Intelligent tea screening method and system

Abstract

The invention relates to the technical field of intelligent tea leaf screening, and discloses an intelligent tea leaf screening method and system, and the method comprises the steps: collecting tea leaf morphological characteristics, calculating tea leaf texture information, fusing multi-modal data through a CNN-SVM hybrid model, building a tea leaf morphological-texture nonlinear mapping model, and generating unified screening characteristics. And calculating the screening feature similarity of different varieties of tea according to the unified screening feature, constructing a feature migration matrix, and converting the features of different tea varieties. And a cross-variety adaptive tea screening model is established, and a domain adaptive loss function is adopted to optimize a screening discrimination boundary, so that the screening precision is improved. According to the method, the screening system can realize self-adaptive migration of screening characteristics among different varieties, so that the cost of retraining is reduced, and the flexibility and expandability of the screening system are improved. The screening model can keep high accuracy in screening tasks of different varieties, and efficient screening can still be carried out even for new tea varieties which are never seen.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent tea screening, and in particular to an intelligent tea screening method and system. Background Art

[0002] Tea screening is crucial in the tea processing process, affecting the quality, uniformity and market value of tea. Traditional tea screening methods mainly rely on mechanical vibration screening, air separation and manual sorting. However, these methods usually have problems such as low screening accuracy, low efficiency and poor adaptability to tea varieties. In recent years, with the development of artificial intelligence, big data analysis and machine vision technology, intelligent screening has become a research hotspot in the field of tea processing. For example, computer vision combined with image recognition technology can be used for automatic identification of tea morphology, while deep learning models can improve the accuracy of screening classification. In addition, multimodal perception methods based on hyperspectral imaging and sensor technology have been applied to the field of food sorting to obtain more comprehensive physical property data. These technological advances provide new possibilities for improving the accuracy and adaptability of tea screening. However, existing methods still have certain limitations when facing the screening of different varieties of tea, especially the weak generalization ability under small sample conditions, which is difficult to meet the needs of industrialization.

[0003] Although the existing tea screening technology has made some progress, there are still many shortcomings. First, the screening method based on image processing mainly relies on the color and shape characteristics of tea leaves, lacks effective modeling of the internal texture characteristics of tea leaves, and has low adaptability to specific varieties. Secondly, the current intelligent screening system mainly uses a single data source, such as RGB images or near-infrared spectra, and fails to fully utilize the complementarity of multimodal data, which affects the screening accuracy. In addition, traditional machine learning methods (such as support vector machines SVM or random forests) are prone to feature redundancy when processing high-dimensional and multimodal data, and it is difficult to efficiently fuse morphological and texture information. Furthermore, the existing screening models are often trained based on large-scale annotated data, while there are many tea varieties and the data collection cost is high, resulting in poor adaptability of the screening system to small sample data. When facing newly introduced tea varieties, traditional methods usually need to retrain the model, which increases the maintenance cost of the screening system. Finally, the existing screening methods lack a cross-variety adaptation mechanism and fail to build a unified screening feature space, making it difficult to migrate screening models between different varieties. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the existing tea screening method has low screening accuracy, fails to effectively combine multimodal data, and is difficult to adapt to the screening needs of different varieties; and how to improve screening accuracy through multimodal perception and transfer learning, and maintain efficient screening capabilities in a small sample environment.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: an intelligent tea screening method, comprising: collecting tea morphological features, calculating tea texture information, fusing multimodal data through a CNN-SVM hybrid model, establishing a nonlinear mapping model of tea morphology-texture, and generating unified screening features; Calculate the similarity of screening features of different tea varieties based on the unified screening features, construct a feature migration matrix, and transform the features of different tea varieties; A cross-variety adaptive tea screening model was established, and a domain adaptive loss function was used to optimize the screening discrimination boundary to improve the screening accuracy.

[0007] As a preferred solution of the intelligent tea screening method of the present invention, the collecting of tea morphological characteristics and calculating tea texture information comprises: Use hyperspectral imaging to obtain the morphological characteristics of tea leaves, including their color, shape, and texture; A tactile pressure sensor is used to measure the deformation of tea leaves under different pressures, and the texture information of the tea leaves, including the stiffness and deformation coefficient of the tea leaves particles, is calculated through a nonlinear elastic model.

[0008] As a preferred solution of the intelligent tea screening method of the present invention, the nonlinear elastic model includes: Tactile pressure data through the pressure matrix The relationship between the force and deformation at each measuring point is expressed as follows: ; in, It represents the texture deformation coefficient of the tea sample, indicating the degree of nonlinear deformation of the tea under stress. Indicates the maximum pressure applied to the tea leaves. , Represents the parameters of the hyperelastic model and represents the material properties. Represents the relative deformation rate. Represents the pressure data matrix collected by the tactile sensor. Represents the calculation function of texture characteristics, comprehensive deformation coefficient and pressure data. Represents the tea texture feature matrix.

[0009] As a preferred solution of the intelligent tea screening method of the present invention, the method of fusing multimodal data through a CNN-SVM hybrid model to establish a nonlinear mapping model of tea morphology-texture includes: Extracting deep features from hyperspectral data using CNN , SVM is used for classification boundary optimization as: ; in, Represents the hyperspectral morphological features extracted by CNN. Represents the CNN feature extraction function, which obtains high-dimensional features through convolution operations. Represents the feature fusion function, which is used to perform nonlinear mapping in the feature space. Represents the final morphological-textural fusion characteristics. Indicates the tea screening category output by SVM classification. Represents the SVM classification function, which performs classification based on fusion features; Represents the raw data of hyperspectral image; The nonlinear mapping model of tea morphology and texture is established to generate unified screening features. After data fusion, support vector regression + Laplace kernel is used for nonlinear modeling. The nonlinear mapping uses the Laplace kernel function to map the fusion features to the screening space, which is expressed as: ; in, Represents the final unified screening characteristics. Represents the weight coefficient of the SVR model. Represents the Laplace kernel parameter, which controls the smoothness of the feature map. Represents the L1 norm distance metric, which is used to measure the similarity between data points. Represents the number of training data points. Indicates The fused feature vector of training samples.

[0010] As a preferred solution of the intelligent tea screening method of the present invention, the method of calculating the similarity of screening characteristics of different varieties of tea according to the unified screening characteristics and constructing the feature migration matrix includes: In order to measure the distribution differences of different varieties of tea in morphology and texture, the screening feature similarity matrix was calculated. The feature similarity between tea varieties was measured by Mahalanobis distance, and the probability distribution difference of screening features was calculated by kernel density estimation, which is expressed as: ; in, Represents the Mahalanobis distance between varieties i and j. Represents the screening feature vector of variety i. Represents the screening feature vector of variety j. Represents the similarity of screening features between varieties i and j. Represents the KDE kernel bandwidth parameter, which controls the smoothness. Represents the characteristic mean between varieties i and j. Sample points representing feature distribution; represents the joint covariance matrix, which measures the correlation of features; Based on screening feature similarity , construct the feature migration matrix, convert the screening features of the new tea variety to the screening feature space of the existing variety, and use the weighted feature projection method to construct the feature transformation relationship from variety i to variety j, which is expressed as: ; in, Represents the element of the feature migration matrix from variety i to variety j. Represents the feature similarity calculated in the first step. Indicates the number of existing tea varieties. Represents the feature contribution weight of variety k, ensuring maximum information retention when features are transformed. Represents the feature distribution adjustment coefficient, which controls the nonlinearity of feature projection. represents the Euclidean distance, which measures the difference in screening characteristics between varieties i and k; The value range is 0~1, and a value close to 1 indicates that the characteristics between varieties i and j can be directly transferred.

[0011] As a preferred solution of the intelligent tea screening method of the present invention, the characteristics of converting different tea varieties include: Using the Laplace feature mapping method, the screening features of the new tea variety n are converted to the screening space of the existing varieties, expressed as: ; in, Represents the screening characteristics of the new tea variety n after conversion. Represents the transformation coefficient from variety n to variety j in the feature migration matrix. Represents the feature divergence measure between varieties n and j. Represents the L1 norm distance, ensuring that local similarity is maintained when features are transformed.

[0012] As a preferred solution of the intelligent tea screening method of the present invention, the establishment of a cross-variety adaptive tea screening model includes: A deep neural network is used to extract screening features, and feature adversarial training is used to reduce the differences in feature distribution between different varieties, so that new varieties can be seamlessly adapted to the existing screening model, which can be expressed as: ; in, Represents the screening feature output of the new tea variety n, which is used for final discrimination after DNN processing. and Represents the trainable weight matrices of DNN, which are used for nonlinear transformations respectively. and Represents the bias term. Represents the Sigmoid activation function, ensuring output normalization. Represents the hyperbolic tangent activation function, which improves the feature expression ability;

[0013] As a preferred solution of the intelligent tea screening method of the present invention, the method of optimizing the screening discrimination boundary by using a domain adaptive loss function to improve the screening accuracy includes: There is inter-domain deviation in the feature distribution of different varieties. The domain adaptive loss function is introduced to optimize the screening decision boundary, so that the screening model can maintain a high screening accuracy between different varieties. Combining the maximum mean difference and adversarial loss, both feature alignment and classification stability are optimized, expressed as: ; in, represents the domain adaptation loss function. Indicates the number of training samples. and Represents the final screening characteristics of varieties i and j. Represents the Euclidean distance metric, which is used to calculate the MMD loss to align the feature distributions of different varieties. represents the loss weight hyperparameter. Represents the category label of the training sample. and Represents the classifier weights and bias terms. Represents the adversarial loss term, ensuring that the features can be used for screening and classification; The value range is [0,+∞). The smaller the value, the closer the screening characteristic distribution of different varieties is, and the better the adaptation effect is.

[0014] As a preferred embodiment of the intelligent tea screening method of the present invention, wherein: The optimized screening discrimination boundary includes: The maximum margin optimization method is used to ensure that the screening model is robust across different varieties; The Lagrange dual optimization is used to optimize the discrimination boundary of the screening classification so that different varieties can maintain classification stability during the screening process, which is expressed as: ; in, represents the discriminant boundary optimization loss. and represents the Lagrange multiplier. and Represents the category label of the training sample. Represents the kernel function, which measures the nonlinear similarity between screening features; The value range is from negative infinity to positive infinity. During the optimization process, the smaller the value, the clearer the judgment boundary and the higher the screening accuracy. The cross-variety screening model with optimized discrimination boundary is suitable for intelligent screening tasks of different tea varieties.

[0015] An intelligent tea screening system, characterized in that it includes: The multimodal perception feature extraction module collects tea morphological features, calculates tea texture information, fuses multimodal data through the CNN-SVM hybrid model, establishes a nonlinear mapping model of tea morphology-texture, and generates unified screening features; Construct a tea variety feature migration matrix module, calculate the screening feature similarity of different tea varieties based on the unified screening feature, construct a feature migration matrix, and convert the features of different tea varieties; The screening optimization module in a small sample environment establishes a cross-variety adaptive tea screening model, and uses a domain adaptive loss function to optimize the screening discrimination boundary to improve the screening accuracy.

[0016] Beneficial effects of the present invention: Collect tea morphological characteristics and texture information and establish a nonlinear mapping model. Through this step, the morphological and texture information of tea leaves is accurately quantified. After data fusion, a unified screening feature is formed, which enhances the adaptability of the screening system among varieties. Compared with the traditional single image recognition method, this step significantly improves the screening accuracy and reduces the screening misjudgment problem caused by differences in tea varieties, providing solid data support for subsequent feature migration; Calculate the similarity of screening features of different varieties of tea and construct a feature migration matrix. Through this step, the screening system changes from "specific variety adaptation" to "cross-variety adaptation", and can automatically learn the screening features of new tea varieties, achieve rapid migration, and reduce the cost of data collection and model training. The implementation of this step enables the screening system to have adaptive learning capabilities, overcomes the drawback that traditional screening methods cannot be universal across different varieties, and improves the scalability and practicality of intelligent screening; A cross-variety adaptive tea screening model is established to optimize the screening discrimination boundary. Through this step, the screening system not only has the ability to adapt to different varieties, but also can maintain high precision during the screening process of different varieties, avoiding the decline in screening performance due to the drift of variety characteristics. Compared with the traditional fixed screening standard, the screening model of the present invention continuously optimizes the screening decision boundary through adaptive learning, improves the stability of the model, and enables it to still have high screening accuracy and generalization ability in the case of small samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them: Figure 1 This is an overall flow chart of an intelligent tea screening method and system provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0019] Example 1, reference Figure 1 , as an embodiment of the present invention, provides an intelligent tea screening method, comprising: S1: Collect tea morphological features, calculate tea texture information, fuse multimodal data through CNN-SVM hybrid model, establish a nonlinear mapping model of tea morphology-texture, and generate unified screening features.

[0020] Hyperspectral imaging is used to obtain the morphological characteristics of tea leaves, including their color, shape, and texture.

[0021] It should be noted that hyperspectral imaging (HSI) can obtain the reflectance information of tea leaves in different spectral bands to extract their morphological characteristics, including color, shape, and texture. A high-dimensional spectral data matrix is ​​used to represent tea samples, and high-value spectral bands are screened through information entropy to remove redundant data and improve data quality.

[0022] Furthermore, assuming that the tea reflectance data collected by the hyperspectral imaging system is a three-dimensional tensor R (wavelength × space × sampling point), the band data is first screened by information entropy to reduce redundant data and improve classification efficiency, which is expressed as: ; in, Indicates wavelength The information entropy value at the wavelength measures the amount of effective information carried by the wavelength. represents the normalized probability density of the i-th spectral band. It represents the optimal wavelength set selected for subsequent feature extraction. Represents a hyperspectral data matrix, containing tea reflectance information at different wavelengths. Represents the hyperspectral feature extraction function, which is used to filter out spectral bands with high information content. Represents the filtered hyperspectral morphological feature matrix.

[0023] It should be noted that the defects of the existing technology include that RGB images are greatly affected by ambient lighting, and images collected under different lighting conditions have color deviations, which affects the consistency of screening. Low-dimensional image data makes it difficult to accurately distinguish tea varieties. Color and shape information alone cannot accurately reflect the internal quality of tea, which can easily lead to misclassification. The morphological feature extraction method is single, and traditional screening systems often classify based on manually set color thresholds or morphological features, which cannot adapt to complex variety changes.

[0024] Our invention introduces hyperspectral imaging and expands it to multiple spectral channels, enabling the screening system to capture more detailed information and improve the ability to distinguish different varieties. We use spectral normalization methods (such as standard normal variable transformation SNV and principal component analysis PCA) to reduce the impact of light changes on screening accuracy and ensure feature stability. We combine deep learning models (CNN) for automatic feature extraction to avoid the limitations of traditional manual feature engineering and make screening more intelligent.

[0025] Our invention improves the robustness of the screening system, is not affected by ambient light, and can maintain consistent screening results even under different collection conditions. It improves the ability to distinguish varieties. Hyperspectral data provides richer spectral information, enabling the screening system to classify based on finer-grained features. Automated screening feature extraction, without the need for manual threshold setting, improves adaptability, and enables the system to handle screening tasks of different varieties more efficiently.

[0026] A tactile pressure sensor is used to measure the deformation of tea leaves under different pressures, and the texture information of the tea leaves, including the stiffness and deformation coefficient of the tea leaves particles, is calculated through a nonlinear elastic model.

[0027] Furthermore, the texture characteristics of tea leaves, such as softness, elasticity, and particle size, will affect the screening effect. We use tactile pressure sensors to measure the deformation of tea leaves under different pressures. Through the nonlinear elastic model, we calculate the key parameters of tea particles, such as stiffness and deformation coefficient, and convert them into numerical features that can be used for machine learning modeling.

[0028] Tactile pressure data through the pressure matrix The relationship between the force and deformation at each measuring point is expressed as follows: ; in, It represents the texture deformation coefficient of the tea sample, indicating the degree of nonlinear deformation of the tea under stress. Indicates the maximum pressure applied to the tea leaves. , Represents the parameters of the hyperelastic model and represents the material properties. Represents the relative deformation rate. Represents the pressure data matrix collected by the tactile sensor. Represents the calculation function of texture characteristics, comprehensive deformation coefficient and pressure data. Represents the tea texture feature matrix.

[0029] Furthermore, the defects of the existing technology include that a single visual feature cannot characterize the physical texture of tea, and the screening system relies only on morphological features, which leads to the misclassification of some varieties due to their similar morphology. Traditional texture measurement methods are inefficient. For example, manual kneading tests are highly subjective, with inconsistent standards, and are difficult to apply to automated screening systems. Existing sensing technologies lack nonlinear modeling capabilities, and traditional hardness measurement methods (such as compression tests) can only provide simple mechanical parameters and cannot accurately describe the dynamic deformation characteristics of tea.

[0030] Our invention uses tactile pressure sensors to measure the deformation data of tea leaves in real time after being stressed, and obtain physical characteristics such as stiffness and elasticity. Combined with the modified Mooney-Rivlin hyperelastic model, a nonlinear relationship is established, enabling the system to accurately characterize the complex mechanical behavior of tea leaves. Using data fusion technology, texture information is combined with hyperspectral features to avoid the limitations of classification based on a single data source and improve screening accuracy.

[0031] Our invention significantly improves the accuracy of the screening model, avoiding misjudgment due to morphological similarity, especially for high-end tea varieties. It enhances the adaptability of the screening system, and can accurately classify tea particles by texture characteristics regardless of their size. Automated measurement reduces manual intervention, greatly improves the efficiency of the screening system, and makes it more suitable for large-scale industrial applications.

[0032] It should be noted that CNN is used to extract the morphological features of hyperspectral images, SVM is used for classification, and hyperspectral and tactile features are combined in the fusion layer to achieve nonlinear mapping. The key is to extract key morphological features through convolution operations and use feature projection functions to map high-dimensional data to a unified screening feature space.

[0033] Extracting deep features from hyperspectral data using CNN , SVM is used for classification boundary optimization as: ; in, Represents the hyperspectral morphological features extracted by CNN. Represents the CNN feature extraction function, which obtains high-dimensional features through convolution operations. Represents the feature fusion function, which is used to perform nonlinear mapping in the feature space. Represents the final morphological-textural fusion characteristics. Represents the tea screening category output by SVM classification. Represents the SVM classification function, which performs classification based on fusion features. Represents the raw data of hyperspectral image.

[0034] It should be noted that through the above data fusion process, a mathematical model needs to be established to map the morphological and texture characteristics of tea leaves to the screening feature space, which will eventually be used for subsequent transfer learning and variety adaptation. Support vector regression (SVR) + Laplace kernel is used for nonlinear modeling to ensure the smoothness and generalization ability of feature mapping.

[0035] The nonlinear mapping model of tea morphology and texture is established to generate unified screening features. After data fusion, support vector regression + Laplace kernel is used for nonlinear modeling. The nonlinear mapping uses the Laplace kernel function to map the fusion features to the screening space, which is expressed as: ; in, Represents the final unified screening characteristics. Represents the weight coefficient of the SVR model. Represents the Laplace kernel parameter, which controls the smoothness of the feature map. Represents the L1 norm distance metric, which is used to measure the similarity between data points. Represents the number of training data points. Indicates The fused feature vector of training samples.

[0036] This step finally outputs a unified screening feature matrix ,This matrix represents the morphological and texture characteristics of tea, and is subsequently used to construct a feature migration matrix to achieve cross-variety screening adaptation.

[0037] It should be noted that the defects of existing technologies include the difficulty of multimodal data fusion, and the difficulty of traditional methods to effectively combine visual features and mechanical features. The feature dimension is too high and the computational complexity is high. Existing machine learning methods are difficult to efficiently process high-dimensional, multimodal data. Linear classification methods perform poorly on high-dimensional data, and traditional screening models (such as SVM or decision trees) are prone to classification errors when faced with complex tea varieties.

[0038] Our invention uses CNN to extract hyperspectral morphological features, reducing the need for manual feature engineering and improving the automation of classification. SVM is used for classification optimization to ensure that the screening system still has good generalization capabilities in a small sample environment. Laplace kernel regression is introduced to construct a nonlinear mapping model, so that morphological features and texture features can be jointly analyzed in the same screening space, improving the accuracy of classification.

[0039] We invented the deep fusion of multimodal data to improve the adaptability of the screening system to different varieties, so that the system can classify based on richer information. Improve the screening accuracy, enhance the ability to distinguish complex varieties through nonlinear mapping, and reduce screening misjudgments caused by single features. Improve computing efficiency. The combination of CNN+SVM performs well in high-dimensional data processing, ensuring that the screening system can operate efficiently in practical applications.

[0040] Preferably, this step obtains the morphological information (color, shape, texture, etc.) of tea leaves through hyperspectral imaging, and uses tactile pressure sensors to collect the physical texture of tea leaves (stiffness, deformation coefficient, etc.). Subsequently, the CNN-SVM hybrid model is used to fuse these heterogeneous data, construct a nonlinear mapping model of tea leaf morphology-texture, and finally generate a unified screening feature. Hyperspectral imaging can provide fine-grained spectral data far exceeding that of ordinary RGB cameras, enabling the screening system to capture the microstructure and quality characteristics of tea leaves and improve classification accuracy. Tactile pressure sensors make up for the defect that it is difficult to identify the internal texture information of tea leaves by visual features alone, thereby improving the generalization ability of the screening model for different varieties. The CNN-SVM hybrid model realizes the deep fusion of hyperspectral data and tactile data, and establishes a unified feature space through nonlinear mapping, laying the foundation for subsequent feature migration.

[0041] Furthermore, through this step, the morphology and texture information of tea leaves are accurately quantified, and after data fusion, a unified screening feature is formed, which enhances the adaptability of the screening system among varieties. Compared with the traditional single image recognition method, this step significantly improves the screening accuracy and reduces the screening misjudgment problem caused by differences in tea varieties, providing solid data support for subsequent feature migration.

[0042] S2: Calculate the similarity of screening features of different tea varieties based on the unified screening features, construct a feature migration matrix, and convert the features of different tea varieties.

[0043] In order to measure the distribution differences of different varieties of tea in morphology and texture, the screening feature similarity matrix was calculated. The feature similarity between tea varieties was measured by Mahalanobis distance, and the probability distribution difference of screening features was calculated by kernel density estimation, which is expressed as: ; in, Represents the Mahalanobis distance between varieties i and j. Represents the screening feature vector of variety i. Represents the screening feature vector of variety j. Represents the similarity of screening features between varieties i and j. Represents the KDE kernel bandwidth parameter, which controls the smoothness. Represents the characteristic mean between varieties i and j. Represents the sample points of the feature distribution. represents the joint covariance matrix, which measures the correlation of features.

[0044] It should be noted that the Mahalanobis distance is suitable for measuring the similarity of data of different categories, while the kernel density estimation can reduce data noise and improve the robustness of feature matching.

[0045] Furthermore, the defects of the existing technology include that the traditional screening method is based on fixed classification boundaries and cannot be adaptively adjusted according to the similarity between varieties, resulting in the possibility that newly introduced varieties may be misclassified. Traditional measurement methods such as Euclidean distance are unstable in high-dimensional space and it is difficult to effectively measure the similarity of high-dimensional feature data. Ignoring the influence of data distribution, the existing method only relies on point-to-point similarity calculation, but does not fully consider the overall distribution pattern of tea characteristics, which affects the reliability of feature similarity.

[0046] We invented the use of Mahalanobis distance instead of Euclidean distance, so that the correlation between features can be considered when calculating similarity, improving the stability of feature matching. Combined with kernel density estimation (KDE), the distribution modeling of screening features between varieties is carried out to avoid calculation errors caused by feature data distribution deviation. The feature similarity matrix is ​​constructed so that the features of different varieties can be quantified, providing an accurate calculation basis for subsequent feature migration.

[0047] The screening system we invented can accurately measure the similarity between different tea varieties and improve the adaptability of new tea varieties. The calculation is more stable, avoiding the common distance failure problem in high-dimensional data and improving the generalization ability of the screening model. It provides a basis for feature migration. The subsequent steps can use the similarity matrix to perform feature conversion so that new tea varieties can match the existing screening model.

[0048] Based on screening feature similarity , construct the feature migration matrix, convert the screening features of the new tea variety to the screening feature space of the existing variety, and adopt the feature weight adjustment strategy to ensure that the converted features retain the maximum amount of information while reducing the feature distribution deviation. Use the weighted feature projection method to construct the feature transformation relationship from variety i to variety j, expressed as: ; in, Represents the element of the feature migration matrix from variety i to variety j. Represents the feature similarity calculated in the first step. Indicates the number of existing tea varieties. Represents the feature contribution weight of variety k, ensuring maximum information retention when features are transformed. Represents the feature distribution adjustment coefficient, which controls the nonlinearity of feature projection. Represents the Euclidean distance, which measures the difference in screening characteristics between varieties i and k.

[0049] The value range is 0~1, and a value close to 1 indicates that the characteristics between varieties i and j can be directly transferred.

[0050] Furthermore, based on the constructed feature transfer matrix, the feature conversion of new tea varieties is realized so that it can adapt to the existing screening model. The feature mapping function based on Laplace regularization is adopted to ensure that the converted features maintain the screening discriminative power.

[0051] It should be noted that the defects of the existing technology include that the existing screening methods lack cross-variety adaptability. When new tea varieties are introduced, the model must be retrained, resulting in excessively high maintenance costs for the screening system. Simple feature weighting methods are prone to information loss. Traditional weighting methods are prone to information redundancy or loss in high-dimensional feature spaces, affecting screening effects. Feature projection is unstable, and some methods will cause feature distribution deviation during the mapping process, affecting the accuracy of subsequent screening.

[0052] We invented the construction of a weighted feature projection matrix to ensure that the original feature information is retained to the greatest extent when converting the features of different varieties, thereby improving adaptability. Combined with regularization optimization, we prevent data distribution deviation during feature projection and ensure the stability of screening features. We dynamically adjust the migration weights to ensure that the screening features of different varieties can effectively match the existing screening model and improve adaptability.

[0053] Our invention improves the cross-variety adaptability of the screening system, allowing efficient conversion of features between different varieties and improving the screening accuracy of new tea varieties. It reduces the need for model retraining, reduces the maintenance cost of the screening system, and improves the flexibility of the system. It enhances the generalization ability of the screening model to ensure stable operation in a multi-variety environment.

[0054] Using the Laplace feature mapping method, the screening features of the new tea variety n are converted to the screening space of the existing varieties, expressed as: ; in, Represents the screening characteristics of the new tea variety n after conversion. Represents the transformation coefficient from variety n to variety j in the feature migration matrix. Represents the feature divergence measure between varieties n and j. Represents the L1 norm distance, ensuring that local similarity is maintained when features are transformed.

[0055] Preferably, based on the unified screening features extracted in S1, this step calculates the similarity of screening features of different tea varieties, and constructs a feature migration matrix based on this, which is used for the conversion of screening features of different varieties, so that new tea varieties can adapt to the existing screening model. This process uses Mahalanobis distance to calculate the similarity of features of different varieties, and combines weighted feature projection technology to establish a mapping relationship between varieties, thereby reducing the feature distribution deviation of cross-variety screening.

[0056] Feature similarity calculation can effectively measure the deviation of screening features between different varieties, so that the screening system is no longer limited to specific varieties, but can be generalized to a variety of tea types.

[0057] The construction of the feature transfer matrix enables newly introduced tea varieties to be quickly adapted to the existing screening system with small samples or even no samples, greatly reducing the cost of model retraining.

[0058] Weighted feature projection optimizes the accuracy of feature migration, ensuring minimal information loss during screening, while enhancing the stability of cross-variety screening.

[0059] Furthermore, through this step, the screening system changes from "specific variety adaptation" to "cross-variety adaptation", and can automatically learn the screening characteristics of new tea varieties, achieve rapid migration, and reduce the cost of data collection and model training. The implementation of this step enables the screening system to have adaptive learning capabilities, overcomes the drawbacks of traditional screening methods that cannot be universal across different varieties, and improves the scalability and practicality of intelligent screening.

[0060] S3: Establish a cross-variety adaptive tea screening model, use domain adaptive loss function to optimize the screening discrimination boundary, and improve screening accuracy.

[0061] It should be noted that after the feature migration is completed in the second step, the converted screening features Screening and discrimination are still needed. The goal of this step is to establish a cross-variety screening model based on deep neural networks so that new tea varieties can be adapted to existing screening categories. We use a two-stream neural network structure, one branch is used to process known varieties, and the other branch is used to process newly introduced varieties, and finally they are fused in the shared layer to optimize the classification ability.

[0062] A deep neural network is used to extract screening features, and feature adversarial training is used to reduce the differences in feature distribution between different varieties, so that new varieties can be seamlessly adapted to the existing screening model, which can be expressed as: ; in, Represents the screening feature output of the new tea variety n, which is used for final discrimination after DNN processing. and Represents the trainable weight matrices of DNN, which are used for nonlinear transformations respectively. and Represents the bias term. Represents the Sigmoid activation function, ensuring output normalization. Represents the hyperbolic tangent activation function, which improves the feature expression ability.

[0063] It should be noted that the defects of the existing technology include that the existing screening model is limited to specific varieties and is difficult to generalize and apply between different tea varieties. There is a lack of feature fusion mechanism, and the traditional screening method only relies on a single feature and cannot fully utilize multimodal information for adaptation. The risk of model overfitting is high, and when facing new tea varieties, the traditional method is prone to the problem of reduced screening accuracy.

[0064] Our invention uses DNN for feature fusion, jointly models morphological features and texture features, and improves the generalization ability of the model. Combined with the transfer learning strategy, the screening rules of new varieties are optimized through the data of existing varieties to improve the adaptability of the model. Using the self-supervised learning method, the model can still maintain a high screening accuracy under the condition of a small amount of data.

[0065] Our invention improves the generalization ability of the screening system, making it applicable to different tea varieties and reducing the cost of model update. It enhances the screening accuracy, combines multimodal data for discrimination, and improves the stability of the screening effect. It reduces the dependence on large-scale training data, so that the screening system can still operate efficiently in a small sample environment.

[0066] There is inter-domain deviation in the feature distribution of different varieties. A domain adaptive loss function is introduced to optimize the screening decision boundary so that the screening model can maintain a high screening accuracy among different varieties.

[0067] Combining the maximum mean difference and adversarial loss, both feature alignment and classification stability are optimized, expressed as: ; in, represents the domain adaptation loss function. Indicates the number of training samples. and Represents the final screening characteristics of varieties i and j. Represents the Euclidean distance metric, which is used to calculate the MMD loss to align the feature distributions of different varieties. represents the loss weight hyperparameter. Represents the category label of the training sample. and Represents the classifier weights and bias terms. Represents the adversarial loss term, ensuring that features can be used for screening classification.

[0068] The value range is [0,+∞). The smaller the value, the closer the screening characteristic distribution of different varieties is, and the better the adaptation effect is.

[0069] The maximum margin optimization method is used to ensure that the screening model is robust among different varieties.

[0070] The Lagrange dual optimization is used to optimize the discrimination boundary of the screening classification so that different varieties can maintain classification stability during the screening process, which is expressed as: ; in, represents the discriminant boundary optimization loss. and represents the Lagrange multiplier. and Represents the category label of the training sample. Represents the kernel function, which measures the nonlinear similarity between screening features.

[0071] The value range is from negative infinity to positive infinity. During the optimization process, the smaller the value, the clearer the judgment boundary and the higher the screening accuracy.

[0072] The cross-variety screening model with optimized discrimination boundary is suitable for intelligent screening tasks of different tea varieties.

[0073] Furthermore, the defects of the existing technology include that the screening boundary is greatly affected by the data distribution, and the screening accuracy of different varieties may decrease due to distribution deviation. There is a lack of adaptive adjustment mechanism, and it is difficult for the existing method to adaptively adjust the screening rules according to the characteristics of new varieties. The optimization method is too static, and the traditional method cannot dynamically optimize the screening decision boundary.

[0074] Our invention uses MMD to reduce the feature deviation between varieties and ensure that the screening rules are consistent among different varieties. Combined with adversarial training, the robustness of the model is improved by generating adversarial samples. The domain adaptation method is used to optimize the screening rules to ensure that the model can dynamically adapt to new tea varieties.

[0075] Our invention improves the robustness of the screening system and reduces the classification error between varieties. It improves the screening accuracy and ensures that different tea varieties can be stably classified. It also enhances the adaptive ability so that the system can adapt to the ever-changing tea varieties.

[0076] Preferably, with the support of the feature transfer matrix, this step establishes a cross-variety adaptive tea screening model, and optimizes the screening discrimination boundary through the domain adaptive loss function to improve the screening accuracy. Specifically, this step uses a deep neural network (DNN) to extract tea screening features, and uses the maximum mean difference (MMD) to calculate the feature distribution differences between different varieties, and combines the adversarial loss function (Adversarial Loss) training, so that the screening model can maintain stable classification performance, even in the face of new tea varieties that have never been seen before.

[0077] The deep neural network (DNN) further optimizes the screening features, giving it stronger discrimination capabilities between different varieties.

[0078] The domain adaptation loss function reduces the feature distribution deviation between different varieties through the maximum mean difference (MMD), which enables the screening system to adapt to new tea varieties without being affected by the drift of variety features.

[0079] Adversarial loss training makes the discrimination boundary of the screening model more robust through adaptive learning, thereby improving the screening accuracy in a small sample environment.

[0080] Through this step, the screening system not only has the ability to adapt across varieties, but also can maintain high precision during the screening process of different varieties, avoiding the decline in screening performance due to the drift of variety characteristics. Compared with the traditional fixed screening standard, the screening model of the present invention continuously optimizes the screening decision boundary through adaptive learning, improves the stability of the model, and still has high screening accuracy and generalization ability in the case of small samples.

[0081] The above embodiments also include an intelligent tea screening system, specifically: The multimodal perception feature extraction module collects tea morphological features, calculates tea texture information, fuses multimodal data through the CNN-SVM hybrid model, establishes a nonlinear mapping model of tea morphology-texture, and generates unified screening features.

[0082] Construct a tea variety feature migration matrix module, calculate the screening feature similarity of different tea varieties based on the unified screening features, construct a feature migration matrix, and convert the features of different tea varieties.

[0083] The screening optimization module in a small sample environment establishes a cross-variety adaptive tea screening model, and uses a domain adaptive loss function to optimize the screening discrimination boundary to improve the screening accuracy.

[0084] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0085] A computer-readable storage medium stores a computer program, which implements the steps of the method described above when executed by a processor.

[0086] The computer device may be a server. The computer device includes a processor, a memory, an input / output interface (I / O for short) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data cluster data of the power monitoring system. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for intelligent screening of tea leaves is implemented.

[0087] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0088] Example 2 is an embodiment of the present invention, which provides an intelligent tea screening method and system. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0089] The purpose of this experiment is to verify the effectiveness of the intelligent tea screening method proposed in this invention in terms of screening accuracy, feature migration capability and adaptability to different tea varieties. The experiment uses hyperspectral imaging and tactile sensors to obtain the morphological and texture characteristics of tea leaves, and establishes a tea morphology-texture mapping model through a CNN-SVM hybrid model. The feature migration matrix is ​​then used to achieve cross-variety screening optimization, and finally the screening discrimination boundary is optimized through a domain adaptive loss function to improve screening accuracy.

[0090] In order to verify the effectiveness of the proposed intelligent tea screening method, the experiment conducted screening tests among multiple tea varieties, used hyperspectral imaging technology to extract tea morphological characteristics, and combined with tactile pressure sensors to measure tea texture information. The experiment selected six different varieties of tea (Longjing, Biluochun, Dahongpao, Tieguanyin, Pu'er, Baihao Yinzhen), and fused hyperspectral data with tactile data through a CNN-SVM hybrid model to establish a tea morphology-texture mapping model.

[0091] The experimental equipment includes: hyperspectral imaging equipment (collecting hyperspectral data in the wavelength range of 400nm-1000nm), tactile pressure sensors (measuring the stiffness and deformation coefficient of tea leaves), computer vision processing systems (used for CNN model training and classification calculations), and screening devices (used to screen different categories of tea leaves).

[0092] The experimental steps are as follows: Hyperspectral data of six kinds of tea were collected to analyze morphological features such as color, shape, and texture. The deformation of tea leaves under different pressures was measured by tactile sensors, and its stiffness and deformation coefficient were calculated. The deep features of hyperspectral data were extracted using a CNN network, and redundant information was removed. The extracted features were optimized and classified using an SVM classification model to improve recognition accuracy. The morphological and texture data were nonlinearly mapped using an SVR model and Laplace kernel to generate unified screening features. The screening feature similarity of different varieties of tea was calculated, and a feature migration matrix was constructed. The similarity of screening features of different varieties of tea was calculated using the Mahalanobis distance to reduce the classification error between different varieties. The screening decision was optimized using a deep neural network (DNN) to improve the classification adaptability between different varieties. The screening discrimination boundary was optimized using a domain adaptive loss function to improve classification accuracy. The maximum mean difference (MMD) was used to optimize the cross-variety screening accuracy, so that new tea varieties can adapt to existing classification models.

[0093] The experiment used 600 tea samples for testing, 100 samples of each variety, and recorded the experimental data parameters, as shown in Table 1 and Table 2.

[0094] Table 1 Experimental data

[0095] Table 2 Calculation data

[0096] Experimental data show that through intelligent screening methods, the characteristics of different varieties of tea in terms of color, shape, texture and texture are accurately extracted and used for classification optimization.

[0097] The screening characteristic value is used to represent the comprehensive characteristics of tea during the screening process. The higher the value, the more obvious the screening characteristics are and the easier it is to classify. Observing the data, Dahongpao has the highest screening characteristic value (0.88), indicating that its morphological and texture characteristics are more obvious during the screening process and easier to distinguish. Baihao Yinzhen has the lowest value (0.80), indicating that its characteristics are less obvious and may require a more sophisticated classification strategy.

[0098] The feature similarity is used to measure the degree of feature similarity between different varieties. The higher the value, the stronger the transferability between varieties. The highest value appeared in Dahongpao (0.94), indicating that its screening features are highly similar to other varieties, while Baihao Yinzhen (0.85) has the lowest feature similarity, indicating that its characteristics are relatively independent and difficult to transfer to other varieties.

[0099] The migration matrix weight is used to measure the importance of different varieties of tea in cross-variety screening. The higher the weight, the more its characteristics are retained during the migration process. Da Hong Pao has the highest migration matrix weight (0.91), indicating that its screening characteristics can be well migrated to other varieties, while Bai Hao Yin Zhen has the lowest weight (0.81), indicating that its screening characteristics are difficult to migrate.

[0100] The classification boundary optimization loss measures the optimization of the classification boundary of the screening model for different varieties. The smaller the value, the higher the classification stability of the model. Dahongpao (0.012) and Tieguanyin (0.014) have the lowest loss, indicating that their classification stability is good and the screening boundary is clear. Baihao Yinzhen (0.022) has the largest loss, indicating that its classification boundary optimization is more difficult.

[0101] The domain adaptation loss is used to measure the adaptability of the screening model to different varieties. The smaller the value, the more similar the distribution of characteristics of different varieties is, and the stronger the adaptability of the screening system is. Dahongpao has the lowest domain adaptation loss (0.038), indicating that its screening characteristics are highly matched with the model, while Baihao Yinzhen has the highest loss (0.050), indicating that the screening characteristics of this variety are quite different from those of other varieties, and the adaptation strategy needs to be further optimized.

[0102] The final classification accuracy of the method of the present invention was higher than 90%, with Pu'er having the highest accuracy (95.1%), indicating that its screening characteristics were the most stable and easy to classify, while Baihao Yinzhen had the lowest accuracy (91.6%), indicating that there were still some challenges in the classification process of this variety.

[0103] Through the intelligent screening method of the present invention, the screening characteristics of tea leaves are accurately modeled, and the screening characteristics of different varieties are clearly different, which can adapt to the screening needs of different varieties. Compared with the traditional classification method based on color and shape, the present invention uses hyperspectral data, tactile sensor information, multimodal fusion and deep learning to achieve higher classification accuracy, more stable screening boundaries, and stronger cross-variety adaptation capabilities, providing a smarter and more efficient screening solution for the tea processing industry.

[0104] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligent screening of tea leaves, characterized in that: include: Collect tea morphological features, calculate tea texture information, fuse multimodal data through CNN-SVM hybrid model, establish a nonlinear mapping model of tea morphology-texture, and generate unified screening features; Calculate the similarity of screening features of different tea varieties based on the unified screening features, construct a feature migration matrix, and transform the features of different tea varieties; A cross-variety adaptive tea screening model was established, and a domain adaptive loss function was used to optimize the screening discrimination boundary to improve the screening accuracy.

2. The intelligent tea screening method according to claim 1, characterized in that: The collecting of tea morphological characteristics and calculating of tea texture information comprises: Use hyperspectral imaging to obtain the morphological characteristics of tea leaves, including their color, shape, and texture; A tactile pressure sensor is used to measure the deformation of tea leaves under different pressures, and the texture information of the tea leaves, including the stiffness and deformation coefficient of the tea leaves particles, is calculated through a nonlinear elastic model.

3. The intelligent tea screening method according to claim 2, characterized in that: The nonlinear elastic model includes: Tactile pressure data through the pressure matrix The relationship between the force and deformation at each measuring point is expressed as follows: ； in, The texture deformation coefficient of the tea sample indicates the nonlinear deformation degree of the tea under stress; Indicates the maximum pressure applied to the tea leaves; , Parameters representing the hyperelastic model, representing material properties; Represents the relative deformation rate; represents the pressure data matrix collected by the tactile sensor; Represents the calculation function of texture characteristics, integrated deformation coefficient and pressure data; Represents the tea texture feature matrix.

4. The intelligent tea screening method according to claim 3, characterized in that: The method of fusing multimodal data through the CNN-SVM hybrid model to establish a nonlinear mapping model of tea morphology-texture includes: Extracting deep features from hyperspectral data using CNN , SVM is used for classification boundary optimization as: ； in, represents the hyperspectral morphological features extracted by CNN; Represents the raw data of hyperspectral image; Represents the CNN feature extraction function, which obtains high-dimensional features through convolution operations; Represents the feature fusion function, which is used for nonlinear mapping in feature space; It represents the final morphological-textural fusion characteristics; Indicates the tea screening category output by SVM classification; Represents the SVM classification function, which performs classification based on fusion features; The nonlinear mapping model of tea morphology and texture is established to generate unified screening features. After data fusion, support vector regression + Laplace kernel is used for nonlinear modeling. The nonlinear mapping uses the Laplace kernel function to map the fusion features to the screening space, which is expressed as: ； in, represents the final unified screening characteristics; Represents the weight coefficient of the SVR model; represents the Laplace kernel parameter, which controls the smoothness of the feature map; Represents the L1 norm distance metric, which is used to measure the similarity between data points; Represents the number of training data points; Indicates The fused feature vector of training samples.

5. The intelligent tea screening method according to claim 4, characterized in that: The method of calculating the similarity of screening features of different varieties of tea according to the unified screening features and constructing a feature migration matrix comprises: In order to measure the distribution differences of different varieties of tea in morphology and texture, the screening feature similarity matrix was calculated. The feature similarity between tea varieties was measured by Mahalanobis distance, and the probability distribution difference of screening features was calculated by kernel density estimation, which is expressed as: ； in, represents the Mahalanobis distance between varieties i and j; Represents the screening feature vector of variety i; Represents the screening feature vector of variety j; Indicates the similarity of screening characteristics between varieties i and j; represents the KDE kernel bandwidth parameter, controlling the smoothness; represents the characteristic mean between varieties i and j; Sample points representing feature distribution; represents the joint covariance matrix, which measures the correlation of features; Based on screening feature similarity , construct the feature migration matrix, convert the screening features of the new tea variety to the screening feature space of the existing variety, and use the weighted feature projection method to construct the feature transformation relationship from variety i to variety j, which is expressed as: ； in, Represents the element of the feature migration matrix from variety i to variety j; Indicates the feature similarity calculated in the first step; Indicates the number of existing tea varieties; represents the feature contribution weight of variety k, ensuring maximum information retention when features are transformed; Represents the feature distribution adjustment coefficient, which controls the nonlinearity of feature projection; represents the Euclidean distance, which measures the difference in screening characteristics between varieties i and k; The value range is 0~1, and a value close to 1 indicates that the characteristics between varieties i and j can be directly transferred.

6. The intelligent tea screening method according to claim 5, characterized in that: The characteristics of converting different tea varieties include: Using the Laplace feature mapping method, the screening features of the new tea variety n are converted to the screening space of the existing varieties, expressed as: ； in, represents the screening characteristics of the new tea variety n after conversion; represents the transformation coefficient from variety n to variety j in the feature migration matrix; represents the characteristic divergence measure between varieties n and j; Represents the L1 norm distance, ensuring that local similarity is maintained when features are transformed.

7. The intelligent tea screening method according to claim 6, characterized in that: The establishment of a cross-variety adaptation screening tea model comprises: A deep neural network is used to extract screening features, and feature adversarial training is used to reduce the differences in feature distribution between different varieties, so that new varieties can be seamlessly adapted to the existing screening model, which can be expressed as: ； in, It represents the screening feature output of the new tea variety n, which is used for final discrimination after being processed by DNN; and Represents the trainable weight matrix of DNN, which is used for nonlinear transformation; and represents the bias term; Represents the Sigmoid activation function, ensuring output normalization; Represents the hyperbolic tangent activation function, which improves the feature expression ability.

8. The intelligent tea screening method according to claim 7, characterized in that: The method of optimizing the screening discrimination boundary by using a domain adaptive loss function to improve the screening accuracy includes: There is inter-domain deviation in the feature distribution of different varieties. The domain adaptive loss function is introduced to optimize the screening decision boundary, so that the screening model can maintain a high screening accuracy between different varieties. Combining the maximum mean difference and adversarial loss, both feature alignment and classification stability are optimized, expressed as: ； in, represents the domain adaptation loss function; Indicates the number of training samples; and It represents the final screening characteristics of varieties i and j; Represents the Euclidean distance metric, which is used to calculate the MMD loss to align the feature distributions of different varieties; represents the loss weight hyperparameter; Represents the category label of the training sample; and Represents the classifier weights and bias terms; Represents the adversarial loss term, ensuring that the features can be used for screening and classification; The value range is [0,+∞). The smaller the value, the closer the screening characteristic distribution of different varieties is, and the better the adaptation effect is.

9. The intelligent tea screening method according to claim 8, characterized in that: The optimized screening discrimination boundary includes: The maximum margin optimization method is used to ensure that the screening model is robust across different varieties; The Lagrange dual optimization is used to optimize the discrimination boundary of the screening classification so that different varieties can maintain classification stability during the screening process, which is expressed as: ； in, represents the discriminant boundary optimization loss; and represents the Lagrange multiplier; and Represents the category label of the training sample; Represents the kernel function, which measures the nonlinear similarity between screening features; The value range is from negative infinity to positive infinity. During the optimization process, the smaller the value, the clearer the judgment boundary and the higher the screening accuracy. The cross-variety screening model with optimized discrimination boundary is suitable for intelligent screening tasks of different tea varieties.

10. An intelligent tea screening system using the method according to any one of claims 1 to 9, characterized in that: The multimodal perception feature extraction module collects tea morphological features, calculates tea texture information, fuses multimodal data through the CNN-SVM hybrid model, establishes a nonlinear mapping model of tea morphology-texture, and generates unified screening features; Construct a tea variety feature migration matrix module, calculate the screening feature similarity of different tea varieties based on the unified screening feature, construct a feature migration matrix, and convert the features of different tea varieties; The screening optimization module in a small sample environment establishes a cross-variety adaptive tea screening model, and uses a domain adaptive loss function to optimize the screening discrimination boundary to improve the screening accuracy.

Images