Main Article Content
Introduction: Stress has become a significant issue in the current society anddamaging human life. This is because of work, family, personal pressures as well as surrounding environments. Image-based stress detection is emerging with the help of computer-aided systems.
Background:Here we proposed ahybrid deep learning technique to detect and analyze the effects of stress. The combination of these two deep neural networks that are Alexnet and Vgg-16 are pre-trained, The main advantages are the input of the Vgg-16 networks is feature maps.
Methods:The first network (DNN-1) input is the raw image, and the second network (DNN-2) input is the Frequency features which are created by wavelet transform. The frequency features help to differentiate the frequency-related information from the images. Finally, these two networks are combined for better classification results. Both the network suitable for detecting stress but the combined network provides better performance in terms of accuracy because the features have been extracted from both the network and combined with its mean value so both features are available for the classification. it will boost the detection results.
Results:Also, we experiment with Alexnet and VGG-16 and combined network separately. the combined or hybrid network provides 96.2 % accuracy which is higher than the separate network and existing machine learning techniques like SVM and KNN as well conventional deep learning techniques.
Conclusion: This method can be used in health care systems for identifying the human stress for further treatments.