Convolutional Neural Network for emotion recognition to assist psychiatrists and psychologists during the COVID-19 pandemic: experts opinion

TL;DR

This paper presents a real-time emotion recognition web application using CNNs to assist mental health professionals during COVID-19, focusing on computational efficiency and usability.

Contribution

It introduces ResmoNet, a lightweight CNN model optimized for low-resource environments, and evaluates its performance and usability in a clinical setting.

Findings

ResmoNet has the lowest parameters, FLOPS, and memory usage.

EDNN slightly outperforms ResmoNet in response time.

The web app received good usability and utility scores from professionals.

Abstract

A web application with real-time emotion recognition for psychologists and psychiatrists is presented. Mental health effects during COVID-19 quarantine need to be handled because society is being emotionally impacted. The human micro-expressions can describe genuine emotions that can be captured by Convolutional Neural Networks (CNN) models. But the challenge is to implement it under the poor performance of a part of society computers and the low speed of internet connection, i.e., improve the computational efficiency and reduce the data transfer. To validate the computational efficiency premise, we compare CNN architectures results, collecting the floating-point operations per second (FLOPS), the Number of Parameters (NP) and accuracy from the MobileNet, PeleeNet, Extended Deep Neural Network (EDNN), Inception- Based Deep Neural Network (IDNN) and our proposed Residual mobile-based Network model (ResmoNet). Also, we compare the trained models results in terms of Main Memory Utilization (MMU) and Response Time to complete the Emotion (RTE) recognition. Besides, we design a data transfer that includes the raw data of emotions and the basic patient information. The web application was evaluated with the System Usability Scale (SUS) and a utility questionnaire by psychologists and psychiatrists. ResmoNet model generated the most reduced NP, FLOPS, and MMU results, only EDNN overcomes ResmoNet in 0.01sec in RTE. The optimizations to our model impacted the accuracy, therefore IDNN and EDNN are 0.02 and 0.05 more accurate than our model respectively. Finally, according to psychologists and psychiatrists, the web application has good usability (73.8 of 100) and utility (3.94 of 5).