Multi-level transfer learning for improving the performance of facial emotion recognition

Abstract

Abstract: The present invention is a multi-level transfer learning for improving the performance of facial emotion recognition is a method claim using transfer learning reduces target-domain data dependency, creating target learners and has greatly enhanced sophisticated technologies in computer vision (CV) and natural language processing (NLP). The transfer learning to ResNet architecture usually involves fine-tuning a source domain model with target domain data and which introduces a fine-tuning-based paradigm multilevel transfer learning (mLTL). An essential facts and principles about the training sequence of linked domain datasets and proved its usefulness by conducting face emotion and named entity identification tasks and MLTL-based deep neural network models outperformed the original models on target tasks, according to experiments.

Specification

Description:Title of the Invention
Multi-level transfer learning for improving the performance of facial emotion recognition

Field of the invention:
The present invention generally relates to the field of a facial emotion recognition, particularly relates to a multi-level transfer learning for improving the performance of facial emotion recognition.

Prior art to the invention:

US20240177523 - Titled - "System and Methods of Predicting Personality and Morals through Facial Emotion Recognition" discloses in the present invention is a human personality and morals prediction system. In particular, the invention discloses a system of predicting personality and morals through facial emotion recognition. A computer-implemented system for predicting personality and morals through facial emotion recognition, comprising a machine learning system that predicts personality characteristics of individuals on the basis of their face also disclosed are methods of tracking the emotional response of the individual's face through facial emotion recognition (FER) while watching a series of 15 short videos of different genres; and calibration of emotional responses for analysis through their facial expression.

None of the above-mentioned prior arts neither teaches nor discloses about a multi-level transfer learning for improving the performance of facial emotion recognition.

wherein, the present invention is a multi-level transfer learning for improving the performance of facial emotion recognition.

Objects of the invention:
The principle objects of the present invention a multi-level transfer learning for improving the performance of facial emotion recognition.

Summary of the invention:
Thus the basic aspect of the present invention provides a transfer learning has several applications, making it a potential machine learning subject. Its efficacy has inspired several methods. Transfer learning improves target learners' performance by transferring information from relevant source domains. So, we may improve model generalization by using data from different domains or activities. Transfer learning reduces target-domain data dependency, creating target learners. Transfer learning has greatly enhanced sophisticated technologies in computer vision (CV) and natural language processing (NLP). Transfer learning to ResNet architecture usually involves fine-tuning a source domain model with target domain data. We introduces a fine-tuning-based paradigm multilevel transfer learning (mLTL). We concluded the essential facts and principles about the training sequence of linked domain datasets and proved its usefulness by conducting face emotion and named entity identification tasks. mLTL based deep neural network models outperformed the original models on target tasks, according to experiments.

Brief description of drawings:
Figure 1 is a representation of a multi-level transfer learning for improving the performance of facial emotion recognition
Figure 2 is a representation of a results in the table


















Detailed description:

The present invention as herein describes about to a foreign object detection in an image analysis

The present invention provides a proposed innovation is implemented in three phases i.e., Dataset utilization, face detection algorithm and output results for each level mLTL. These FER datasets were utilized in the mLTL experiment: An expansion of the Cohn-Kanade (CK) dataset, CK+: + is a foundational dataset for emotions. When comparing CK with CK+, the number of participants increases by 27% and the number of sequences increases by 22%. The phrases aimed at FACS and the emotion are stored for each sequence. The labels have been reviewed and approved. GFE2019: Emotions, the data acquired for this experiment includes 1,275 photographs of consulting with professional e-sports coaches and players. A total of eight primary facial expressions-four happy and four sad-are included in the dataset. Delightful, excited, astonished, and neutral/flow are examples of pleasant feelings, whereas furious, bewildered, sad, and frustrated are examples of negative emotions. Face detection: Dense_FaceLiveNet is a specialized model used in computer vision for real-time face detection and recognition, Dense_FaceLiveNet is designed for applications that require accurate and efficient face detection and recognition in real-time environments. This study's experiment utilized data augmentation to randomly rotate a picture from -5◦ to 5◦, flip it horizontally, and zoom in order to tackle the issue of training data shortage. At random into it by a factor of 1-1.5, which significantly raises the quantity of data used for training in the first dataset. To prevent the training model from being overfit, we averaged the outcomes from each training session and utilized K-Folds cross validation as our training model. The early stopping skill requires a learning rate of 0.01 along with 12 batches, 70 epochs, and a patience value of 15. Complex Face LiveNet. For mLTL and CK+ combinations, GFE2019 is the target dataset because of its huge number of categories and high application value.
Results for DNN model with multi-level transfer learning
C=ck+
G=GFE2019

Transfer learning has several applications, making it a potential machine learning subject. Its efficacy has inspired several methods. Transfer learning improves target learners' performance by transferring information from relevant source domains. So, we may improve model generalization by using data from different domains or activities. Transfer learning reduces target-domain data dependency, creating target learners. Transfer learning has greatly enhanced sophisticated technologies in computer vision (CV) and natural language processing (NLP). Transfer learning to ResNet architecture usually involves fine-tuning a source domain model with target domain data. We introduces a fine-tuning-based paradigm multilevel transfer learning (mLTL). We concluded the essential facts and principles about the training sequence of linked domain datasets and proved its usefulness by conducting face emotion and named entity identification tasks. MLTL-based deep neural network models outperformed the original models on target tasks, according to experiments.

Embodiments of the present invention will now be described in more detail with reference to the drawings. The following description is for convenience of understanding of the present invention, and the present invention is not limited by this.
, Claims:Claims:
1. We claim,
A multi-level transfer learning for improving the performance of facial emotion recognition is a method claim using transfer learning reduces target-domain data dependency, creating target learners and has greatly enhanced sophisticated technologies in computer vision (CV) and natural language processing (NLP),
wherein, the transfer learning to ResNet architecture usually involves fine-tuning a source domain model with target domain data and which introduces a fine-tuning-based paradigm multilevel transfer learning (mLTL),
wherein, an essential facts and principles about the training sequence of linked domain datasets and proved its usefulness by conducting face emotion and named entity identification tasks and MLTL-based deep neural network models outperformed the original models on target tasks, according to experiments.

Documents