source uncertainty. Furthermore, by incorporating environmental information such as ammonia concentration and organic dust levels, the parameters of the interpolation or trend extrapolation fitting formulas can be corrected. This is because these environmental factors directly affect the transmission and reception of spectral signals, potentially leading to surface contamination of the spectral probe or enhanced scattering effects, thus introducing additional distortion. By incorporating these external environmental disturbances into the model correction, the calculation of distortion characteristics can more comprehensively reflect actual operating conditions, improving adaptability to complex environments.

[0068] In some embodiments, the step of calculating the local distortion characteristics of a wavelength range that does not contain an intrinsic spectral fingerprint, based on the distortion mode and in conjunction with the non-intrinsic spectral fingerprint within that range, includes the following after the step: The signal-to-noise ratio of the non-endogenous spectral fingerprint is identified; Based on the signal-to-noise ratio, the weight of the non-endogenous spectral fingerprint in calculating the local distortion characteristics is dynamically adjusted to obtain the second weight; Based on the second weight, and combined with the collected information on ammonia concentration and organic dust content, the local distortion characteristics of this range are corrected.

[0069] Specifically, the signal-to-noise ratio (SNR) for identifying non-endogenous spectral fingerprints refers to quantifying the relative levels of effective signal and noise in non-endogenous spectral fingerprints using signal processing techniques. SNR is a crucial indicator of spectral data quality; a higher SNR indicates more reliable data, while a lower SNR may suggest greater noise interference. Its purpose is to provide a data foundation for subsequent weight adjustments, ensuring the ability to distinguish reliable spectral information from noise when calculating local distortion characteristics.

[0070] The second weight, derived by dynamically adjusting the weight of non-endogenous spectral fingerprints in calculating local distortion characteristics based on the signal-to-noise ratio (SNR), can be understood as adjusting the influence of non-endogenous spectral fingerprints on the calculation of local distortion characteristics in real time according to the identified SNR. For example, when the SNR is high, non-endogenous spectral fingerprints are given a higher weight, indicating that their information is more reliable; conversely, when the SNR is low, they are given a lower weight to reduce the negative impact of noise on the calculation results. The aim is to optimize the contribution of non-endogenous spectral fingerprints in the calculation of local distortion characteristics, reduce noise interference, and improve the accuracy of the calculation.

[0071] In practical applications, based on the second weight and combined with the collected information on ammonia concentration and organic dust content, the local distortion characteristics in this range are corrected. Specifically, this means that, after considering the reliability of the non-endogenous spectral fingerprint (reflected by the second weight), environmental factors (such as ammonia concentration and organic dust content) are further introduced to refine the local distortion characteristics. For example, ammonia and organic dust may absorb or scatter within specific wavelength ranges, thus affecting the spectral signal. By monitoring these environmental parameters in real time and combining them with a model of their impact on spectral distortion, the initially calculated local distortion characteristics can be compensated or corrected. The aim is to eliminate or reduce the interference of environmental factors on the calculation of spectral distortion characteristics, making the corrected spectrum closer to reality.

[0072] This application's solution quantifies the data quality by introducing the identification of the signal-to-noise ratio (SNR) of non-endogenous spectral fingerprints, thus avoiding blind reliance on potentially noise-contaminated spectral information when calculating local distortion characteristics. The introduction of the SNR allows for dynamic adjustment of the weights of non-endogenous spectral fingerprints based on their levels, ensuring that high-quality spectral data plays a greater role in distortion characteristic calculations, while effectively suppressing the influence of low-quality data. Furthermore, by combining this with collected ammonia concentration and organic dust information, targeted corrections can be made to spectral distortions caused by environmental factors. For example, when ammonia concentration is high, its absorption effect on specific bands can be predicted, and local distortion characteristics can be adjusted accordingly, resulting in more accurate and robust calculated local distortion characteristics.

[0073] This application also discloses a spectral feature pattern recognition system for intestinal lesions caused by Salmonella in chickens, the system comprising: The raw spectrum acquisition module is used to acquire the raw spectrum of chicken intestinal tissue; The distortion mode determination module is used to determine the distortion mode on the surface of the spectral probe. A spectral correction module is used to correct the original spectrum according to the distortion mode to obtain a corrected spectrum. The matching module is used to match the corrected spectrum with the standard lesion spectrum and calculate the matching degree between the corrected spectrum and the standard lesion spectrum to obtain the spectral matching degree. The lesion assessment module determines whether there is Salmonella lesions in the chicken intestinal tissue based on the spectral matching degree.

[0074] Specifically, the raw spectral acquisition module can be configured to include a spectral probe and a data acquisition unit. The spectral probe can be a portable spectrometer probe designed to be placed directly in contact with or at close range on the surface of chicken intestinal tissue, illuminating it with a built-in light source and acquiring reflected or transmitted spectra. The data acquisition unit is responsible for converting the analog signals acquired by the probe into digital spectral data. Alternatively, the raw spectral acquisition module can be designed to integrate with a laboratory-grade high-precision spectrometer to scan chicken intestinal tissue samples placed on a sample stage to obtain more detailed spectral data. In practical applications, this module is typically set with fixed acquisition distances, angles, and integration times to ensure data consistency.

[0075] One implementation of the distortion mode determination module involves configuring the system to perform measurements using a standard reference material with known optical properties (e.g., a white plate with stable reflectivity or a calibration solution with specific absorption peaks) before each spectral acquisition. By comparing the theoretical spectrum of the standard reference material with the actual measured spectrum, the module can infer the distortion mode of the probe surface. For example, if the measured white plate reflectance spectrum shows an abnormal decrease in intensity or a peak shift in a specific wavelength region, distortion can be considered to exist in that region. Another approach is to integrate an image recognition or sensor unit to periodically perform physical inspections of the probe, such as microscopic examination of the probe surface for scratches, stains, or thin film deposits, and combine this with historical data analysis to build a distortion mode database.

[0076] The spectral correction module corrects the original spectrum based on the distortion mode, resulting in a corrected spectrum. One correction method involves configuring the module to perform a mathematical transformation on the original spectrum based on a pre-established distortion mode model. For example, if the distortion mode indicates intensity attenuation in a specific wavelength region, the module can compensate for this attenuation by multiplying by a wavelength-dependent correction factor. If the distortion mode indicates a peak shift, the module can adjust using a wavelength calibration algorithm. Another correction method integrates a machine learning processing unit, taking the original spectrum and the corresponding distortion mode as input to train a model that outputs the corrected spectrum. For example, a neural network model can be constructed, taking the original spectrum and distortion mode parameters as input, and outputting the corrected spectrum.

[0077] The matching module is used to match the corrected spectrum with the standard lesion spectrum, calculating the spectral matching degree. One matching method is to configure the module to use spectral similarity algorithms, such as Euclidean distance, correlation coefficient, or spectral angle mapping (SAM). These algorithms calculate the matching degree by quantifying the geometric distance or shape similarity between the two spectral curves. For example, the smaller the Euclidean distance, the more similar the two spectra are, and the higher the matching degree. Another matching method is to configure the module to extract spectral feature parameters (such as the intensity, position, and full width at half maximum of specific peaks) and then compare the similarity of these feature parameters. For example, the intensity ratios of the corrected spectrum and the standard lesion spectrum at multiple key wavelengths can be calculated, and these ratios can be combined to evaluate the matching degree.

[0078] The lesion detection module determines the presence of Salmonella lesions in chicken intestinal tissue based on spectral matching scores. One approach is to configure the module to set a preset matching score threshold. If the calculated spectral matching score is higher than this threshold, Salmonella lesions are identified; if it is lower, no lesions are identified. For example, by testing a large number of known lesion and non-lesion samples, an optimal threshold can be determined to minimize the false positive rate. Another approach is to integrate a classifier model, such as a Support Vector Machine (SVM) or Random Forest, into the module. The spectral matching score is used as input features to train the classifier model and output the lesion detection result.

[0079] The spectral feature pattern recognition system for chicken Salmonella intestinal lesions proposed in this application works by introducing the identification and correction of distortion patterns on the surface of the spectral probe. This effectively solves the problem of spectral signal distortion caused by environmental factors affecting diagnostic accuracy in traditional methods. Specifically, after the raw spectrum of chicken intestinal tissue is acquired by the raw spectrum acquisition module, the system does not directly analyze the raw spectrum. Instead, it first determines the distortion pattern on the surface of the spectral probe through the distortion pattern determination module, quantifying the signal distortion caused by probe contamination or wear. Subsequently, the spectral correction module precisely corrects the raw spectrum based on this distortion pattern, thereby obtaining a corrected spectrum that more realistically and accurately reflects the biochemical state of chicken intestinal tissue. This correction step is crucial, as it eliminates or significantly reduces the non-uniform, wavelength-dependent spectral distortion and stray light effects caused by the non-uniform chemical film on the probe surface, ensuring that subsequent lesion identification is no longer interfered with by these systematic errors. Next, the matching module matches the corrected spectrum with a pre-established standard lesion spectrum and calculates the degree of matching between the two. Because the accuracy of the calibrated spectrum is guaranteed, the matching degree can more reliably reflect the presence of Salmonella lesions in chicken intestinal tissue. Ultimately, the lesion judgment module can make an accurate lesion judgment based on the calculated spectral matching degree. The entire system forms a closed loop, solving the problem of spectral signal distortion at its source, ensuring the accuracy and reliability of subsequent identification, thereby effectively improving the ability to provide early warning and precise control of chicken Salmonella intestinal lesions.

[0080] The spectral feature pattern recognition system for chicken Salmonella intestinal lesions proposed in this application has significant advantages and innovations compared with existing technologies. Traditional spectral recognition systems struggle to accurately and effectively compensate for non-uniform, wavelength-dependent spectral distortions and stray light caused by chemical films on the focusing lens surface. This leads to the failure of the recognition model's definition of the "characteristic wavelength range" for chicken Salmonella intestinal lesions, ultimately resulting in a significant decrease in recognition accuracy when dealing with chickens with early, mild lesions, increasing the risk of missed detections. The system in this application, by introducing a distortion pattern determination module and a spectral correction module, can actively identify and quantify the optical degradation of the probe surface, and accordingly perform targeted compensation and correction of the original spectrum. This systematic approach can effectively eliminate or significantly reduce non-uniform spectral distortions caused by probe surface contamination, allowing the corrected spectrum to more accurately reflect the biochemical information of chicken intestinal tissue. Therefore, the system of this application can significantly improve the accuracy of identifying enteropathogenic lesions of chickens, especially early lesions, and reduce the risk of missed detection, thereby achieving more reliable early warning and precise prevention and control in large-scale chicken farms.

[0081] The foregoing has provided a detailed description of the preferred embodiments of this application. However, this application is not limited to the above-described embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined in this application.

Claims

1. A method for recognizing spectral feature patterns of intestinal lesions caused by Salmonella in chickens, characterized in that, include: Obtain the raw spectrum of chicken intestinal tissue; Determine the distortion mode on the surface of the spectral probe; The original spectrum is corrected according to the distortion mode to obtain the corrected spectrum; The corrected spectrum is matched with the standard lesion spectrum, and the matching degree between the corrected spectrum and the standard lesion spectrum is calculated to obtain the spectral matching degree. Based on the spectral matching degree, it can be determined whether there is chicken salmonella lesions in the chicken intestinal tissue.

2. The method according to claim 1, characterized in that, The step of determining the distortion mode of the spectral probe surface includes: Identify the endogenous spectral fingerprint in the original spectrum, wherein the endogenous spectral fingerprint refers to the spectral fingerprint that is inherent in chicken intestinal tissue under normal physiological conditions and changes only slightly when early Salmonella lesions occur; Calculate the characteristic parameters of the endogenous spectral fingerprint, the characteristic parameters including peak positions; The distortion mode of the spectral probe surface is determined based on the characteristic parameters of the intrinsic spectral fingerprint and the characteristic parameters of the spectral fingerprint corresponding to the standard non-lesion spectrum.

3. The method according to claim 1, characterized in that, The step of matching the corrected spectrum with the standard lesion spectrum and calculating the matching degree between the corrected spectrum and the standard lesion spectrum to obtain the spectral matching degree includes: Multiple spectral features are extracted from the corrected spectrum, including waveform features, relative intensity features, and peak area features. The corrected spectrum is compared with the standard lesion spectrum to obtain the similarity of the waveform features, the relative intensity features, and the peak area features; Based on the distortion pattern, matching weights are assigned to the waveform feature, the relative intensity feature, and the peak area feature; The spectral matching degree is obtained by calculating the matching degree between the corrected spectrum and the standard lesion spectrum based on the matching weight and the similarity.

4. The method according to claim 3, characterized in that, The step of extracting multi-dimensional spectral features from the corrected spectrum, wherein the multi-dimensional spectral features include waveform features, relative intensity features, and peak area features, includes: The curve obtained by taking the second derivative of the corrected spectrum characterizes the waveform features. Extract the intensity of multiple preset wavelengths from the corrected spectrum, and calculate the intensity ratio of multiple specific wavelengths to characterize the relative intensity features; The peak area of ​​the corrected spectrum within the preset wavelength range is calculated to characterize the peak area features.

5. The method according to claim 3, characterized in that, The step of assigning matching weights to the waveform feature, the relative intensity feature, and the peak area feature based on the distortion mode includes: Determine the degree of influence of the distortion mode on the waveform characteristics, the relative intensity characteristics, and the peak area characteristics; The waveform feature, the relative intensity feature, and the peak area feature are assigned matching weights based on the degree of influence, wherein the matching weights of the waveform feature, the relative intensity feature, and the peak area feature are negatively correlated with the degree of influence of the distortion mode.

6. The method according to claim 1, characterized in that, The step of correcting the original spectrum according to the distortion mode to obtain the corrected spectrum includes: The original spectrum is divided into multiple wavelength ranges; Based on the distortion mode, for each wavelength range, the local distortion characteristics of the spectrum within that range are calculated; Based on the aforementioned local distortion characteristics, the original spectrum in each wavelength range is independently adjusted for wavelength, scaled for intensity, and shifted for baseline, thereby correcting the original spectrum. The corrected spectra of all wavelength ranges are stitched together to obtain the corrected spectrum.

7. The method according to claim 6, characterized in that, The step of calculating the local distortion characteristics of the spectrum within each wavelength range based on the distortion mode includes: For wavelength ranges containing intrinsic spectral fingerprints, evaluate the intensity of the intrinsic spectral fingerprint within each wavelength range; The wavelength range where the intensity of the intrinsic spectral fingerprint is lower than a preset threshold is defined as the low intensity range. Based on the local distortion characteristics in adjacent wavelength ranges, the local distortion characteristics of the low intensity range are calculated by interpolation or trend extrapolation. The wavelength range where the intensity of the intrinsic spectral fingerprint is not lower than a preset threshold is defined as the high-intensity range. Based on the difference between the characteristic parameters of the intrinsic spectral fingerprint and the characteristic parameters of the spectral fingerprint corresponding to the standard non-lesion spectrum, and in combination with the distortion mode, the local distortion characteristics of the high-intensity range are calculated. For wavelength ranges that do not contain intrinsic spectral fingerprints, based on the distortion mode and in combination with the non-intrinsic spectral fingerprints within the range, the local distortion characteristics of the range are calculated.

8. The method according to claim 7, characterized in that, The step of determining the wavelength range where the intensity of the intrinsic spectral fingerprint is below a preset threshold as a low-intensity range, and calculating the local distortion characteristics of the low-intensity range by interpolation or trend extrapolation based on the local distortion characteristics within adjacent wavelength ranges, includes: The wavelength range in which the intensity of the intrinsic spectral fingerprint is below a preset threshold is defined as the low intensity range; The degree of uncertainty calculated based on the local distortion characteristics of the adjacent wavelength range; Based on the degree of uncertainty, the weights of the waveform characteristics, relative intensity characteristics, and peak area characteristics of the adjacent wavelength intervals to the low intensity interval are dynamically adjusted to obtain the first weight; Based on the first weight, and combined with the collected ammonia concentration and organic dust information, the parameters of the interpolation or trend extrapolation fitting formula are corrected to obtain the local distortion characteristics of the low-intensity range.

9. The method according to claim 7, characterized in that, The step of calculating the local distortion characteristics of a wavelength range that does not contain intrinsic spectral fingerprints, based on the distortion mode and in conjunction with the non-intrinsic spectral fingerprints within that range, includes the following: The signal-to-noise ratio of the non-endogenous spectral fingerprint is identified; Based on the signal-to-noise ratio, the weight of the non-endogenous spectral fingerprint in calculating the local distortion characteristics is dynamically adjusted to obtain the second weight; Based on the second weight, and combined with the collected information on ammonia concentration and organic dust content, the local distortion characteristics of this range are corrected.

10. A system for recognizing spectral features of intestinal lesions caused by Salmonella in chickens, characterized in that, The system includes: The raw spectrum acquisition module is used to acquire the raw spectrum of chicken intestinal tissue; The distortion mode determination module is used to determine the distortion mode on the surface of the spectral probe. A spectral correction module is used to correct the original spectrum according to the distortion mode to obtain a corrected spectrum. The matching module is used to match the corrected spectrum with the standard lesion spectrum and calculate the matching degree between the corrected spectrum and the standard lesion spectrum to obtain the spectral matching degree. The lesion assessment module determines whether there is Salmonella lesions in the chicken intestinal tissue based on the spectral matching degree.

Images