A method, system, and device for accurately predicting human emotions based on facial expression recognition.

Abstract

This invention discloses a method for accurately predicting human emotions based on facial expression recognition, comprising the following steps: 1) acquiring a facial image and preprocessing the image; 2) determining the facial position and key points in the image; 3) extracting deep features from the facial image; 4) processing temporal features; 5) modeling the model space and global relationships; 6) using an emotion classification model to obtain the probability distribution of each emotion; 7) outputting the emotion label with the highest probability as the final prediction or providing the probability of each emotion for user reference. This invention can automatically learn features from images, better capture subtle changes in facial expressions, and improve the accuracy of emotion prediction. The introduction of a deep learning model improves the model's generalization ability, enabling it to maintain high prediction accuracy in different scenarios and better handle complex emotional states. It can achieve efficient processing in large-scale data and real-time application scenarios, improving real-time performance and efficiency.

Description

Technical Field

[0001] This invention relates to the fields of computer vision, deep learning, and sentiment analysis, and in particular to a method for accurately predicting human emotions based on recognizing facial expressions. Background Technology

[0002] In the fields of computer vision and sentiment analysis, several methods and devices exist aimed at recognizing facial expressions and predicting emotional states. Most of these techniques are based on image processing and pattern recognition methods, utilizing traditional feature extraction and machine learning techniques. While these methods have achieved some success, they also have limitations:

[0003] 1. Limitations of traditional feature extraction methods: Many existing technologies still rely on traditional, hand-designed feature extraction methods, which struggle to fully capture the complex and subtle changes in facial expressions, thus affecting the accuracy of emotion prediction.

[0004] 2. Insufficient model generalization ability: Some existing technologies have limited model generalization ability, and their performance often degrades under different lighting, facial expressions, angles, etc., making it difficult to adapt to the variability of the real world.

[0005] 3. Challenges in identifying complex emotional states: Many emotions are complex and diverse, and traditional methods have difficulty accurately predicting complex emotional states.

[0006] 4. Real-time performance and efficiency issues: Some existing technologies exhibit insufficient processing speed in large-scale data and real-time application scenarios. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method for accurately predicting human emotions based on facial expression recognition, overcoming the limitations of traditional methods mentioned in the background art, and improving the accuracy, precision and real-time performance of facial expression recognition and emotion prediction.

[0008] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0009] A method for accurately predicting human emotions based on facial expression recognition includes the following steps:

[0010] (1) Obtain face images and preprocess the images;

[0011] (2) Determine the location of faces and important key points in the image using face detection algorithms;

[0012] (3) Use deep learning and trained VGG16 model to extract deep features from face images;

[0013] (4) The extracted features are fed into the LSTM model for temporal feature processing;

[0014] (5) The temporal features processed by the LSTM model are fused in parallel with the deep features extracted by the VGG16 model to form fused features, which are then fed into the Transformer for model space and global relationship modeling, and the final feature output is obtained from the Transformer.

[0015] (6) Based on the output of Transformer, construct an emotion classification model and use the softmax activation function to obtain the probability distribution of each emotion;

[0016] (7) Output the sentiment label with the highest probability as the final prediction or provide the probability of each sentiment for the user's reference.

[0017] Furthermore, the VGG16 model performs deep feature extraction on face images, including the following steps:

[0018] (1) The input image is processed through multiple convolutional layers, and a 3x3 convolutional kernel is used for convolution operation;

[0019] (2) Formula for convolution operation: Y=X*K, where X is the input feature map, K is the convolution kernel, and * indicates convolution operation;

[0020] (3) Next, each convolution operation is followed by a ReLU activation function and a 2x2 max pooling operation;

[0021] (4) Finally, the image features are processed through three fully connected layers.

[0022] Further, the steps of temporal feature processing include:

[0023] (1) Use each frame output of VGG16 as the input of LSTM;

[0024] (2) LSTM processes time-series data through its internal “memory units” to provide enhanced time-series features for the next step of Transformer.

[0025] Furthermore, the deep learning and training steps for the VGG16 model include:

[0026] (1) Forward propagation: Input image, extract features step by step through convolution and pooling layers to obtain the final feature representation;

[0027] (2) Calculate the loss: Use a loss function to measure the difference between the predicted output and the actual label;

[0028] (3) Backpropagation: The gradient of the parameters is calculated through backpropagation;

[0029] (4) Parameter update: Update parameters using an optimization algorithm;

[0030] (5) Iterative training: Repeat the above steps and iterate multiple times until convergence.

[0031] The present invention also provides a system for accurately predicting human emotions based on recognizing facial expressions, which is equipped with or stores a program that can implement the above-mentioned method for accurately predicting human emotions based on recognizing facial expressions.

[0032] The present invention also provides a device for accurately predicting human emotions based on recognizing facial expressions, for implementing the above method, including an image input source module for sequentially transmitting data, a preprocessing module, a face detection and key point localization module, a feature extraction module, an emotion classification model module, an emotion prediction module, and an output module;

[0033] The image input source module is used to acquire the face image to be identified, providing a data foundation for subsequent facial expression recognition and emotion prediction;

[0034] The preprocessing module is used to process face images through a series of steps, including cropping, scale normalization, and grayscale conversion. These steps transform the images into standardized input data, providing clear and consistent data for feature extraction and the establishment of sentiment classification models.

[0035] A face detection and key point localization module is used to locate face regions in an image and locate key points of the face.

[0036] The feature extraction module uses a deep convolutional neural network as the core method to extract high-level features from the preprocessed face image.

[0037] The emotion classification model module is used to build emotion classification models and predict the emotional state corresponding to facial expressions;

[0038] The emotion prediction module is used to input the facial features to be predicted into the emotion classification module, and through model learning and reasoning, generate the emotion state prediction result corresponding to the input face.

[0039] The output module is used to present the prediction results to the user in a readable manner.

[0040] The advantages of this invention are:

[0041] This invention introduces advanced deep learning technology to construct an emotion classification model for facial expressions, thereby overcoming the shortcomings of existing technologies. Specific innovations include:

[0042] 1. Automatic Feature Learning: This invention uses deep learning technology to automatically learn features from images, better capture subtle changes in facial expressions, and improve the accuracy of emotion prediction.

[0043] 2. Optimize model generalization ability: The introduction of deep learning models improves the generalization ability of the model, enabling it to maintain high prediction accuracy in different scenarios.

[0044] 3. Multi-emotional state recognition: Through deep learning models, this invention can better handle complex emotional states and identify subtle emotional changes.

[0045] 4. Improved real-time performance and efficiency: This invention, combined with modern computer hardware, enables efficient processing of large-scale data and real-time application scenarios, thus improving real-time performance and efficiency. Attached Figure Description

[0046] Figure 1 This is a diagram illustrating the training process of VGG16.

[0047] Figure 2 A schematic diagram of the VGG16 model construction. Detailed Implementation

[0048] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0049] One: A method for accurately predicting human emotions based on facial expression recognition

[0050] When it comes to technical solutions based on accurate facial expression recognition to predict emotions, we need to go through a series of steps. First, we acquire facial images from various sources, including using cameras, image files, video streams, and so on.

[0051] The acquired face images undergo preprocessing, including resizing and pixel value normalization, to facilitate feature extraction and subsequent processing. To ensure that each image can be correctly read and calculated by the model, all images are uniformly resized to 224x224 pixels and normalized between 0 and 1. Next, using a deep learning face detector and keypoint locator, we can accurately locate the face and important keypoints, such as eyes, nose, and mouth. To accurately capture faces and related keypoints from images, MTCNN (Multi-Task Cascaded Convolutional Network) is used for face and keypoint detection. This method can not only accurately detect faces but also find keypoints on the face, such as the corners of the eyes, the tip of the nose, and the corners of the mouth. These keypoints play a crucial role in the feature alignment process.

[0052] In the feature extraction process, we chose convolutional neural networks (CNNs) as the key component. Among them, VGG16 (Visual Geometry Group 16 layers) is a highly regarded CNN model that performs well in the field of image recognition. The VGG16 model consists of 16 convolutional layers and 3 fully connected layers. Its core idea is to gradually increase the depth of the network to capture different levels of image features.

[0053] VGG16 works by progressively reducing the size of feature maps through a series of convolutional and pooling layers, thereby extracting multi-level features from an image. In the convolutional layers, VGG16 uses small 3x3 convolutional kernels, combining multiple convolutions and activation functions to capture a variety of features in the image. Subsequently, pooling layers (typically with 2x2 pooling kernels) are used to reduce the spatial resolution of the feature maps while increasing the model's receptive field, thus better capturing the structure in the image.

[0054] The training process for VGG16 involves the following steps (see appendix). Figure 1 ):

[0055] Initialization parameters: Randomly initialize the weight parameters of the convolutional and fully connected layers.

[0056] Forward propagation: The input image is processed through convolutional and pooling layers to extract features step by step, resulting in the final feature representation.

[0057] Calculate the loss: Use a loss function to measure the difference between the predicted output and the actual label, such as cross-entropy loss.

[0058] Backpropagation: The gradient of the parameters is calculated through backpropagation.

[0059] Parameter update: Update parameters using optimization algorithms (such as gradient descent).

[0060] Iterative training: Repeat the above steps multiple times until convergence.

[0061] In the next step of the method, we use preprocessed and VGG16-processed face images as input for feature extraction. The convolutional and pooling operations of VGG16 transform the images into high-level features, which are captured at deeper levels of the network and accurately represent key information about facial expressions. This approach allows us to automatically extract useful features about emotional states from images, laying the foundation for subsequent emotion classification tasks.

[0062] However, simply using image features is not enough. When processing consecutive video frames, the temporal continuity of emotions also needs to be considered. Therefore, LSTM, a neural network that can handle temporal correlations, is used. The feature output obtained from the VGG16 model is used as the input of LSTM. After LSTM processing, a new temporal feature vector is output.

[0063] To better model the spatial and global relationships in emotion recognition, the Encoder part of the Transformer was further employed (avoiding the Decoder part, as decoding is not required in emotion recognition). The output of the LSTM was concatenated with the features of the VGG16 to form a fused feature, which was then used as the input to the Transformer. After the Transformer's self-attention computation, the final emotion recognition feature was output.

[0064] In the model building phase for sentiment classification, a fully connected layer was added to the output of the Transformer, and Softmax was used for classification. During model training, cross-entropy loss was used, and backpropagation was employed to fine-tune the weights of all layers, thereby achieving accurate sentiment recognition.

[0065] We concatenate the features extracted from VGG16 into a sentiment classification model. This model will be trained using deep learning techniques to predict sentiment states from the features, a step that may include convolutional layers, pooling layers, fully connected layers, etc., thus constructing an end-to-end sentiment classification model.

[0066] Here, VGG16's feature extraction capabilities will enable the model to better capture sentiment-related information, thereby improving prediction accuracy.

[0067] In summary, the method for accurately recognizing facial expressions to predict emotions involves multiple steps, employing convolutional neural networks, particularly VGG16, to extract high-level features from images. VGG16, with its deep structure and multi-level feature representation capabilities, enables us to effectively capture key information from facial expressions. By combining the features extracted by VGG16 with a sentiment classification model, we can construct a powerful sentiment prediction system to predict the emotional state in facial images with high accuracy.

[0068] II. Devices and Systems:

[0069] In order to apply the technology based on accurate facial expression recognition to predict emotions to real-world scenarios, we designed a complete device and system. The device contains multiple key modules to carry different systems and functions, which work together to achieve the goal of emotion prediction.

[0070] First, the image input source module is the starting point of the device, used to acquire the face image to be recognized. This module provides various input methods, such as camera, image file, and video stream, offering a rich data source for the system. Through these input methods, users can easily input face images into the system, providing a data foundation for subsequent facial expression recognition and emotion prediction.

[0071] Next, the face detection and keypoint localization module plays a crucial role. Employing advanced face detection algorithms, such as convolutional neural network-based methods, this module accurately locates face regions in the image. Subsequently, a keypoint locator is used to locate key facial features, such as the eyes, nose, and mouth. This step provides accurate face region and keypoint location information for subsequent preprocessing and feature extraction modules.

[0072] In the preprocessing module, the acquired face images undergo a series of processing steps, such as cropping, scale normalization, and grayscale conversion. These steps transform the images into standardized input data, providing clear and consistent data for feature extraction and the establishment of sentiment classification models.

[0073] The feature extraction module employs a deep convolutional neural network (CNN) as its core method to extract high-level features from preprocessed face images. Specifically, we chose the VGG16 model, which progressively reduces the size of the feature maps through a series of convolutional and pooling layers to capture image features at different levels. This model's structure includes 16 convolutional layers and 3 fully connected layers, using small 3x3 convolutional kernels to capture multi-level features. After training, the VGG16 model is able to better capture emotional features in facial expressions.

[0074] The emotion classification model module is the core of the device, constructing an emotion classification model to predict the emotional state corresponding to facial expressions. The model consists of convolutional layers, pooling layers, and fully connected layers to learn and predict emotional states (such as facial expressions) from features extracted from VGG16. Figure 2 The model construction considers feature extraction at different scales and introduces adversarial learning methods. The model is trained and optimized using multi-task labeled data to improve its generalization ability and robustness.

[0075] The sentiment prediction module inputs the facial features to be predicted into the sentiment classification model. Through the model's learning and reasoning, it generates a predicted sentiment state corresponding to the input face. These predictions may be presented in the form of probability distributions or specific sentiment labels, providing users with comprehensive sentiment information.

[0076] Finally, the output module presents the prediction results to the user in a readable format, which may include labels for facial emotion states or display the probability distribution of the prediction results. Through this device, users can input facial images into the system and obtain real-time, accurate predictions of facial emotions, realizing the practical application of a technical solution for accurately recognizing facial expressions to predict emotions. The entire device integrates technologies such as deep learning and computer vision, bringing innovation and practicality to the field of emotion recognition.

[0077] VGG16 model construction (see appendix) Figure 2 ):

[0078] VGG16 + LSTM + Transformer fusion model

[0079] Algorithm steps

[0080] 1. Data Preparation

[0081] o Obtain face images from the data source.

[0082] o Perform image preprocessing, such as resizing and normalization.

[0083] 2. Face and key point detection

[0084] o Uses advanced face detection algorithms to determine the location of faces in an image.

[0085] o Locate key points of the face, such as eyes, nose, and mouth.

[0086] 3. Feature extraction: VGG16

[0087] We utilize the VGG16 model to extract deep features from face images. VGG16 is a deep convolutional neural network with multiple convolutional and fully connected layers, ensuring in-depth extraction of key features from images.

[0088] VGG16 Feature Extraction:

[0089] The input image is passed through multiple convolutional layers, with convolution operations performed using 3x3 convolutional kernels.

[0090] The formula for convolution operation is: where X is the input feature map, K is the convolution kernel, and * indicates the convolution operation.

[0091] Next, each convolution operation is followed by a ReLU activation function and a 2x2 max pooling operation.

[0092] Ultimately, the image features are processed through three fully connected layers.

[0093] 4. Temporal Feature Processing: LSTM

[0094] The extracted features are fed into the LSTM model. LSTM is specifically designed for processing time-series data and can effectively capture the temporal correlations and contextual information of emotional changes in consecutive face images (such as video frames).

[0095] LSTM is used to process temporal information when there is a series of image data (such as consecutive frames in a video).

[0096] LSTM timing analysis:

[0097] LSTM contains three gates: input gate, forget gate, and output gate. These gates determine how information is updated in the LSTM cell.

[0098] The formula for calculating the gate size is: ρ(W) i *X t + U i * h t-1 +b i ), where ρ is the sigmoid function, W i U i , and b i These are weights and biases.

[0099] o uses each frame output of VGG16 as input to LSTM.

[0100] oLSTM processes timing data through its internal "memory cells," providing enhanced timing features for the next step of the Transformer.

[0101] 5. Spatial and global relationship modeling: Transformer

[0102] oTransformer primarily handles spatial and global relationships. Spatial relationships here can refer to the relationships between key regions within a face, while global relationships can refer to the relationship between the entire face and the background, etc.

[0103] The output of oLSTM is used as the input of Transformer.

[0104] oTransformer analyzes various relationships in the input features through its self-attention mechanism.

[0105] Self-attention mechanism: The extracted features are fed into the Transformer model. The Transformer's self-attention mechanism allows for the capture of long-term dependencies, further weighing and identifying key emotional components in the image.

[0106] a. Calculate Query, Key, Value (Q, K, V)

[0107] b. Calculate the attention score using dot-product:

[0108] c. Obtain attention weights by normalizing using the softmax function.

[0109] d. Use these weights to calculate the weighted sum of values:

[0110] Attention Output=softmax(Score)V

[0111] o Obtain the final feature output from the Transformer.

[0112] 6. Sentiment Classification

[0113] Based on the output of the Transformer, a sentiment classification model can be built, such as one or more fully connected layers.

[0114] The probability distribution of each emotion is obtained by using the softmax activation function.

[0115] 7. Output and Explanation

[0116] The output sentiment label with the highest probability is used as the final prediction.

[0117] Optionally, the probability of each emotion can be provided for the user's reference.

[0118] This fusion model combines VGG16's powerful image feature extraction capabilities, LSTM's advantages in temporal data processing, and Transformer's excellent performance in capturing global and spatial relationships. By fusing these three models, a more comprehensive and accurate feature representation can be provided for sentiment prediction.

[0119] Thirdly, this invention has the following complementary advantages:

[0120] 1. VGG16:

[0121] o Deep feature extraction: VGG16 is able to extract a variety of features from low to high levels due to its deep structure.

[0122] o Local feature detection: Early convolutional layers can effectively detect low-level features such as edges and textures, which is especially crucial in emotion detection.

[0123] 2. LSTM:

[0124] o Temporal data processing: LSTM can capture long-term temporal dependencies, making it particularly suitable for processing facial expression changes in consecutive frames.

[0125] o Contextual relationships: It can remember past information and take current input into account, thus better understanding changes in emotion.

[0126] 3. Transformer:

[0127] o Self-attention mechanism: It can capture long-distance dependencies and is particularly suitable for understanding global information in images.

[0128] Parallel computation: Unlike traditional RNNs, all parts of the Transformer model can be computed in parallel, improving efficiency.

[0129] Four: The following key points should be noted when implementing this invention:

[0130] 1. Integration Method: A key issue is how to ensure that the outputs of the three can be effectively combined.

[0131] 2. Computational complexity: Since three models are used, it is necessary to ensure the computational efficiency of the entire model;

[0132] 3. Data alignment: It is crucial to ensure that the output of the VGG16 is aligned with the inputs of the LSTM and Transformer;

[0133] 4. Hyperparameter tuning: Due to the complexity of the model, a lot of hyperparameter tuning may be required to obtain the best performance.

[0134] Fifth: The intelligent emotion analysis device and method proposed in this invention have many advantages and significant effects in accurately recognizing facial expressions to predict emotions.

[0135] Compared to the problems existing in the field of emotion recognition, this invention brings significant improvements and innovations. The following are the beneficial effects of the technical solution of this invention:

[0136] First, this invention employs deep learning technology to fully capture subtle yet crucial changes in facial expressions, thereby achieving more accurate emotion prediction. By introducing the VGG16 model, the system can automatically learn emotion-related features from images, improving prediction accuracy and better capturing subtle differences between different emotional states.

[0137] Secondly, the deep learning model of this invention can better handle complex and diverse emotional states and distinguish subtle emotional changes. Compared to the difficulties traditional methods face when dealing with complex emotions, this technical solution allows the system to accurately identify and predict multiple emotional states, bringing more comprehensive emotional information to practical applications.

[0138] Furthermore, deep learning-based models possess strong generalization capabilities, maintaining stable prediction performance under varying lighting conditions, angles, and scenarios. Compared to the performance degradation of existing technologies in specific environments, the model of this invention maintains high accuracy across diverse situations, enhancing the system's practicality.

[0139] Meanwhile, this invention fully leverages modern computer hardware support, enabling efficient processing in large-scale data and real-time application scenarios. Compared to the processing speed issues in existing technologies, this solution significantly improves real-time performance and efficiency, allowing users to quickly obtain sentiment prediction results.

[0140] Furthermore, this invention provides a complete end-to-end solution, covering the entire process from image input to emotion prediction. Users can input facial images into the system using this device and obtain high-quality emotion prediction results without cumbersome operating procedures, providing users with a convenient and practical way to obtain information about facial emotions.

[0141] Finally, this invention encompasses multiple cutting-edge technological fields such as computer vision, deep learning, and sentiment analysis, and has broad application value in areas such as emotion recognition, human-computer interfaces, and psychological research. The innovative technical solution can not only be used for sentiment analysis but also contribute to promoting the development and research of related fields.

[0142] In summary, the intelligent emotion analysis device and method demonstrate outstanding advantages and effects in areas such as accurate emotion prediction, multi-emotional state recognition, improved model generalization ability, real-time performance and efficiency, end-to-end solutions, and cross-domain application value. These advantages make this invention innovative and practical in the field of emotion recognition technology, providing users with a more accurate, efficient, and comprehensive method for facial emotion analysis.

[0143] This invention closely integrates knowledge and methods from fields such as artificial intelligence, image processing, pattern recognition, and human-computer interaction. It aims to achieve efficient and accurate facial expression recognition through advanced image processing algorithms, deep neural network models, and emotion classification technology, and to predict a person's emotional state accordingly. This innovation encompasses the intersection of computer science and artificial intelligence, aiming to provide more accurate and intelligent solutions for fields such as emotion recognition, human-computer interfaces, and psychological research.

[0144] In the field of computer vision, the face detection and key point localization technology of this invention plays a crucial role, effectively extracting facial regions and feature points, laying a solid foundation for subsequent sentiment analysis. Simultaneously, in the field of deep learning, this invention constructs a deep neural network model targeting facial expression features. Through training on a large amount of data, it can automatically identify, classify, and predict various emotional states. Furthermore, related technologies in the field of sentiment analysis are also an indispensable part of this invention, ensuring that the model can accurately capture the relationship between facial expressions and emotions.

[0145] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A method for accurately predicting human emotions based on facial expression recognition, characterized in that, Includes the following steps: (1) Obtain face images and preprocess the images; (2) Determine the location of faces and important key points in the image using face detection algorithms; (3) Use deep learning and trained VGG16 model to extract deep features from face images; (4) The extracted features are fed into the LSTM model for temporal feature processing; (5) The temporal features processed by the LSTM model are fused in parallel with the deep features extracted by the VGG16 model to form fused features, which are then fed into the Transformer for model space and global relationship modeling, and the final feature output is obtained from the Transformer. (6) Based on the output of Transformer, construct an emotion classification model and use the softmax activation function to obtain the probability distribution of each emotion; (7) Output the sentiment label with the highest probability as the final prediction or provide the probability of each sentiment for the user's reference.

2. The method for accurately predicting human emotions based on facial expression recognition according to claim 1, characterized in that, The steps of deep feature extraction from face images using the VGG16 model include: (1) The input image is processed through multiple convolutional layers, and a 3x3 convolutional kernel is used for convolution operation; (2) Formula for convolution operation: Y=X*K, where X is the input feature map, K is the convolution kernel, and * indicates convolution operation; (3) Next, each convolution operation is followed by a ReLU activation function and a 2x2 max pooling operation; (4) Finally, the image features are processed through three fully connected layers.

3. The method for accurately predicting human emotions based on facial expression recognition according to claim 1, characterized in that, The steps of time series feature processing include: (1) Use each frame output of VGG16 as the input of LSTM; (2) LSTM processes time-series data through its internal "memory units" to provide enhanced time-series features for the next step of Transformer.

4. The method for accurately predicting human emotions based on facial expression recognition according to any one of claims 1-3, characterized in that, The deep learning and training steps for the VGG16 model include: (1) Forward propagation: Input image, extract features step by step through convolution and pooling layers to obtain the final feature representation; (2) Calculate the loss: Use a loss function to measure the difference between the predicted output and the actual label; (3) Backpropagation: The gradient of the parameters is calculated through backpropagation; (4) Parameter update: Update parameters using an optimization algorithm; (5) Iterative training: Repeat the above steps and iterate multiple times until convergence.

5. A system for accurately predicting human emotions based on facial expression recognition, characterized in that: It carries or stores a program that can implement the method for accurately predicting human emotions based on recognizing facial expressions as described in any one of claims 1-4.

6. An apparatus for implementing the method for accurately predicting human emotions based on facial expression recognition as described in claim 1, characterized in that: It includes an image input source module that transmits data in sequence, a preprocessing module, a face detection and key point localization module, a feature extraction module, an emotion classification model module, an emotion prediction module, and an output module; The image input source module is used to acquire the face image to be identified, providing a data foundation for subsequent facial expression recognition and emotion prediction; The preprocessing module is used to process face images through a series of steps, including cropping, scale normalization, and grayscale conversion. These steps transform the images into standardized input data, providing clear and consistent data for feature extraction and the establishment of sentiment classification models. A face detection and key point localization module is used to locate face regions in an image and locate key points of the face. The feature extraction module uses a deep convolutional neural network as the core method to extract high-level features from the preprocessed face image. The emotion classification model module is used to build emotion classification models and predict the emotional state corresponding to facial expressions; The emotion prediction module is used to input the facial features to be predicted into the emotion classification module, and through model learning and reasoning, generate the emotion state prediction result corresponding to the input face. The output module is used to present the prediction results to the user in a readable manner.

Images