An emotion recognition system applied to cosmetic body dysmorphic disorder

Abstract

This invention belongs to the field of emotion recognition and medical auxiliary diagnosis and treatment technology, and discloses an emotion recognition system applied to cosmetic body dysmorphic disorder. The system includes an image acquisition module that acquires and de-identifies patient facial videos during pre-operative, intra-operative, and post-operative follow-up stages; a face detection module that locates the face region using a YOLO network adapted to mask occlusion, supine posture, and complex lighting; an expression feature extraction module that extracts multi-level features in parallel using ResNet101-SE, VGG16, and EfficientNet-B0; a feature fusion and discrimination module that normalizes the feature vectors and inputs them into a gradient boosting decision tree classifier, adjusting the output emotion recognition result based on class weights; a risk assessment module that correlates the recognition result with BDD-YBOCS, SAS, and SDS scale scores, generating a risk warning when a threshold is exceeded; and a result processing module that stores data in a time series and interacts with a medical information system to establish a continuous mental health record in the electronic medical record.

Description

Technical Field

[0001] This invention belongs to the field of emotion recognition and medical auxiliary diagnosis and treatment technology, specifically an emotion recognition system applied to cosmetic body dysmorphic disorder. Background Technology

[0002] In the field of cosmetic medicine, Body Dysmorphic Disorder (BDD) is a common psychological disorder characterized by excessive concern and denial of one's own appearance defects. These patients often exhibit long-term negative emotions such as anxiety, depression, and rejection. Their emotional state fluctuates continuously during preoperative consultations, intraoperative communication, and postoperative follow-ups. If the patient's emotional changes at different stages are not identified and intervened in a timely manner, it may not only aggravate psychological risks but also easily lead to treatment dissatisfaction or even doctor-patient disputes. Therefore, it is crucial to continuously, objectively, and traceably identify and assess the emotions of BDD patients in cosmetic medicine practice.

[0003] Currently, psychological scales such as BDD-YBOCS, SAS, and SDS are commonly used in clinical practice to assess patients' psychological state. However, these methods have limitations: scales rely on patient self-reporting, which is highly subjective and easily influenced by avoidance or exaggeration; scale assessments are often conducted at fixed time points, making it difficult to dynamically reflect continuous emotional changes before, during, and after surgery; scales cannot capture immediate facial expression fluctuations, resulting in delayed assessments and a high risk of missed judgments. On the other hand, some existing studies have introduced computer vision and facial expression recognition technologies, but these mostly rely on standardized datasets and fail to consider the special characteristics of medical aesthetic scenarios, such as mask occlusion, supine position, and uneven lighting, leading to decreased detection and recognition accuracy. Furthermore, existing methods often use a single convolutional neural network for feature extraction and classification, making it difficult to simultaneously consider both superficial texture and deep semantic features, and lacking the ability to recognize minority emotions such as contempt and disgust. In practical applications, these shortcomings make it difficult for existing technologies to provide doctors with stable and reliable emotional analysis support. Summary of the Invention

[0004] The purpose of this invention is to provide an emotion recognition system for cosmetic body dysmorphic disorder, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an emotion recognition system applied to cosmetic body dysmorphic disorder, the system comprising: The image acquisition module is configured to acquire facial video streams of patients before, during and after medical aesthetic procedures, and to perform de-identification processing at the acquisition end, outputting only image data related to emotion recognition. The face detection module is configured to receive data output by the image acquisition module and use a YOLO detection network trained on a sample set including mask occlusion, supine posture and illumination changes to determine the face region and generate a standardized cropped image. The facial expression feature extraction module is configured to receive the cropped image and extract multi-level facial expression features using an improved ResNet101-SE network, a VGG16 network, and an EfficientNet-B0 network, respectively. The ResNet101-SE network introduces a channel attention mechanism after the residual block to enhance the features of the periorbital and periorbital regions and outputs the multi-path feature vectors to the fusion module. The feature fusion and discrimination module is configured to normalize and concatenate the multi-path feature vectors, input them into the gradient boosting decision tree classifier, and output the emotion recognition result. During the training process, the class weights are adjusted based on the sample distribution to ensure the stability of minority emotion recognition. The risk assessment module is configured to receive the emotion recognition results and perform correlation analysis with the scores of the BDD-YBOCS, SAS and SDS scales. When the emotion fluctuation index and the scale score both exceed the threshold, a risk warning signal is generated. The results processing module is configured to receive the risk warning signal and the emotion recognition result, and store them in time sequence to form a structured data file with timestamps, so as to interact with the medical information system and establish a continuous mental health record in the electronic medical record.

[0006] Preferably, the image acquisition module includes: Camera unit: Installed in medical aesthetic clinics and follow-up terminals, it can continuously collect video streams during the preoperative consultation, intraoperative operation and postoperative follow-up stages to ensure coverage of the entire treatment cycle. Typical characteristics of patients with cosmetic body dysmorphic disorder (BDD) are high anxiety before surgery, emotional tension during surgery, and easy dissatisfaction or depression after surgery. Therefore, full-cycle collection is necessary. Data preprocessing unit: At the acquisition end, the video frames are directly de-identified to remove individual identity features such as background, hairline, moles, and scars, and only the expression-related areas are retained to form de-identified image data, which is directly transmitted to the face detection module; Traditional manual questionnaires rely on subjective questions and answers, lack real-time data processing, and often overlook the dynamic emotional changes of patients at different stages. This system, through full-cycle, front-end anonymized data collection, can achieve more continuous and objective data recording than manual consultations.

[0007] Preferably, the face detection module includes: Detection Unit: Receives desensitized image data and uses a YOLO network trained under conditions including mask occlusion, supine posture, and complex lighting to locate the face region. In cosmetic clinical scenarios, patients wearing masks or performing procedures while supine often cause traditional detection algorithms to fail. This system introduces these actual interference samples for training to ensure the robustness of the detection results. The output cropped images have a uniform size and resolution to ensure that the input for subsequent feature extraction is standardized; YOLO detection not only locates individual frames but also optimizes detection boxes by combining the continuity between frames in a video sequence, reducing detection jumps caused by momentary occlusion and making emotion recognition results more stable.

[0008] Preferably, the facial expression feature extraction module includes: The facial expression feature extraction module includes an improved ResNet101-SE network, a VGG16 network, and an EfficientNet-B0 network: VGG16: Extracts low-level texture and contour features; EfficientNet-B0: Captures mid-level semantic features with less computation; Improved ResNet101-SE: Embedded channel attention mechanism after residual block to enhance subtle facial features in the periorbital and perioral regions. This design directly corresponds to the typical manifestations of BDD patients: excessive focus on local flaws, and related emotional reactions are often concentrated in the periorbital and perioral regions. The module outputs multiple feature vectors and passes them to the feature fusion and discrimination module; Unlike existing single-network technologies, this system uses a three-network complementarity and channel attention enhancement to make the recognition of subtle facial expressions more accurate, and is especially suitable for discovering the worry, rejection and denial emotions commonly found in BDD patients.

[0009] Preferably, the feature fusion and discrimination module includes: Feature processing unit: Normalizes and concatenates the three feature vectors, and introduces a weighting strategy to make the channel attention-enhanced features account for a higher proportion in the fusion, rather than simply concatenating them; Classification unit: Gradient boosting decision tree (GBDT) is used as the classifier. Through multiple rounds of iterative learning, complex nonlinear relationships are learned to improve the accuracy of the classification boundary. Training optimization unit: To address the common class imbalance problem in BDD patient datasets, the weights of minority emotions such as contempt and disgust are increased to maintain model balance during training; Compared to the common Softmax, this system emphasizes weighted fusion + GBDT classification, which can solve the problems of feature redundancy and neglect of minority classes.

[0010] Preferably, the risk assessment module includes: Input unit: Receives emotion recognition results and scores from three scales: BDD-YBOCS, SAS, and SDS; Analysis Unit: In the time dimension, the emotion recognition sequence is matched with the scale results. The characteristic of BDD patients is that the scale score is consistently high and accompanied by frequent negative emotion fluctuations. This system can detect the consistency between this psychological scale and facial expression dynamics. Early warning unit: When mood fluctuations are frequent and the scale scores exceed the threshold at the same time, a risk warning signal is automatically generated to alert clinicians that there may be a high-risk psychological state. Existing manual assessment scales rely on patient self-reporting, which carries the risk of subjective bias and missed judgments. This system significantly reduces misjudgments by using automatic emotion recognition and scale cross-validation, and can especially help doctors discover hidden psychological risks.

[0011] Preferably, the result processing module includes: Input unit: Receives risk warning signals and emotion recognition results; Storage unit: Store the two in time sequence to generate a structured data file with timestamps, forming a complete trajectory of emotional changes; Interactive Unit: It interfaces with the hospital's HIS / EMR system to establish a continuous mental health record in the electronic medical record. Doctors can access this record to view the long-term trend of the patient's emotions and mental state when making follow-up or surgical decisions. Existing solutions can only output single results. This system achieves a clinical closed loop through time-series archiving and deep interaction with the medical system, providing doctors with decision-making basis for continuous tracking.

[0012] The beneficial effects of this invention are as follows: 1. This invention fully considers the clinical characteristics of patients with cosmetic body dysmorphic disorder in the image acquisition and preprocessing stages. By capturing facial videos throughout the preoperative, intraoperative, and postoperative follow-up processes, it captures the patient's real emotional fluctuations at different stages, such as anxiety, tension, and postoperative dissatisfaction. The data preprocessing unit at the acquisition end directly performs de-identification operations to remove individual features unrelated to emotion recognition, thereby ensuring the objectivity of facial expression data while protecting privacy. Combined with a YOLO detection network trained under conditions of mask occlusion, supine position, and complex lighting, it achieves stable detection and cropping of facial regions in cosmetic treatment scenarios. This design overcomes the lag of scale assessment and the fragility of traditional detection methods, providing a continuous and reliable data foundation for subsequent accurate identification.

[0013] 2. This invention employs a multi-network collaborative structure in the feature extraction and classification stages. It uses ResNet101-SE, VGG16, and EfficientNet-B0 to extract subtle movements, shallow textures, and deep semantic features, respectively. A channel attention mechanism is introduced into ResNet101-SE to enhance the sensitivity of features in the periorbital and perioral regions. These multi-path features are fused and then input into a gradient-boosting decision tree classifier. During training, the class weights are dynamically adjusted based on the sample distribution to ensure that minority emotions such as contempt and disgust can be effectively identified. Compared to existing schemes that rely on a single convolutional network and Softmax classification, this invention maintains the stability and clinical usability of the recognition results under conditions of class imbalance and complex scenarios, thus more realistically reflecting the psychological state of BDD patients.

[0014] 3. This invention, through the combination of a risk assessment module and a result processing module, establishes a correspondence between automated emotion recognition and clinical scale data for the first time, realizing a dual judgment mechanism of emotional dynamics and scale scores. When the recognition results show frequent emotional fluctuations and the scores of the BDD-YBOCS, SAS, or SDS scales simultaneously exceed the set threshold, the system automatically triggers a risk warning. The recognition results and warning signals are stored in a time-series format and a structured file with a timestamp is generated. Through the interaction unit, it connects with the hospital information system to form a continuous mental health record. This solution breaks through the traditional model that relies on a single scale or a single recognition result, realizing long-term traceable closed-loop management, providing objective evidence for doctors to carry out intervention and follow-up, and significantly reducing the risk of missed and misjudgments. Attached Figure Description

[0015] Figure 1 This is a flowchart of the emotion recognition system of the present invention applied to cosmetic body dysmorphic disorder. Detailed Implementation

[0016] 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.

[0017] like Figure 1 As shown, this embodiment of the invention provides an emotion recognition system applied to cosmetic body dysmorphic disorder, the system comprising: The image acquisition module is configured to acquire facial video streams of patients before, during and after medical aesthetic procedures, and to perform de-identification processing at the acquisition end, outputting only image data related to emotion recognition. The face detection module is configured to receive data output from the image acquisition module and use a YOLO detection network trained on a sample set including mask occlusion, supine posture, and lighting changes to determine the face region and generate a standardized cropped image. The facial expression feature extraction module is configured to receive cropped images and extract multi-level facial expression features using improved ResNet101-SE network, VGG16 network and EfficientNet-B0 network respectively. The ResNet101-SE network introduces a channel attention mechanism after the residual block to enhance the features of the periorbital and periorbital regions and outputs the multi-path feature vectors to the fusion module. The feature fusion and discrimination module is configured to normalize and concatenate multiple feature vectors, input them into a gradient boosting decision tree classifier, and output emotion recognition results. During training, the class weights are adjusted based on the sample distribution to ensure the stability of minority emotion recognition. The risk assessment module is configured to receive emotion recognition results and perform correlation analysis with BDD-YBOCS, SAS and SDS scale scores. When the emotion fluctuation index and the scale score both exceed the threshold, a risk warning signal is generated. The results processing module is configured to receive risk warning signals and emotion recognition results, and store them in time sequence to form a structured data file with timestamps, so as to interact with the medical information system and establish a continuous mental health profile in the electronic medical record. Example:

[0018] In this embodiment, the system sequentially includes an image acquisition module, a face detection module, an expression feature extraction module, a feature fusion and discrimination module, a risk assessment module, and a result processing module. The modules communicate and work collaboratively with each other through data flow.

[0019] First, the camera unit in the image acquisition module is installed in the consultation area of ​​the clinic and the follow-up terminal to collect video streams before, during and after the patient's operation. The acquisition end is equipped with a data preprocessing unit, which directly removes identity features such as hairline, moles and scars during the acquisition process, retaining only the expression area, generating desensitized image data and outputting it to the downstream module. This ensures continuous data acquisition and avoids excessive transmission of privacy information, solving the shortcomings of traditional scale assessments that are lagging and discontinuous.

[0020] Secondly, the face detection module receives desensitized image data and calls the YOLO detection network trained on clinical scenario sample sets including mask occlusion, supine position, and complex lighting to achieve robust detection and cropping of face regions. To reduce detection instability caused by instantaneous occlusion or lighting fluctuations, this module optimizes the detection box by combining the inter-frame continuity of the video sequence and outputs a cropped image of uniform size as input for subsequent feature extraction.

[0021] In the facial expression feature extraction module, three networks, ResNet101-SE, VGG16, and EfficientNet-B0, are set to run in parallel. VGG16 is used to obtain shallow texture and contour features, EfficientNet-B0 is responsible for the efficient extraction of mid-level semantic features, and ResNet101-SE introduces a channel attention mechanism in the residual block to focus on enhancing the subtle motion features in areas such as the periorbital region and the periorbital region. This multi-network collaborative extraction method combines local sensitive features with global expression, which is more in line with the subtle emotional expressions commonly seen in BDD patients.

[0022] After receiving multiple feature vectors, the feature fusion and discrimination module performs normalization and weighted fusion by the feature processing unit, and then inputs the result into the gradient boosting decision tree classifier for emotion discrimination. During model training, the training optimization unit dynamically adjusts the class weights according to the sample distribution, so that minority emotions such as contempt and disgust can be effectively identified, avoiding the decrease in accuracy of the traditional Softmax classification method under class imbalance.

[0023] The input unit of the risk assessment module simultaneously receives emotion recognition results and scores from scales such as BDD-YBOCS, SAS, and SDS. The analysis unit compares the correlation between emotion fluctuations and scale scores through time series modeling. When the emotional state fluctuates frequently and the scale scores all exceed the set threshold, the early warning unit generates a risk warning signal to alert doctors that the patient has a high psychological risk.

[0024] Finally, the results processing module receives the risk warning signal and the emotion recognition result, the storage unit stores the data in time sequence and generates a structured file with timestamps, and the interaction unit writes it into the hospital information system to form a continuous mental health record in the electronic medical record, which is convenient for doctors to access at any time.

[0025] Through the above embodiments, the present invention achieves privacy protection at the data collection end, robust face detection under complex conditions, multi-network collaborative feature extraction and minority class emotion recognition, dual verification and early warning of scales and emotions, and structured archiving of results in clinical scenarios, forming a complete diagnosis and treatment closed loop, effectively making up for the shortcomings of existing scale assessment and single recognition methods.

[0026] The image acquisition module includes: Camera unit: Installed in medical aesthetic clinics and follow-up terminals, it can continuously collect video streams during the preoperative consultation, intraoperative operation and postoperative follow-up stages to ensure coverage of the entire treatment cycle. Typical characteristics of patients with cosmetic body dysmorphic disorder (BDD) are high anxiety before surgery, emotional tension during surgery, and easy dissatisfaction or depression after surgery. Therefore, full-cycle collection is necessary. Data preprocessing unit: At the acquisition end, the video frames are directly de-identified to remove individual identity features such as background, hairline, moles, and scars, and only the expression-related areas are retained to form de-identified image data, which is directly transmitted to the face detection module; Traditional manual questionnaires rely on subjective questions and answers, lack real-time data processing, and often overlook the dynamic emotional changes of patients at different stages. This system, through full-cycle, front-end anonymized data collection, can achieve more continuous and objective data recording than manual consultations.

[0027] The face detection module includes: Detection Unit: Receives desensitized image data and uses a YOLO network trained under conditions including mask occlusion, supine posture, and complex lighting to locate the face region. In cosmetic clinical scenarios, patients wearing masks or performing procedures while supine often cause traditional detection algorithms to fail. This system introduces these actual interference samples for training to ensure the robustness of the detection results. The output cropped images have a uniform size and resolution to ensure that the input for subsequent feature extraction is standardized; YOLO detection not only locates individual frames but also optimizes detection boxes by combining the continuity between frames in a video sequence, reducing detection jumps caused by momentary occlusion and making emotion recognition results more stable.

[0028] The facial expression feature extraction module includes: The facial expression feature extraction module includes an improved ResNet101-SE network, a VGG16 network, and an EfficientNet-B0 network: VGG16: Extracts low-level texture and contour features; EfficientNet-B0: Captures mid-level semantic features with less computation; Improved ResNet101-SE: Embedded channel attention mechanism after residual block to enhance subtle facial features in the periorbital and perioral regions. This design directly corresponds to the typical manifestations of BDD patients: excessive focus on local flaws, and related emotional reactions are often concentrated in the periorbital and perioral regions. The module outputs multiple feature vectors and passes them to the feature fusion and discrimination module; Unlike existing single-network technologies, this system uses a three-network complementarity and channel attention enhancement to make the recognition of subtle facial expressions more accurate, and is especially suitable for discovering the worry, rejection and denial emotions commonly found in BDD patients.

[0029] The feature fusion and discrimination module includes: Feature processing unit: Normalizes and concatenates the three feature vectors, and introduces a weighting strategy to make the channel attention-enhanced features account for a higher proportion in the fusion, rather than simply concatenating them; Classification unit: Gradient boosting decision tree (GBDT) is used as the classifier. Through multiple rounds of iterative learning, complex nonlinear relationships are learned to improve the accuracy of the classification boundary. Training optimization unit: To address the common class imbalance problem in BDD patient datasets, the weights of minority emotions such as contempt and disgust are increased to maintain model balance during training; Compared to the common Softmax, this system emphasizes weighted fusion + GBDT classification, which can solve the problems of feature redundancy and neglect of minority classes.

[0030] The risk assessment module includes: Input unit: Receives emotion recognition results and scores from three scales: BDD-YBOCS, SAS, and SDS; Analysis Unit: In the time dimension, the emotion recognition sequence is matched with the scale results. The characteristic of BDD patients is that the scale score is consistently high and accompanied by frequent negative emotion fluctuations. This system can detect the consistency between this psychological scale and facial expression dynamics. Early warning unit: When mood fluctuations are frequent and the scale scores exceed the threshold at the same time, a risk warning signal is automatically generated to alert clinicians that there may be a high-risk psychological state. Existing manual assessment scales rely on patient self-reporting, which carries the risk of subjective bias and missed judgments. This system significantly reduces misjudgments by using automatic emotion recognition and scale cross-validation, and can especially help doctors discover hidden psychological risks.

[0031] The result processing module includes: Input unit: Receives risk warning signals and emotion recognition results; Storage unit: Store the two in time sequence to generate a structured data file with timestamps, forming a complete trajectory of emotional changes; Interactive Unit: It interfaces with the hospital's HIS / EMR system to establish a continuous mental health record in the electronic medical record. Doctors can access this record to view the long-term trend of the patient's emotions and mental state when making follow-up or surgical decisions. Existing solutions can only output single results. This system achieves a clinical closed loop through time-series archiving and deep interaction with the medical system, providing doctors with decision-making basis for continuous tracking.

[0032] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0033] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An emotion recognition system applied to cosmetic body dysmorphic disorder, characterized in that: The system includes: The image acquisition module is configured to acquire facial video streams of patients before, during and after medical aesthetic procedures, and to perform de-identification processing at the acquisition end, outputting only image data related to emotion recognition. The face detection module is configured to receive data output by the image acquisition module and use a YOLO detection network trained on a sample set including mask occlusion, supine posture and illumination changes to determine the face region and generate a standardized cropped image. The facial expression feature extraction module is configured to receive the cropped image and extract multi-level facial expression features using an improved ResNet101-SE network, a VGG16 network, and an EfficientNet-B0 network, respectively. The ResNet101-SE network introduces a channel attention mechanism after the residual block to enhance the features of the periorbital and periorbital regions and outputs the multi-path feature vectors to the fusion module. The feature fusion and discrimination module is configured to normalize and concatenate the multi-path feature vectors, input them into the gradient boosting decision tree classifier, and output the emotion recognition result. During the training process, the class weights are adjusted based on the sample distribution to ensure the stability of minority emotion recognition. The risk assessment module is configured to receive the emotion recognition results and perform correlation analysis with the scores of the BDD-YBOCS, SAS and SDS scales. When the emotion fluctuation index and the scale score both exceed the threshold, a risk warning signal is generated. The results processing module is configured to receive the risk warning signal and the emotion recognition result, and store them in time sequence to form a structured data file with timestamps, so as to interact with the medical information system and establish a continuous mental health record in the electronic medical record.

2. The emotion recognition system for cosmetic body dysmorphic disorder according to claim 1, characterized in that: The image acquisition module includes: The camera unit captures facial video streams during the preoperative consultation, intraoperative communication, and postoperative follow-up stages of the patient. The image acquisition module has a data preprocessing unit at the acquisition end. The data preprocessing unit performs de-identification processing on the acquired video stream to remove identity information that is not related to emotion recognition, generates de-identified image data, and transmits the de-identified image data to the face detection module.

3. The emotion recognition system for cosmetic body dysmorphic disorder according to claim 2, characterized in that: The face detection module includes: The detection unit receives the desensitized image data output by the image acquisition module, and locates the face region in the image based on the YOLO detection network trained on a sample set including mask occlusion, supine posture, and lighting changes, generates a standardized cropped image, and transmits the cropped image as input to the expression feature extraction module.

4. The emotion recognition system for cosmetic body dysmorphic disorder according to claim 3, characterized in that: The facial expression feature extraction module includes: Improved ResNet101-SE network, VGG16 network, and EfficientNet-B0 network; The expression feature extraction module receives the cropped image output by the face detection module and performs feature extraction in the three networks respectively. The ResNet101-SE network sets a channel attention mechanism after the residual block to enhance the subtle movement features of the eye and mouth regions. The expression feature extraction module outputs the multi-path feature vectors obtained by the network to the feature fusion and discrimination module.

5. An emotion recognition system for cosmetic body dysmorphic disorder according to claim 4, characterized in that: The feature fusion and discrimination module includes: The system includes a feature processing unit, a classification unit, and a training optimization unit. The feature processing unit receives the multi-path feature vectors output by the facial expression feature extraction module and performs normalization and concatenation on them. The classification unit inputs the processed feature vector into the gradient boosting decision tree classifier to output the emotion recognition result. The training optimization unit adjusts the weights of each category based on the sample distribution during the model training process to improve the recognition stability of the minority emotions of contempt and disgust.

6. The emotion recognition system for cosmetic body dysmorphic disorder according to claim 5, characterized in that: The risk assessment module includes: The system includes an input unit, an analysis unit, and an early warning unit. The input unit receives the emotion recognition results and BDD-YBOCS, SAS, and SDS scale scores output by the feature fusion and discrimination module. The analysis unit performs correlation analysis on the emotion recognition results and the scale scores. The early warning unit generates a risk warning signal and transmits it to the result processing module when it detects that the emotion fluctuation index and the scale score both exceed a preset threshold.

7. An emotion recognition system for cosmetic body dysmorphic disorder according to claim 6, characterized in that: The result processing module includes: The system includes an input unit, a storage unit, and an interaction unit. The input unit receives risk warning signals and emotion recognition results output by the risk assessment module. The storage unit stores the risk warning signals and emotion recognition results in a time sequence and generates a structured data file with a timestamp. The interaction unit transmits the structured data file to the medical information system to establish a continuous mental health profile in the electronic medical record.

Images