Emotion Detection Using Deep Learning

EasyChair Preprint 10416

Abstract

An important area of research is the detection and recognition of emotions using visual features extracted from facial expressions. This paper presents a project that focuses on developing an emotion recognition model using the ResNet50 deep learning architecture and training it on the AffectNet dataset. The achieved accuracy of 87% demonstrates the effectiveness of the proposed approach. The applications of this project are wide ranging from human-computer interaction to psychology, medicine, education and crime detection.

The paper also highlights future directions for improving model accuracy and performance. Suggestions include modifying the model layers and exploring larger and more diverse datasets to improve the training process. Additionally, the integration of multimodal data, such as combining facial expressions with voice analysis or physiological signals, holds promise for improving the robustness and accuracy of emotion recognition systems. Cultural and contextual factors that influence emotional expressions should be taken into account to develop more culturally sensitive models. Additionally, optimizing the model for real-time deployment on resource-constrained devices such as smartphones or wearables can expand the practical applications of emotion recognition systems.

By focusing on these future research directions, this paper aims to contribute to the development of emotion detection and recognition systems, which will ultimately lead to more accurate and versatile applications in various fields.

Keyphrases: Emotion Detection, facial expression recognition, visual features

@booklet{EasyChair:10416,
  author    = {Shraddha Belhekar and Priya Patil and Snehal Hulule and Tanvi Ghare and Dhammjyoti Dhawase},
  title     = {Emotion Detection Using Deep Learning},
  howpublished = {EasyChair Preprint 10416},
  year      = {EasyChair, 2023}}