A method for evaluating bird's nest and a system for removing impurities from bird's nest during the impurity removal process.

Abstract

This invention discloses a method and system for evaluating bird's nests during the impurity removal process, belonging to the field of bird's nest testing. Addressing the problems of easy adulteration during the existing bird's nest preparation process and the difficulty of detection during processing, this invention provides a method for evaluating bird's nests during the impurity removal process. The method includes obtaining a first bird's nest sample based on an impurity sample; constructing a first detection model to test the first bird's nest sample and obtaining a first detection result of the component characteristics of the first bird's nest sample; obtaining a second bird's nest sample based on a clean bird's nest sample and constructing a second detection model to test the second bird's nest sample and obtaining a second detection result of the physical characteristics of the second bird's nest sample; and obtaining an evaluation result of the bird's nest based on the first and second detection results. This invention uses the first and second detection models to perform targeted detection on impurity samples and clean bird's nests respectively, thereby increasing the comprehensiveness and randomness of bird's nest sampling while reducing bird's nest loss and outputting reliable bird's nest evaluation results. This greatly ensures the accuracy and reliability of bird's nest testing, without affecting the processing efficiency of bird's nests.

Description

Technical Field

[0001] This invention belongs to the field of bird's nest processing technology, and more specifically, relates to a bird's nest evaluation method and a bird's nest impurity removal system in the bird's nest impurity removal process. Background Technology

[0002] As a precious tonic, the identification of genuine bird's nest and its quality assessment have always been major challenges for the industry. Traditional manual testing methods suffer from high subjectivity, low efficiency, and difficulty in meeting the demands of large-scale production. Furthermore, raw bird's nest often contains various impurities such as feathers and soil, severely affecting product quality. Meanwhile, adulterated bird's nest is rampant in the market, using chemical additives to imitate genuine products, making identification extremely difficult. In addition, key quality indicators such as sialic acid content require complex chemical analysis for accurate measurement, which is time-consuming and labor-intensive. These factors lead to inefficiency and difficulty in quality control at all stages of the bird's nest supply chain, failing to meet market demand for high-quality products. How to achieve automated detection and identification of bird's nest while ensuring accuracy, and improve production efficiency and quality control, is a core technical challenge that the bird's nest industry urgently needs to solve. Summary of the Invention

[0003] To address the problems of adulteration in the existing bird's nest production process and the difficulty of detection during processing, this invention provides a bird's nest evaluation method and system for the impurity removal process. This invention removes impurities from initial bird's nests to obtain clean bird's nest and impurity samples. A first detection model and a second detection model are then used to perform targeted detection on the impurity samples and clean bird's nest, respectively. This approach increases the comprehensiveness and randomness of bird's nest sampling while reducing losses, and outputs reliable bird's nest evaluation results, greatly ensuring the accuracy and reliability of bird's nest detection without affecting processing efficiency.

[0004] To solve the above problems, the present invention adopts the following technical solution.

[0005] A method for evaluating bird's nest in the process of removing impurities from bird's nest involves removing impurities from the initial bird's nest to obtain clean bird's nest and impurity samples.

[0006] The impurity sample was screened and filtered to obtain the first bird's nest sample;

[0007] A first detection model is constructed, and the first bird's nest sample is detected based on the first detection model to obtain a first detection result, wherein the first detection result includes the detection result of the component characteristics of the first bird's nest sample;

[0008] Obtain a second bird's nest sample from the cleaned bird's nest;

[0009] A second detection model is constructed, and the second bird's nest sample is detected based on the second detection model to obtain a second detection result, which includes the detection results of the physical characteristics of the second bird's nest sample.

[0010] The first and second test results are analyzed and processed to obtain the evaluation results of the bird's nest.

[0011] By employing the above technical solution, a first test sample is obtained using impurity samples, thereby enabling the evaluation of the compositional characteristics of bird's nest based on the test results of the first test sample. Simultaneously, impurity samples that would otherwise be discarded can be reused, saving on bird's nest losses during testing and reducing the cost of bird's nest waste during the testing process. Furthermore, since impurities in bird's nest typically exhibit random and widespread distribution, the sampling of the first bird's nest sample can be completed more comprehensively, avoiding the risk of missed detections due to insufficient sampling quantity or poor randomness, thus improving the reliability of the first bird's nest sample test.

[0012] Furthermore, since the first bird's nest sample obtained through the impurity sample contains a large amount of impurities, it affects the accuracy of the first bird's nest sample detection. Therefore, a second detection model is used to detect the second bird's nest sample obtained from the clean bird's nest and evaluate its physical characteristics to obtain a second detection result. Finally, through the analysis and processing of the first and second detection results, the obtained bird's nest evaluation result has extremely high comprehensiveness, reliability and accuracy.

[0013] Separating the two inspections can reduce the cost of building a single inspection model and improve inspection efficiency, making it suitable for quality monitoring in mass production processes.

[0014] In some embodiments, the impurity treatment of the initial bird's nest includes:

[0015] Obtain the original image of the initial bird's nest;

[0016] The original image is preprocessed to obtain the first image of the bird's nest;

[0017] An impurity identification model is used to identify the first image and mark the impurity locations and the corresponding de-noising regions; the de-noising regions completely cover the impurity locations.

[0018] Vacuum adsorption is used to adsorb all the impurity removal areas and obtain the impurity sample;

[0019] The remaining initial bird's nest is the clean bird's nest.

[0020] By adopting the above technical solution, the location of impurities can be accurately identified and located, and the impurity removal area can completely cover the impurity location. Thus, when the impurity removal area is removed by vacuum adsorption, not only can the impurities be completely removed to ensure the quality of the remaining clean bird's nest, but a certain amount of bird's nest will inevitably be carried along, so that the impurity sample contains a certain amount of bird's nest, thereby achieving cost characteristic detection of the first bird's nest sample.

[0021] In some implementations, obtaining the original image of the initial bird's nest includes:

[0022] Obtain dried bird's nest and soak and clean it;

[0023] The processed dried bird's nest is manually laid on a sorting tray and then manually de-feathered and de-impure to obtain the initial bird's nest;

[0024] The sorting tray containing the initial bird's nest is placed in an automatic sorting device;

[0025] The automatic sorting device provides supplemental lighting from the bottom of the sorting tray and uses a high-resolution camera to acquire the original image of the initial bird's nest from above the sorting tray.

[0026] Using the above technical solution, a large number of impurities in dried bird's nest can be accurately removed manually without causing damage to the bird's nest itself. Then, the initial bird's nest is laid on a sorting tray to obtain the original image, so that small impurities that are not easily detected or ignored by humans can be identified from the original image, so as to obtain impurity samples in the future.

[0027] In some implementations, constructing a first detection model includes:

[0028] According to the first ratio range, pure bird's nest and standard impurities were mixed to obtain training samples;

[0029] Chemical composition data of bird's nest samples were obtained using hyperspectral imaging technology, and a multidimensional feature matrix containing sialic acid content and glycoprotein structure was generated.

[0030] A convolutional neural network model is used to preprocess the multidimensional feature matrix to generate a first standard feature set, which includes feature vectors of sialic acid content and glycoprotein structure.

[0031] Based on the first standard feature set, the principal component analysis algorithm is applied to reduce the dimensionality, retaining the main features of sialic acid content and glycoprotein structure, and obtaining a low-dimensional feature vector.

[0032] Based on the low-dimensional feature vector, a first detection model is constructed to identify the similarity between the target sample and pure bird's nest and standard impurities mixed in a first ratio.

[0033] Using the above technical solution, pure bird's nest and standard impurities are mixed in a first ratio range to obtain training samples. Then, a convolutional neural network model is used to learn from the training samples, so that the first detection model can more accurately identify the chemical composition characteristics of pure bird's nest containing impurities. On the other hand, the first detection model only retains the identification and training of the main features of bird's nest sialic acid content and glycoprotein structure, thereby greatly reducing the training difficulty of the first detection model and reducing the construction cost of the first detection model.

[0034] In some embodiments, the first proportion ranges from 5% to 30%, and the standard impurities are obtained through a manual impurity removal process for bird's nest, including eggshells, feathers, and / or dust particles.

[0035] The raw spectral data of bird's nest samples were obtained by using hyperspectral imaging technology, and band selection algorithms were used to extract bands related to chemical components to obtain a preliminary spectral feature set.

[0036] Based on the preliminary spectral feature set, principal component analysis algorithm is used to separate the features related to sialic acid content and glycoprotein structure to generate a chemical component feature subset; for the chemical component feature subset, matrix splicing method is used to integrate the data of sialic acid content and glycoprotein structure to generate a multidimensional feature matrix;

[0037] A convolutional neural network is used to preprocess the multidimensional feature matrix, extracting local features using convolutional and pooling layers to obtain a first feature matrix. If the dimension of the first feature matrix exceeds a preset threshold, principal component analysis (PCA) is used to reduce its dimensionality, resulting in a second feature matrix. Based on the second feature matrix, a deep convolutional neural network is used to extract features related to sialic acid content and glycoprotein structure, generating feature vectors.

[0038] The feature vectors are normalized using a standardization algorithm, and the first standard feature set is generated by using the z-score standardization method.

[0039] By employing the aforementioned technical features, on the one hand, when the impurity content in pure bird's nest is too high, it is easily removed manually, and excessive impurities can easily prevent the first detection model from accurately identifying the component characteristics of the bird's nest in the training samples. On the other hand, if the impurity content in pure bird's nest is too low, the impurities have less impact on the detection results, and the analysis model for pure bird's nest can be used directly for detection without the need to construct an additional first detection model. Therefore, determining the first proportion range can meet the detection requirements of the first detection sample.

[0040] In some implementations, obtaining the first detection result includes:

[0041] The sialic acid content and glycoprotein characteristic spectrum were obtained from the first bird's nest sample using hyperspectral imaging technology to generate the first characteristic dataset;

[0042] The first feature dataset is identified, and the first detection result is generated based on the similarity of the features;

[0043] The first test result includes the adulteration risk level of the first bird's nest sample, which includes low risk, medium risk and high risk.

[0044] In some implementations, the detection process of the first detection model includes:

[0045] The proportion of impurities in the first bird's nest sample is estimated. Based on the first image, the first total area of ​​all impurities and the second total area of ​​all impurity-removed areas are calculated. The proportion of impurities is equal to the first total area and the second total area.

[0046] Based on the impurity ratio, the first detection model uses a low-dimensional feature vector whose first ratio is closest to the impurity ratio to perform similarity recognition in order to generate the first detection result.

[0047] When the similarity is greater than 90%, the output adulteration risk level is low.

[0048] When the similarity is less than or equal to 90% and greater than or equal to 70%, the output adulteration risk level is medium risk.

[0049] When the similarity is less than 70%, the output adulteration risk level is high.

[0050] Using the above technical solution, the adulteration risk level is set as low risk, medium risk, and high risk. Since the cost and proportion of impurities in the actual first bird's nest sample deviate from the cost and proportion of standard impurities in the first detection model, when the similarity deviation is less than 10%, it can be judged as low risk; when the similarity deviation is greater than 30%, the deviation far exceeds the impact of impurities, so it can be judged as high risk; in other cases, it can be judged as medium risk, which can be used as a reference for the bird's nest evaluation results.

[0051] In some implementations, the location and quantity of the second bird's nest sample are determined based on the adulteration risk level in the first detection result;

[0052] When the adulteration risk level is medium or high, bird's nest samples are obtained from adjacent positions of each of the impurity areas to obtain the second bird's nest sample;

[0053] When the adulteration risk level is low, bird's nest samples are randomly obtained from the clean bird's nest to obtain a first number of second bird's nest samples, wherein the first number is 10% to 30% of the number of first bird's nest samples.

[0054] Using the above technical solution, when the adulteration risk level is medium or high, it indicates that there is a high probability of adulterants in the impurity area. Therefore, it is necessary to obtain bird's nest samples from adjacent positions of each impurity area to obtain the second bird's nest sample. Thus, the second detection model is used for accurate detection and identification to ensure the accuracy and reliability of the bird's nest evaluation results.

[0055] When the adulteration risk level is low, it means that no adulterants are present in the impurity area. Therefore, a second bird's nest sample can be randomly obtained from the clean bird's nest for a second test to enhance the accuracy and reliability of the bird's nest evaluation results. At the same time, since pure bird's nest has a high value, the first quantity is set to 10% to 30% of the first bird's nest sample quantity to reduce the loss of the second bird's nest sample.

[0056] Furthermore, the physical properties of the second bird's nest sample are tested, which greatly reduces the damage to the second bird's nest sample during testing, so that the second bird's nest sample can be reused or collected and processed as a secondary product.

[0057] In some implementations, a second detection model is constructed, including:

[0058] Reference values ​​of physical characteristics are obtained from a pre-established database of standard parameters for authentic bird's nests, and a second standard feature set is generated; the second standard feature set includes feature vectors of fiber structure and density distribution.

[0059] A second detection model is constructed based on the second standard feature set to identify whether the target sample is genuine bird's nest.

[0060] In some implementations, the analysis and processing of the first and second detection results includes:

[0061] Calculate the first deviation value D1 of the component characteristics of the first bird's nest sample in the first test result, and the second deviation value D2 of the physical characteristics of the second bird's nest sample in the second test result;

[0062] Calculate the first sample size C1 of the first bird's nest sample and the second sample size C2 of the second bird's nest sample;

[0063] The adulteration probability score S is obtained by weighting the results using preset weights W1 and W2 (representing the weights of the first and second detection results, respectively) with the formula S = (W1*D1*C1 + W2*D2*C2) / (C1+C2); and W2 is greater than W1.

[0064] Based on the adulteration probability score S, the results of the assessment of whether bird's nest is adulterated and the probability of adulteration are obtained.

[0065] By employing the above technical solution, it is possible to accurately assess whether bird's nest is adulterated and the assessment result of the possibility of adulteration. Furthermore, the weight W2 of the second test result using clean bird's nest as the second sample is greater than the weight W1 of the first test result using an impurity sample as the first sample, thereby increasing the reliability of the clean bird's nest test result. Simultaneously, the parameter of the test sample size is increased. Therefore, when the sample size in the second test result is small, sporadic errors are more likely to occur, thus increasing the sample size parameter reduces the reliability of the test result when the sample size is small. This makes the adulteration probability score S more accurate and reliable.

[0066] This application also provides a bird's nest impurity removal system for performing the bird's nest evaluation method as described in any of the above embodiments, including an automatic screening device and a bird's nest testing device;

[0067] The automatic screening equipment includes a sorting table, a vacuum suction head movably disposed above the sorting table, and a photographing device for acquiring images of the sorting table.

[0068] The sorting table is used to hold the initial bird's nest, and the vacuum suction head adsorbs the initial bird's nest on the sorting table based on the image acquired by the imaging device.

[0069] The vacuum adsorption head is connected to the bird's nest detection device, which is used to adsorb bird's nest samples by the vacuum adsorption head.

[0070] In some embodiments, the photographing device is configured as a high-resolution CCD / CMOS camera with more than 5 megapixels; and the photographing device is connected to a photoelectric sensor, which triggers the photographing device to take a picture when the photoelectric sensor detects that the sorting table is carrying the initial bird's nest.

[0071] The bottom of the sorting table is equipped with an LED fill light to illuminate the initial bird's nests carried on the sorting table;

[0072] The bird's nest testing equipment is configured as a hyperspectral detection device.

[0073] By adopting the above technical solution, the bird's nest impurity removal system can not only accurately separate impurities in the initial bird's nest, but also make full use of the separated impurity samples to quickly and accurately evaluate the bird's nest; at the same time, the detection process can greatly reduce the loss of bird's nest, and the sampling process has high randomness and representativeness, which can greatly improve the reliability and representativeness of the detection results, and greatly avoid the situation and risk of adulteration in the initial bird's nest. Attached Figure Description

[0074] Figure 1 This is a flowchart illustrating the bird's nest evaluation method according to an embodiment of this application. Detailed Implementation

[0075] The following specific embodiments illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Although the description of this application is presented in conjunction with preferred embodiments, this does not mean that the features of this application are limited to this embodiment. On the contrary, the purpose of describing the application in conjunction with embodiments is to cover other options or modifications that may be derived based on the claims of this application. To provide a thorough understanding of this application, many specific details will be included in the following description. This application may also be implemented without using these details. Furthermore, to avoid confusion or obscuring the focus of this application, some specific details will be omitted in the description. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0076] In the description of this application, it should be noted that the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more. Unless otherwise explicitly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0077] Please see Figure 1 , Figure 1 This is a flowchart illustrating the bird's nest evaluation method according to an embodiment of this application.

[0078] like Figure 1 As shown in the embodiment of this application, a method for evaluating bird's nest in the process of removing impurities from bird's nest is provided, which removes impurities from the initial bird's nest to obtain clean bird's nest and impurity samples;

[0079] The impurity sample was screened and filtered to obtain the first bird's nest sample;

[0080] A first detection model is constructed, and the first bird's nest sample is detected based on the first detection model to obtain a first detection result, wherein the first detection result includes the detection result of the component characteristics of the first bird's nest sample;

[0081] Obtain a second bird's nest sample from the cleaned bird's nest;

[0082] A second detection model is constructed, and the second bird's nest sample is detected based on the second detection model to obtain a second detection result, which includes the detection results of the physical characteristics of the second bird's nest sample.

[0083] The first and second test results are analyzed and processed to obtain the evaluation results of the bird's nest.

[0084] It should be noted that during the production and processing of bird's nest, because bird's nest is a precious tonic, it is easy for fake bird's nest or inferior bird's nest to be mixed into the raw materials, thus reducing the quality of the bird's nest products leaving the factory. In existing technologies, bird's nest is usually sampled and tested at its source and at the time of product delivery to identify whether it has been adulterated. However, this sampling method cannot comprehensively cover all bird's nest products, and the bird's nest samples used in the testing process need to be discarded. When the bird's nest variety is particularly precious, such as blood bird's nest, this testing method can easily result in significant losses of valuable bird's nest. Furthermore, the testing method has a long cycle, with test results typically taking several hours to several days, making it difficult to meet the requirements of real-time testing and accurately locate bird's nest carrying adulterants.

[0085] Based on the above-mentioned actual situation, this application utilizes impurity samples to obtain the first test sample, thereby enabling a preliminary assessment of the compositional characteristics of bird's nest through the test results of the first test sample. Simultaneously, it allows for the reuse of impurity samples that would otherwise be discarded, saving on bird's nest losses during testing and reducing the cost of bird's nest loss in the testing process. Furthermore, since impurities in bird's nest typically exhibit random and widespread distribution, it allows for a more comprehensive sampling of the first bird's nest sample, avoiding the risk of missed detections due to insufficient sampling quantity or poor randomness, and improving the reliability of the first bird's nest sample test.

[0086] It should be noted that the distribution and quantity of normal impurities in the initial bird's nest are highly random, allowing a certain amount of impurity samples to be adsorbed during the impurity removal process. Furthermore, the sampling distribution and quantity of impurity samples are sufficiently representative, greatly reducing the possibility of adulteration in the bird's nest.

[0087] Furthermore, since the first bird's nest sample obtained through the impurity sample contains a large amount of impurities, it affects the accuracy of the first bird's nest sample detection. Therefore, a second detection model is used to detect the second bird's nest sample obtained from the clean bird's nest and evaluate its physical characteristics to obtain a second detection result. Finally, through the analysis and processing of the first and second detection results, the obtained bird's nest evaluation result has extremely high comprehensiveness, reliability and accuracy.

[0088] Furthermore, since there are many dimensions used to evaluate bird's nest detection, while evaluating all dimensions can enhance the accuracy of bird's nest detection, the detection speed is slow, and the model training difficulty and cost are extremely high. Therefore, this application constructs a first detection model based on the component characteristics of bird's nest, and a second detection model based on the physical characteristics of bird's nest. This reduces the training difficulty of the first and second detection models. At the same time, it enables targeted detection of different bird's nest samples, ensuring the reliability and accuracy of the first and second detection models.

[0089] In one embodiment, screening and filtering the impurity sample, including rinsing with water to reduce the impurity content in the sample, can improve the purity of the first bird's nest sample, thereby improving the accuracy of the detection results of the first detection model.

[0090] In one embodiment, the impurity treatment of the initial bird's nest includes:

[0091] Obtain the original image of the initial bird's nest;

[0092] The original image is preprocessed to obtain the first image of the bird's nest;

[0093] An impurity identification model is used to identify the first image and mark the impurity locations and the corresponding de-noising regions; the de-noising regions completely cover the impurity locations.

[0094] Vacuum adsorption is used to adsorb all the impurity removal areas and obtain the impurity sample;

[0095] The remaining initial bird's nest is the clean bird's nest.

[0096] This allows for accurate identification and location of impurities through the first image, ensuring that the impurity removal area completely covers the impurity location. Thus, when the impurity removal area is removed by vacuum adsorption, not only can the impurities be completely removed, ensuring the quality of the remaining clean bird's nest, but a certain amount of bird's nest will inevitably be carried along, resulting in a certain amount of bird's nest in the impurity sample, thereby achieving cost characteristic detection of the first bird's nest sample.

[0097] Among them, the method of identifying and marking the location of impurities based on the first image is an existing technology that can be mastered by those skilled in the art, and the identification accuracy of the existing technology can reach more than 99%, while the identification rate can be less than a few seconds.

[0098] In one embodiment, a vacuum adsorption gun can be used to adsorb all impurity samples in the impurity removal area. In this case, the shape and size of the impurity removal area are set to be the same as the adsorption port of the vacuum adsorption gun, and it is ensured that the impurity removal area can completely cover the impurity location.

[0099] In one implementation, different impurity removal zones can be staggered or overlapped to ensure that each impurity location is adsorbed once, thereby improving the cleanliness of the bird's nest. At the same time, the vacuum adsorption method can reduce the mechanical damage to the bird's nest itself caused by the impurity removal process, which is beneficial to ensuring the quality of the bird's nest.

[0100] In one implementation, obtaining the original image of the initial bird's nest includes:

[0101] Obtain dried bird's nest and soak and clean it;

[0102] The processed dried bird's nest is manually laid on a sorting tray and then manually de-feathered and de-impure to obtain the initial bird's nest;

[0103] The sorting tray containing the initial bird's nest is placed in an automatic sorting device;

[0104] The automatic sorting device provides supplemental lighting from the bottom of the sorting tray and uses a high-resolution camera to acquire the original image of the initial bird's nest from above the sorting tray.

[0105] Manual methods can precisely remove a large number of impurities from dried bird's nests, such as feathers, shells, and particles, and this method is less likely to cause damage to the bird's nest itself. However, dried bird's nests contain many large particles of feathers, eggshells, and other particles, often located within the folds of the nest, making mechanical removal difficult. Directly using large-area adsorption methods for impurity removal would easily lead to significant losses and damage to the physical structure of the clean bird's nest, thus reducing its quality. Therefore, in obtaining the initial image of the bird's nest, manual impurity removal is necessary to remove obvious large particles and impurities embedded within the nest.

[0106] However, manual impurity removal is limited by human limitations and is prone to omissions and errors. Especially for inconspicuous and small-sized impurities, the possibility of omissions is extremely high. Therefore, it is necessary to lay the initial bird's nest on a sorting tray to obtain the original image. This allows for the identification of small impurities that are difficult to detect or ignore manually, so as to obtain impurity samples later.

[0107] In one embodiment, the initial bird's nests are laid flat in the sorting tray. This facilitates obtaining the original image of the initial bird's nests and reduces the overlap between them. On the other hand, it also helps to reduce the amount of clean bird's nests carried during the subsequent impurity adsorption process.

[0108] In one implementation, constructing a first detection model includes:

[0109] According to the first ratio range, pure bird's nest and standard impurities were mixed to obtain training samples;

[0110] Chemical composition data of bird's nest samples were obtained using hyperspectral imaging technology, and a multidimensional feature matrix containing sialic acid content and glycoprotein structure was generated.

[0111] A convolutional neural network model is used to preprocess the multidimensional feature matrix to generate a first standard feature set, which includes feature vectors of sialic acid content and glycoprotein structure.

[0112] Based on the first standard feature set, the principal component analysis algorithm is applied to reduce the dimensionality, retaining the main features of sialic acid content and glycoprotein structure, and obtaining a low-dimensional feature vector.

[0113] Based on the low-dimensional feature vector, a first detection model is constructed to identify the similarity between the target sample and pure bird's nest and standard impurities mixed in a first ratio.

[0114] In this process, pure bird's nest and standard impurities are mixed in a first ratio range to obtain training samples. Then, a convolutional neural network model is used to learn from the training samples, enabling the first detection model to more accurately identify the chemical composition characteristics of pure bird's nest containing impurities. On the other hand, the first detection model only retains the identification and training of the main features of bird's nest sialic acid content and glycoprotein structure, thereby greatly reducing the training difficulty of the first detection model and reducing the construction cost of the first detection model.

[0115] In one embodiment, the first proportion ranges from 5% to 30%, and the standard impurities are obtained through a manual impurity removal process for bird's nest, including eggshells, feathers, and / or dust particles.

[0116] The raw spectral data of bird's nest samples were obtained by using hyperspectral imaging technology, and band selection algorithms were used to extract bands related to chemical components to obtain a preliminary spectral feature set.

[0117] Based on the preliminary spectral feature set, principal component analysis algorithm is used to separate the features related to sialic acid content and glycoprotein structure to generate a chemical component feature subset; for the chemical component feature subset, matrix splicing method is used to integrate the data of sialic acid content and glycoprotein structure to generate a multidimensional feature matrix;

[0118] A convolutional neural network is used to preprocess the multidimensional feature matrix, extracting local features using convolutional and pooling layers to obtain a first feature matrix. If the dimension of the first feature matrix exceeds a preset threshold, principal component analysis (PCA) is used to reduce its dimensionality, resulting in a second feature matrix. Based on the second feature matrix, a deep convolutional neural network is used to extract features related to sialic acid content and glycoprotein structure, generating feature vectors.

[0119] The feature vectors are normalized using a standardization algorithm, and the first standard feature set is generated by using the z-score standardization method.

[0120] It should be noted that when the impurity content in pure bird's nest is too high, it is easy to remove impurities manually. Furthermore, excessive impurities can easily interfere with and affect the primary detection model due to variations in the types and components of impurities, preventing the model from accurately identifying the compositional characteristics of the bird's nest in the training samples. Conversely, if the impurity content in pure bird's nest is too low, the impurities have less impact on the detection results, allowing for direct detection using the analytical model for pure bird's nest without the need to construct a separate primary detection model. Therefore, determining the first proportion range can meet the detection requirements of the first detection sample.

[0121] In one implementation, obtaining the first detection result includes:

[0122] The sialic acid content and glycoprotein characteristic spectrum were obtained from the first bird's nest sample using hyperspectral imaging technology to generate the first characteristic dataset;

[0123] The first feature dataset is identified, and the first detection result is generated based on the similarity of the features;

[0124] The first test result includes the adulteration risk level of the first bird's nest sample, which includes low risk, medium risk and high risk.

[0125] It should be noted that the first detection model performs real-time detection and identification during the bird's nest processing, therefore its detection accuracy is relatively low. However, in actual production, overly precise detection results are not necessary. Therefore, the first detection model is used to obtain the adulteration risk level, and the results are used to assess whether the bird's nest is adulterated, as well as subsequent processing operations. This ensures that the detection process does not affect the bird's nest processing efficiency.

[0126] In one implementation, the detection process of the first detection model includes:

[0127] The proportion of impurities in the first bird's nest sample is estimated. Based on the first image, the first total area of ​​all impurities and the second total area of ​​all impurity-removed areas are calculated. The proportion of impurities is equal to the first total area and the second total area.

[0128] Based on the impurity ratio, the first detection model uses a low-dimensional feature vector whose first ratio is closest to the impurity ratio to perform similarity recognition in order to generate the first detection result.

[0129] When the similarity is greater than 90%, the output adulteration risk level is low.

[0130] When the similarity is less than or equal to 90% and greater than or equal to 70%, the output adulteration risk level is medium risk.

[0131] When the similarity is less than 70%, the output adulteration risk level is high.

[0132] It should be noted that different quantities and types of impurities have varying impacts on the detection results of the first detection model. Therefore, the adulteration risk level output by the first detection model is set as low risk, medium risk, and high risk. Because the cost and proportion of impurities in the actual first bird's nest sample deviate from the cost and proportion of standard impurities in the first detection model, a similarity deviation of less than 10% can be judged as low risk; when the similarity deviation is greater than 30%, the deviation far exceeds the impact of the impurities, thus it can be judged as high risk; the remaining cases can be judged as medium risk, which can be used as a reference for the bird's nest evaluation results.

[0133] In one implementation, the acquisition location and quantity of the second bird's nest sample are determined based on the adulteration risk level in the first detection result;

[0134] When the adulteration risk level is medium or high, bird's nest samples are obtained from adjacent positions of each of the impurity areas to obtain the second bird's nest sample;

[0135] When the adulteration risk level is low, bird's nest samples are randomly obtained from the clean bird's nest to obtain a first number of second bird's nest samples, wherein the first number is 10% to 30% of the number of first bird's nest samples.

[0136] It should be noted that when the adulteration risk level is low, the possibility of adulteration in bird's nest is relatively small, thus reducing the quantity of the first sample. On the one hand, this reduces the loss of clean bird's nest during the bird's nest testing process. On the other hand, the sampling process for the second bird's nest sample takes a certain amount of time, and reducing the quantity of the first sample can reduce the sampling time for the second bird's nest sample, thereby ensuring the efficiency of the bird's nest processing process.

[0137] When the adulteration risk level is medium or high, it indicates that there is a high probability of adulterants in the impurity area. Therefore, it is necessary to obtain bird's nest samples from adjacent positions of each impurity area to obtain the second bird's nest sample. This allows for accurate detection and identification through the second detection model, ensuring the accuracy and reliability of the bird's nest evaluation results.

[0138] When the adulteration risk level is low, it means that no adulterants are present in the impurity area. Therefore, a second bird's nest sample can be randomly obtained from the clean bird's nest for a second test to enhance the accuracy and reliability of the bird's nest evaluation results. At the same time, since pure bird's nest has a high value, the first quantity is set to 10% to 30% of the first bird's nest sample quantity to reduce the loss of the second bird's nest sample.

[0139] Furthermore, the physical properties of the second bird's nest sample are tested, which greatly reduces the damage to the second bird's nest sample during testing, so that the second bird's nest sample can be reused or collected and processed as a secondary product.

[0140] It should be noted that in other alternative implementation methods, preset values ​​can also be used to directly obtain the acquisition location and quantity of the second bird's nest sample; this is beneficial to improving the acquisition efficiency of the second bird's nest sample and is suitable for high-efficiency production and processing scenarios.

[0141] In one implementation, a second detection model is constructed, including:

[0142] Reference values ​​of physical characteristics are obtained from a pre-established database of standard parameters for authentic bird's nests, and a second standard feature set is generated; the second standard feature set includes feature vectors of fiber structure and density distribution.

[0143] A second detection model is constructed based on the second standard feature set to identify whether the target sample is genuine bird's nest.

[0144] In one embodiment, the analysis and processing of the first detection result and the second detection result includes:

[0145] Calculate the first deviation value D1 of the component characteristics of the first bird's nest sample in the first test result, and the second deviation value D2 of the physical characteristics of the second bird's nest sample in the second test result;

[0146] Calculate the first sample size C1 of the first bird's nest sample and the second sample size C2 of the second bird's nest sample;

[0147] The adulteration probability score S is obtained by weighting the results using preset weights W1 and W2 (representing the weights of the first and second detection results, respectively) with the formula S = (W1*D1*C1 + W2*D2*C2) / (C1+C2); and W2 is greater than W1.

[0148] Based on the adulteration probability score S, the results of the assessment of whether bird's nest is adulterated and the probability of adulteration are obtained.

[0149] By identifying the adulteration probability score S, the possibility of adulteration in bird's nest can be accurately assessed, along with the assessment result of the adulteration probability. Furthermore, the weight W2 of the second test result using clean bird's nest as the second sample is greater than the weight W1 of the first test result using an impure sample as the first sample, thus increasing the reliability of the clean bird's nest test results. Simultaneously, the sample size parameter is increased. When the sample size in the second test result is small, sporadic errors are more likely to occur, thus increasing the sample size parameter reduces the reliability of the test result. Therefore, the adulteration probability score S is made more accurate and reliable.

[0150] In one implementation, an adulteration threshold of the adulteration probability score S can be determined based on setting up an actual detection control experiment; when the adulteration probability score S is greater than or equal to the adulteration threshold, the bird's nest is assessed to carry adulterants; when the adulteration probability score S is less than the adulteration threshold, the bird's nest is assessed to not carry adulterants.

[0151] It should be noted that the first detection model can output the first deviation value D1 from the first detection result; the second detection model can output the second deviation value D2 from the second detection result.

[0152] Furthermore, in one application scenario, when the first detection model outputs a low-risk result, but the second deviation value D2 of the second detection result is high, the adulteration probability score S can also be greater than the adulteration threshold. This is because the second bird's nest sample obtained by the second detection model in this case is randomly acquired and may have adsorbed adulterants.

[0153] This application also provides a bird's nest impurity removal system for performing the bird's nest evaluation method as described above, including an automatic screening device and a bird's nest testing device;

[0154] The automatic screening equipment includes a sorting table, a vacuum suction head movably mounted above the sorting table, and a photographic device for acquiring images of the sorting table.

[0155] The sorting table is used to hold the initial bird's nests, and the vacuum adsorption head adsorbs the initial bird's nests on the sorting table based on the image obtained by the imaging device.

[0156] The vacuum adsorption head is connected to the bird's nest testing equipment, which is used to adsorb bird's nest samples by the vacuum adsorption head.

[0157] In one embodiment, the photographing device is configured as a high-resolution CCD / CMOS camera with more than 5 million pixels; and the photographing device is connected to a photoelectric sensor, which triggers the photographing device to take a picture when the photoelectric sensor detects that the sorting table is carrying the initial bird's nest.

[0158] LED supplementary lights are installed at the bottom of the sorting table to illuminate the initial bird's nests carried on the sorting table;

[0159] The bird's nest testing equipment is set to a hyperspectral detection device.

[0160] The examples described herein are merely preferred embodiments of the invention and are not intended to limit the concept and scope of the invention. Any modifications and improvements made by those skilled in the art to the technical solutions of the invention without departing from the design concept of the invention should fall within the protection scope of the invention.

Claims

1. A method for evaluating bird's nest in a bird's nest impurity removal process, characterized in that: the initial bird's nest is subjected to impurity removal treatment to obtain clean bird's nest and impurity samples; the impurity samples are subjected to screening and filtering to obtain first bird's nest samples; a first detection model is constructed, and the first bird's nest samples are detected based on the first detection model to obtain first detection results, the first detection results including detection results of component characteristics of the first bird's nest samples; second bird's nest samples are obtained from the clean bird's nest; a second detection model is constructed, and the second bird's nest samples are detected based on the second detection model to obtain second detection results, the second detection results including detection results of physical characteristics of the second bird's nest samples; the first detection results and the second detection results are analyzed and processed to obtain evaluation results of the bird's nest; the second detection model is constructed, including: reference values of physical characteristics are obtained from a pre-established standard parameter database of pure bird's nest to generate a second standard feature set; the second standard feature set includes feature vectors of fiber structure and density distribution; the second detection model is constructed according to the second standard feature set to identify whether the target sample is pure bird's nest; the analysis and processing of the first detection results and the second detection results include: first deviation values D1 of the component characteristics of the first bird's nest samples in the first detection results and second deviation values D2 of the physical characteristics of the second bird's nest samples in the second detection results are calculated; first sample amounts C1 of the first bird's nest samples and second sample amounts C2 of the second bird's nest samples are calculated; preset weights W1 and W2 are used for weighted calculation, and a formula is S = (W1*D1*C1+W2*D2*C2) / (C1+C2), to obtain a fraud possibility score S; and the W2 is greater than the W1; wherein W1 represents the weight of the first detection results, and W2 represents the weight of the second detection results; and based on the fraud possibility score S, evaluation results of whether the bird's nest is adulterated and the adulteration possibility are obtained. The impurity treatment of the initial bird's nest includes: an original image of the initial bird's nest is obtained; the original image is preprocessed to obtain a first image of the bird's nest; an impurity recognition model is used to recognize the first image, mark impurity positions, and mark corresponding impurity removal areas of the impurity positions; the impurity removal areas completely cover the impurity positions; a vacuum adsorption method is used to adsorb all the impurity removal areas, and the impurity samples are obtained; the remaining initial bird's nest is the clean bird's nest. Obtaining the original image of the initial bird's nest includes: dry bird's nest is obtained and subjected to soaking, cleaning treatment; the treated dry bird's nest is manually laid on a selection tray, and manual hair removal and impurity removal treatment are performed to obtain the initial bird's nest; the selection tray carrying the initial bird's nest is placed in an automatic picking device; the automatic picking device supplements light from the bottom of the selection tray, and a high-resolution camera obtains the original image of the initial bird's nest from above the selection tray. The first detection model is constructed, including: pure bird's nest and standard impurities are mixed according to a first proportion range to obtain training samples; ​ 2. The bird's nest evaluation method in the bird's nest impurity removal process as described in claim 1, characterized in that, ​ ​ ​ ​ ​ ​ 3. The method of Claim 2, wherein the method further comprises the step of: 3.

1. evaluating the quality of the bird's nest based on the number of the bird's nest pieces and the number of the bird's nest fragments. ​ ​ ​ ​ ​ 4. The method of Claim 2, wherein the evaluating of the bird's nest is performed by a human operator. ​ ​ The chemical composition data of the bird's nest sample is obtained by hyperspectral imaging technology, and a multi-dimensional feature matrix containing sialic acid content and glycoprotein structure is generated; A convolutional neural network model is used to preprocess the multi-dimensional feature matrix to generate a first standard feature set, which includes feature vectors of sialic acid content and glycoprotein structure; According to the first standard feature set, a principal component analysis algorithm is applied for dimension reduction processing to retain the main features of sialic acid content and glycoprotein structure, and a low-dimensional feature vector is obtained. According to the low-dimensional feature vector, a first detection model is constructed to identify the similarity between the target sample and the pure bird's nest mixed with the standard impurities according to the first proportion.

5. The bird's nest evaluation method in the bird's nest impurity removal process of claim 4, wherein The first proportion range is 5-30%, the standard impurities are obtained by the artificial impurity removal process of the bird's nest, and the standard impurities include eggshells, feathers and / or dust particles; Raw spectral data of the bird's nest sample is obtained by hyperspectral imaging technology, and a waveband selection algorithm is used to extract wavebands related to chemical composition to obtain a preliminary spectral feature set; According to the preliminary spectral feature set, a principal component analysis algorithm is used to separate sialic acid content and glycoprotein structure related features to generate a chemical composition feature subset; for the chemical composition feature subset, a matrix splicing method is used to integrate sialic acid content and glycoprotein structure data to generate a multi-dimensional feature matrix; A convolutional neural network is used to perform preprocessing operations on the multi-dimensional feature matrix, and a convolutional layer and a pooling layer are used to extract local features to obtain a first feature matrix; if the dimension of the first feature matrix exceeds a preset threshold, a dimension reduction algorithm principal component analysis is used to reduce the dimension of the first feature matrix to obtain a second feature matrix; according to the second feature matrix, a deep structure of the convolutional neural network is used to extract features of sialic acid content and glycoprotein structure to generate a feature vector; A normalization algorithm is used to normalize the feature vector, and a z-score standardization method is used to generate a first standard feature set.

6. The method of Claim 4, wherein the evaluating of the bird's nest is performed by a human operator. The first detection result includes: Sialic acid content and glycoprotein feature spectrum are obtained from the first bird's nest sample by hyperspectral imaging technology to generate a first feature data set; The first feature data set is identified, and the first detection result is generated based on the similarity of the features; The first detection result includes the adulteration risk level of the first bird's nest sample, and the adulteration risk level includes low risk, medium risk and high risk.

7. The method of Claim 6, wherein the evaluating of the bird's nest is performed by a human operator. 7 The detection process of the first detection model includes: Estimating the impurity proportion of the first bird's nest sample, calculating the first total area of all impurities and the second total area of all impurity removal areas based on the first image, and the impurity proportion = first total area / second total area; Based on the impurity proportion, the first detection model uses the low-dimensional feature vector closest to the first proportion and the impurity proportion for similarity identification to generate the first detection result; When the similarity is greater than 90%, the adulteration risk level is output as low risk; When the similarity is less than or equal to 90% and greater than or equal to 70%, the adulteration risk level is output as medium risk. When the similarity is less than 70%, the output adulteration risk level is high risk.

8. The method of Claim 6, wherein the evaluating of the bird's nest is performed by a human operator. 8 Based on the adulteration risk level in the first detection result, the acquisition position and acquisition quantity of the second bird's nest sample are determined; When the adulteration risk level is medium risk or high risk, the bird's nest sample is acquired from the position adjacent to each of the impurity regions to obtain the second bird's nest sample; When the adulteration risk level is low risk, the bird's nest sample is randomly acquired from the clean bird's nest to obtain a first quantity of the second bird's nest sample, and the first quantity is 10%-30% of the first bird's nest sample quantity.

9. A bird's nest impurity removing system, characterized in that, An automatic screening device and a bird's nest detection device are used to perform the bird's nest evaluation method according to any one of claims 1-8. The automatic screening device comprises a picking table, a vacuum suction head movably arranged above the picking table, and a photographing device configured to acquire images of the picking table. The picking table is configured to carry the initial bird's nest, and the vacuum suction head is configured to suction the initial bird's nest on the picking table based on the images acquired by the photographing device. The vacuum suction head is connected to the bird's nest detection device, and the bird's nest detection device is configured to detect the bird's nest sample suctioned by the vacuum suction head.

10. The system of claim 9, wherein the system further comprises a controller configured to control the operation of the system. The photographing device is configured as a high-resolution CCD / CMOS camera with a resolution greater than 5 million pixels, and the photographing device is connected to a photoelectric sensor configured to trigger the photographing device to take pictures when the photoelectric sensor senses that the picking table carries the initial bird's nest. The picking table is provided with an LED fill light at the bottom, which is configured to light the initial bird's nest carried by the picking table. The bird's nest detection device is configured as a hyperspectral detection device.

Images