Construction system and method of facial expression recognition training model

Abstract

The invention discloses a construction system and method for a facial expression recognition training model, and the system comprises a database assembly, the database assembly is composed of a data collection module, a data classification module, a data storage module, a data output module, a historical access module and a backup storage module, and the data collection module is in control connection with the data classification module; according to the invention, by using the database component, the data information in the backup storage module is directly called to be output, so that the data extraction efficiency during model training is improved, and the model training is accelerated; by using a feature extraction component, redundant features are removed, key features useful for expression classification are reserved, and feature extraction is optimized, so that the complexity of the features is reduced, and the robustness of subsequent model recognition output is guaranteed; by using the optimization output component, combining the prediction results of the plurality of models and performing weighting processing, the final recognition result is generated, and the stability and the accuracy of the model recognition result are improved.

Description

Technical Field

[0001] This invention relates to the field of facial expression recognition technology, specifically to a system and method for constructing a facial expression recognition training model. Background Technology

[0002] Facial recognition, also known as face recognition, is a biometric-based identification method. It uses a common camera as the information acquisition device to capture facial images of the subject in a non-contact manner. The images are then compared with images in a database after being acquired by a computer system to complete the recognition process. Facial expression recognition is a technology that identifies emotional states by analyzing facial expressions and movements. While some models employ expression recognition techniques, traditional models have shortcomings. First, training traditional expression recognition models requires frequent extraction of large amounts of data, resulting in inefficient data extraction and slowing down model training. Second, the features extracted during training are complex, compromising the robustness of subsequent model outputs. Third, training traditional expression recognition models requires high stability and accuracy in the recognition results. Therefore, designing a system and method for constructing a facial expression recognition training model is essential. Summary of the Invention

[0003] This invention claims priority. The earlier application was filed on November 19, 2024, with application number 2024116532609, and the receiving agency is China (CN).

[0004] The purpose of this invention is to provide a system and method for constructing a facial expression recognition training model, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a system for constructing a facial expression recognition training model, comprising a database component, wherein the database component consists of a data acquisition module, a data classification module, a data storage module, a data output module, a historical access module, and a backup storage module. The data acquisition module is connected to the data classification module, the data classification module is connected to the data storage module, the data storage module is connected to the data output module, the data output module is connected to the historical access module, the historical access module is connected to the backup storage module, and the data output module is connected to the data extraction module in the data extraction component.

[0006] As a further technical solution of the present invention, the data extraction component comprises a data extraction module, an expression recognition module, a data processing module, a standardization module, a normalization module, a grayscale processing module, a data enhancement module, and a data docking module. The data extraction module is connected to the expression recognition module, the expression recognition module is connected to the data processing module, the data processing module is connected to the standardization module, the data processing module is connected to the normalization module, the data processing module is connected to the grayscale processing module, the data processing module is connected to the data enhancement module, and the data processing module is connected to the data docking module.

[0007] As a further technical solution of the present invention, the data docking module is controlled to connect to the data receiving module in the feature extraction component.

[0008] As a further technical solution of the present invention, the feature extraction component comprises a data receiving module, an organ feature module, a texture region module, a predefined module, a feature recognition module, a geometric positioning module, a learning extraction module, a redundancy processing module, a feature selection module, a dimensionality reduction technology module, and a feature output module. The data receiving module is controlled and connected to the organ feature module, which in turn controls and connects to the feature recognition module. The feature recognition module is controlled and connected to the texture region module, which in turn controls and connects to the data receiving module. The data receiving module is controlled and connected to the predefined module, which is also controlled and connected to the feature recognition module. The feature recognition module is controlled and connected to the geometric positioning module, which in turn controls and connects to the redundancy processing module. The redundancy processing module is controlled and connected to the learning extraction module, which is also controlled and connected to the feature recognition module.

[0009] As a further technical solution of the present invention, the redundancy processing module is controlled and connected to a feature selection module, the redundancy processing module is controlled and connected to a dimensionality reduction technology module, the redundancy processing module is controlled and connected to a feature output module, and the feature output module is controlled and connected to a feature receiving module in the optimized output component.

[0010] As a further technical solution of the present invention, the optimized output component comprises a feature receiving module, a receiving and combining module, an ensemble learning module, a weighted averaging module, a result comparison module, a model adjustment module, and a model output module. The feature receiving module is connected to the receiving and combining module, the receiving and combining module is connected to the ensemble learning module, the ensemble learning module is connected to the weighted averaging module, the weighted averaging module is connected to the result comparison module, the result comparison module is connected to the model adjustment module, and the model adjustment module is connected to the model output module.

[0011] A method for constructing a facial expression recognition training model includes the following steps: Step 1: Database establishment and data extraction processing; Step 2: Model feature extraction and training; Step 3: Model optimization and output.

[0012] In step one above, a database for model training is established and its contents are extracted.

[0013] In step two above, features are extracted and trained from the data;

[0014] In step three above, the results of the model output are evaluated, and the model output is optimized.

[0015] As a further technical solution of the present invention, in step one, various facial image data information is collected by the data acquisition module, and classified and defined by the data classification module to define different expression types. The data is then transmitted to the data storage module for storage, completing the establishment of the model training database. The data output module transmits the data in the data storage module according to the request. At the same time, the frequency and pattern of data requests are recorded by the historical access module. The data with high frequency of access is stored in the backup storage module. When the same data request occurs again, the data information in the backup storage module is directly called for output. The data extraction module extracts the data from the established database and transmits it to the expression recognition module for expression recognition, which serves as a comparison reference for the subsequent model output. The data is then transmitted to the data processing module, where it is preprocessed by the standardization module, normalization module, grayscale processing module, and data augmentation module to reduce noise and interference in the data information. The processed data information is then transmitted to the data docking module.

[0016] As a further technical solution of the present invention, in step two, the processed data is transmitted to the data receiving module through the data docking module. The received data is then used for feature localization and extraction through the organ feature module, texture region module, and predefined module to extract information that is beneficial to expression classification. The processed data is then transmitted to the feature recognition module. The feature recognition module uses the geometric localization module to locate the key feature points of the face, and then calculates the geometric parameters such as the distance and angle between these feature points as expression features. The learning extraction module automatically learns the hierarchical features in the image to learn the feature representations useful for expression classification. The extracted and recognized expression feature data is then transmitted to the redundancy processing module. The redundancy processing module performs principal component analysis (PCA) and linear discriminant analysis (LDA) on the data through the feature selection module and dimensionality reduction module to remove redundant features and retain the key features useful for expression classification. Finally, the processed expression classification information is output through the feature output module.

[0017] As a further technical solution of the present invention, in step three, the facial expression recognition information extracted and output by the model is received by the feature receiving module, the facial expression recognition information output by multiple models is combined by the receiving and combining module, and then transmitted to the integrated learning module by the receiving and combining module. The combined facial expression information is then transmitted to the weighted averaging module by the integrated learning module. In the weighted averaging module, the facial expression recognition information of each model is weighted and processed to form the final unique facial expression recognition information, which is then output to the result comparison module. This information is compared with the classified information in the database, and a judgment is made according to the set recognition error threshold. If it is not qualified, the model recognition parameters are adjusted by the model adjustment module, and the recognition training is carried out again until it is qualified, and then it is output by the model output module.

[0018] Compared with existing technologies, the beneficial effects of this invention are as follows: This system and method for constructing a facial expression recognition training model collects various facial image data through a data acquisition module, classifies and defines different expression types through a data classification module, and transmits this data to a data storage module for storage, thus establishing a model training database. A data output module then transmits data from the data storage module upon request. Simultaneously, a historical access module records the frequency and patterns of data requests, storing frequently accessed data through a backup storage module. When the same data request occurs again, the data from the backup storage module is directly retrieved for output, improving the efficiency of data extraction during model training and accelerating model training. A data docking module transmits the processed data to a data receiving module. An organ feature module, a texture region module, and a predefined module perform feature localization and extraction on the received data, extracting information beneficial for expression classification. The processed data is then transmitted to a feature recognition module, where a geometric localization module locates key facial feature points, and then these feature points are calculated. Geometric parameters such as distance and angle between images are used as facial expression features. The learning extraction module automatically learns hierarchical features in the image, acquiring feature representations useful for facial expression classification. The extracted facial expression feature data is then transmitted to a redundancy processing module. This module performs Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) through a feature selection module and a dimensionality reduction module to remove redundant features and retain key features useful for facial expression classification. This optimization of feature extraction reduces feature complexity and ensures the robustness of subsequent model recognition output. A feature receiving module receives the facial expression recognition information extracted by the model. A receiving and combining module combines the facial expression recognition information from multiple models and transmits this combination to an ensemble learning module. The ensemble learning module then transmits the combined facial expression information to a weighted averaging module. In the weighted averaging module, the facial expression recognition information from each model is weighted to form a final, unique facial expression recognition result, which is then output to a result comparison module. This module combines the prediction results of multiple models and performs weighted processing to generate the final recognition result, improving the stability and accuracy of the model's recognition results. Attached Figure Description

[0019] Figure 1 This is a diagram of the overall architecture of the present invention;

[0020] Figure 2 This is a flowchart of the database component in this invention;

[0021] Figure 3 This is a flowchart of the data extraction component in this invention;

[0022] Figure 4This is a flowchart of the feature extraction component in this invention;

[0023] Figure 5 This is a flowchart of the optimized output component in this invention;

[0024] Figure 6 This is a flowchart of the method of the present invention;

[0025] Figure 7 In Example 1, the RAF-DB dataset was used as the dataset. The system and method of the present invention were compared with the baseline model YOLO12n, and the mAP@0.5 comparison chart was obtained.

[0026] Figure 8 In Example 1, the LLD dataset was used as the dataset. The system and method of the present invention were compared with the baseline model YOLO12n, and the mAP@0.5 comparison chart was obtained.

[0027] In the diagram: 1. Database component; 2. Data extraction component; 3. Feature extraction component; 5. Optimized output component; 11. Data acquisition module; 12. Data classification module; 13. Data storage module; 14. Data output module; 15. Historical access module; 16. Backup storage module; 21. Data extraction module; 22. Expression recognition module; 23. Data processing module; 24. Standardization module; 25. Normalization module; 26. Grayscale processing module; 27. Data augmentation module; 28. Data output module; 3 1. Data receiving module; 32. Organ feature module; 33. Texture region module; 34. Predefined module; 35. Feature recognition module; 36. Geometric localization module; 37. Learning extraction module; 38. Redundancy processing module; 39. Feature selection module; 40. Dimensionality reduction technology module; 41. Feature output module; 51. Feature receiving module; 52. Receiver combination module; 53. Ensemble learning module; 54. Weighted average module; 55. Result comparison module; 56. Model adjustment module; 57. Model output module. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Please see the appendix Figure 1 - Appendix Figure 5This invention provides an embodiment of a facial expression recognition training model construction system, comprising a database component 1. The database component 1 consists of a data acquisition module 11, a data classification module 12, a data storage module 13, a data output module 14, a history access module 15, and a backup storage module 16. The data acquisition module 11 is connected to the data classification module 12, which in turn is connected to the data storage module 13. The data storage module 13 is connected to the data output module 14, which is connected to the history access module 15. The history access module 15 is connected to the backup storage module 16, and the data output module 14 is connected to a data extraction component 2. The data extraction module 21 in the data extraction component 2 consists of a data extraction module 21, an expression recognition module 22, a data processing module 23, a standardization module 24, a normalization module 25, a grayscale processing module 26, a data enhancement module 27, and a data docking module 28. The data extraction module 21 is connected to the expression recognition module 22, which in turn is connected to the data processing module 23. The data processing module 23 is connected to the standardization module 24, the normalization module 25, the grayscale processing module 26, and the data enhancement module 27. The data processing module 28 is also connected to... Data docking module 28 controls and connects to data receiving module 31 in feature extraction component 3. Feature extraction component 3 consists of data receiving module 31, organ feature module 32, texture region module 33, predefined module 34, feature recognition module 35, geometric localization module 36, learning extraction module 37, redundancy processing module 38, feature selection module 39, dimensionality reduction technology module 40, and feature output module 41. Data receiving module 31 controls and connects to organ feature module 32, organ feature module 32 controls and connects to feature recognition module 35, feature recognition module 35 controls and connects to texture region module 33, and texture region module 33 controls and connects to data receiving module 31. On block 31, the data receiving module 31 controls and connects to a predefined module 34, which in turn controls and connects to a feature recognition module 35. The feature recognition module 35 controls and connects to a geometric positioning module 36, which in turn controls and connects to a redundancy processing module 38. The redundancy processing module 38 controls and connects to a learning extraction module 37, which in turn controls and connects to the feature recognition module 35. The redundancy processing module 38 also controls and connects to a feature selection module 39, a dimensionality reduction technology module 40, and a feature output module 41. The feature output module 41 controls and connects to the feature receiving module 51 in the optimized output component 5.The optimized output component 5 consists of a feature receiving module 51, a receiving and combining module 52, an ensemble learning module 53, a weighted averaging module 54, a result comparison module 55, a model adjustment module 56, and a model output module 57. The feature receiving module 51 is connected to the receiving and combining module 52, which in turn is connected to the ensemble learning module 53. The ensemble learning module 53 is connected to the weighted averaging module 54, which in turn is connected to the result comparison module 55. The result comparison module 55 is connected to the model adjustment module 56, and the model adjustment module 56 is connected to the model output module 57.

[0030] Please see the appendix Figure 6 The present invention provides an embodiment of a method for constructing a facial expression recognition training model, comprising the following steps: Step 1: database establishment and data extraction processing; Step 2: model feature extraction and training; Step 3: model optimization output;

[0031] In step one above, a database for model training is established and processed. Various facial image data are collected by the data acquisition module 11 and classified and defined by the data classification module 12. Different expression types are defined and transmitted to the data storage module 13 for storage, thus completing the establishment of the model training database. The data output module 14 transmits the data in the data storage module 13 according to the request. At the same time, the frequency and pattern of data requests are recorded by the historical access module 15. The data with high frequency of access is stored in the backup storage module 16. When the same data request occurs again, the data information in the backup storage module 16 is directly called for output. The data extraction module 21 extracts the data from the established database and transmits it to the expression recognition module 22 for expression recognition, which serves as a comparison reference for the subsequent model output. The data is then transmitted to the data processing module 23, where the data is preprocessed by the standardization module 24, normalization module 25, grayscale processing module 26, and data augmentation module 27 to reduce noise and interference in the data information. The processed data information is then transmitted to the data docking module 28.

[0032] In step two above, features in the data are extracted and trained. The processed data is transmitted to the data receiving module 31 through the data docking module 28. The received data is localized and extracted through the organ feature module 32, texture region module 33 and predefined module 34. Information that is useful for expression classification is extracted and the processed data is transmitted to the feature recognition module 35. The feature recognition module 35 uses the geometric positioning module 36 to locate the key feature points of the face, and then calculates the geometric parameters such as the distance and angle between these feature points as expression features. The learning extraction module 37 automatically learns the hierarchical features in the image and learns the feature representations that are useful for expression classification. The extracted and recognized expression feature data is transmitted to the redundancy processing module 38. The redundancy processing module 38 performs principal component analysis (PCA) and linear discriminant analysis (LDA) on the data through the feature selection module 39 and the dimensionality reduction technology module 40 to remove redundant features and retain the key features that are useful for expression classification. The processed expression classification information is output through the feature output module 41.

[0033] In step three above, the model output is evaluated and the model output is optimized. The feature receiving module 51 receives the facial expression recognition information extracted by the model, the receiving and combining module 52 combines the facial expression recognition information output by multiple models, and the receiving and combining module 52 transmits it to the ensemble learning module 53. The ensemble learning module 53 then transmits the combined facial expression information to the weighted averaging module 54. In the weighted averaging module 54, the facial expression recognition information of each model is weighted and processed to form the final unique facial expression recognition information, which is then output to the result comparison module 55. This result is compared with the classified information in the database, and a judgment is made based on the set recognition error threshold. If the result is not qualified, the model adjustment module 56 adjusts the model recognition parameters and retrains the model until it is qualified, and then outputs the result through the model output module 57.

[0034] Working Principle: During operation, various facial image data are collected by the data acquisition module 11 and classified and defined by the data classification module 12. Different expression types are defined and then transmitted to the data storage module 13 for storage, completing the establishment of the model training database. The data output module 14 transmits the data from the data storage module 13 according to requests. Simultaneously, the historical access module 15 records the frequency and pattern of data requests, and the frequently accessed data is stored in the backup storage module 16. When the same data request occurs again, the data information in the backup storage module 16 is directly called for output, which improves the efficiency of data extraction during model training and speeds up model training. The processed data is transmitted to the data receiving module 31 through the data docking module 28. The organ feature module 32, texture region module 33, and predefined module 34 perform feature localization and extraction on the received data, extracting information that is helpful for expression classification. The processed data is then transmitted to the feature recognition module 35. The feature recognition module 35 uses the geometric positioning module 36 to locate the key feature points of the face, and then calculates the geometric parameters such as the distance and angle between these feature points. Using these as facial expression features, the learning extraction module 37 automatically learns hierarchical features in the image, learning feature representations useful for facial expression classification. The extracted facial expression feature data is then transmitted to the redundancy processing module 38. The redundancy processing module 38 performs Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) on these features through the feature selection module 39 and the dimensionality reduction module 40 to remove redundant features, retaining key features useful for facial expression classification. This optimizes feature extraction, reduces feature complexity, and ensures the robustness of subsequent model recognition output. The feature receiving module 51 receives the model's input... The output facial expression recognition information is obtained and combined by the receiving and combining module 52. The combined information is then transmitted to the ensemble learning module 53, and the combined facial expression information is transmitted to the weighted averaging module 54. In the weighted averaging module 54, the facial expression recognition information of each model is weighted and processed to form the final unique facial expression recognition information, which is then output to the result comparison module 55. The prediction results of multiple models are combined and weighted to generate the final recognition result, which improves the stability and accuracy of the model recognition result.

[0035] Example 1

[0036] Ablation experiments were conducted using different datasets:

[0037] Table 1 shows the results of ablation experiments based on the RAF-DB dataset using the system and method of this invention. Figure 7 Using the RAF-DB dataset as the dataset, the system and method of this invention are compared with the baseline model YOLO12n, as shown in the mAP@0.5 comparison chart.

[0038] Table 1

[0039] Baseline C3k2_star A2C2f_MCA LRFE ATFL P(%) R(%) F1 mAP@0.5(%) Params / M √ 78.4 78.6 0.78 83.8 2.5 √ √ 78.8 80.7 0.79 85.9 2.5 √ √ 78.6 82.7 0.81 86.9 2.5 √ √ 81.2 80.8 0.81 86.8 2.8 √ √ 78.7 81.3 0.80 86.7 2.5 √ √ √ 80.1 81.5 0.81 87.0 2.8 √ √ √ √ 80.6 81,7 0.81 87.3 3.0 √ √ √ √ √ 81.8 81.9 0.82 87.6 3.0

[0040] Table 2 shows the results of ablation experiments based on the LLD dataset using the system and method of this invention. Figure 8 Using the LLD dataset as the dataset, the system and method of this invention are compared with the baseline model YOLO12n, as shown in the mAP@0.5 comparison chart.

[0041] Table 2

[0042] Baseline C3k2_star A2C2f_MCA LRFE ATFL P(%) R(%) F1 mAP@0.5(%) Params / M √ 87.3 82.8 0.85 90.9 2.5 √ √ 91.0 87.5 0.89 94.0 2.5 √ √ 89.6 87.3 0.88 93.2 2.5 √ √ 92.4 87.4 0.90 94.4 2.8 √ √ 90.1 87.9 0.89 93.8 2.5 √ √ √ 89.2 89.6 0.90 94.2 2.8 √ √ √ √ 92.3 89.4 0.91 95.0 3.0 √ √ √ √ √ 91.9 91.2 0.92 95.9 3.0

[0043] according to Figure 7 and Figure 8 Experimental results show that the method of the present invention can significantly improve detection accuracy.

[0044] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A system for constructing a facial expression recognition training model, comprising a database component (1), characterized in that: The database component (1) consists of a data acquisition module (11), a data classification module (12), a data storage module (13), a data output module (14), a historical access module (15), and a backup storage module (16). The data acquisition module (11) is connected to the data classification module (12), the data classification module (12) is connected to the data storage module (13), the data storage module (13) is connected to the data output module (14), the data output module (14) is connected to the historical access module (15), the historical access module (15) is connected to the backup storage module (16), and the data output module (14) is connected to the data extraction module (21) in the data extraction component (2).

2. The system for constructing a facial expression recognition training model according to claim 1, characterized in that: The data extraction component (2) consists of a data extraction module (21), an expression recognition module (22), a data processing module (23), a standardization module (24), a normalization module (25), a grayscale processing module (26), a data enhancement module (27), and a data docking module (28). The data extraction module (21) is connected to the expression recognition module (22), the expression recognition module (22) is connected to the data processing module (23), the data processing module (23) is connected to the standardization module (24), the data processing module (23) is connected to the normalization module (25), the data processing module (23) is connected to the grayscale processing module (26), the data processing module (23) is connected to the data enhancement module (27), and the data processing module (23) is connected to the data docking module (28).

3. The system for constructing a facial expression recognition training model according to claim 2, characterized in that: The data docking module (28) controls the data receiving module (31) in the feature extraction component (3).

4. The system for constructing a facial expression recognition training model according to claim 3, characterized in that: The feature extraction component (3) consists of a data receiving module (31), an organ feature module (32), a texture region module (33), a predefined module (34), a feature recognition module (35), a geometric positioning module (36), a learning extraction module (37), a redundancy processing module (38), a feature selection module (39), a dimensionality reduction technology module (40), and a feature output module (41). The data receiving module (31) is connected to the organ feature module (32), and the organ feature module (32) is connected to the feature recognition module (35). The feature recognition module (35) controls... The texture region module (33) is connected to the data receiving module (31), the data receiving module (31) is connected to the predefined module (34), the predefined module (34) is connected to the feature recognition module (35), the feature recognition module (35) is connected to the geometric positioning module (36), the geometric positioning module (36) is connected to the redundancy processing module (38), the redundancy processing module (38) is connected to the learning extraction module (37), and the learning extraction module (37) is connected to the feature recognition module (35).

5. The system for constructing a facial expression recognition training model according to claim 4, characterized in that: The redundancy processing module (38) is connected to a feature selection module (39), a dimensionality reduction technology module (40), a feature output module (41), and a feature receiving module (51) in the optimized output component (5).

6. The system for constructing a facial expression recognition training model according to claim 5, characterized in that: The optimized output component (5) consists of a feature receiving module (51), a receiving and combining module (52), an ensemble learning module (53), a weighted averaging module (54), a result comparison module (55), a model adjustment module (56), and a model output module (57). The feature receiving module (51) is connected to the receiving and combining module (52), the receiving and combining module (52) is connected to the ensemble learning module (53), the ensemble learning module (53) is connected to the weighted averaging module (54), the weighted averaging module (54) is connected to the result comparison module (55), the result comparison module (55) is connected to the model adjustment module (56), and the model adjustment module (56) is connected to the model output module (57).

7. A method for constructing a facial expression recognition training model, comprising the following steps: Step 1: Database establishment and data extraction processing; Step 2: Model feature extraction and training; Step 3: Model optimization and output; Its key features are: In step one above, a database for model training is established and its contents are extracted. In step two above, features are extracted and trained from the data; In step three above, the results of the model output are evaluated, and the model output is optimized.

8. The method for constructing a facial expression recognition training model according to claim 7, characterized in that: In step one, various facial image data information is collected by the data acquisition module (11), and classified and defined by the data classification module (12) to define different expression types. The data is then transmitted to the data storage module (13) for storage, thus completing the establishment of the model training database. The data in the data storage module (13) is then transmitted to the data storage module (13) according to the request through the data output module (14). At the same time, the frequency and pattern of data requests are recorded by the historical access module (15), and the data with high frequency of access is stored through the backup storage module (16) for the next occurrence of the same data request. When the data is in the backup storage module (16), the data information is directly called and output. The data is extracted from the database by the data extraction module (21) and transmitted to the expression recognition module (22) for expression recognition. As a comparison reference for the subsequent model output, the data is transmitted to the data processing module (23). The data is preprocessed by the standardization module (24), normalization module (25), grayscale processing module (26) and data enhancement module (27) to reduce noise and interference in the data information. The processed data information is transmitted to the data docking module (28).

9. The method for constructing a facial expression recognition training model according to claim 7, characterized in that: In step two, the processed data is transmitted to the data receiving module (31) through the data docking module (28). The received data is located and extracted by the organ feature module (32), texture region module (33) and predefined module (34). Information that is useful for expression classification is extracted and the processed data is transmitted to the feature recognition module (35). The feature recognition module (35) uses the geometric positioning module (36) to locate the key feature points of the face and then calculates the distance, angle and other geometric parameters between these feature points as expression features. The learning extraction module (37) automatically learns the hierarchical features in the image and learns the feature representation that is useful for expression classification. The extracted and recognized expression feature data is transmitted to the redundancy processing module (38). The redundancy processing module (38) performs principal component analysis (PCA) and linear discriminant analysis (LDA) on the data through the feature selection module (39) and dimensionality reduction technology module (40) to remove redundant features and retain the key features that are useful for expression classification. The processed expression classification information is output through the feature output module (41).

10. The method for constructing a facial expression recognition training model according to claim 7, characterized in that: In step three, the facial expression recognition information extracted and output by the model is received by the feature receiving module (51), the facial expression recognition information output by multiple models is combined by the receiving and combining module (52), and then transmitted to the integrated learning module (53) by the receiving and combining module (52). The combined facial expression information is then transmitted to the weighted averaging module (54) by the integrated learning module (53). The facial expression recognition information of each model is weighted in the weighted averaging module (54) to form the final unique facial expression recognition information, which is then output to the result comparison module (55). The result is compared with the classified information in the database and judged according to the set recognition error threshold. If the result is not qualified, the model recognition parameters are adjusted by the model adjustment module (56) and the recognition training is repeated until the result is qualified. The result is then output by the model output module (57).

Images