A novel IoT-based deep neural network for COVID-19 detection using a soft-attention mechanism

Learning models from COVID-19 data are conducive to understand this disease. However, the scarcity of labeled data presents certain challenges. Previous works have exploited existing deep neural network models that are pre-trained on large datasets like the ImageNet dataset. However, the generalization of the pre-trained models remains a challenge. The objective of this study is to develop an accurate and reliable model that improves diagnostic accuracy and reduces the chances of misdiagnosis. This, in turn, enables appropriate and timely medical interventions for COVID-19 patients. In this paper, a novel framework is proposed to monitor and predict COVID-19 cases that relies on (1) a layered software architecture and (2) a deep neural network model for data processing. The proposed deep neural network model is based on a pre-trained RegNet model. However, the RegNet has limitations in effectively capturing complex shapes. The receptive field may not handle enough shape. To address this issue, we construct a new block using commonly used convolutional and max-pooling layers. It also incorporates the attention mechanism. This mechanism allows us to control a large receptive field with limited computational resources, highlight relevant features and enhance the discriminative power of the model. Comparative experiments using four different benchmark datasets have shown promising results. The proposed model exhibits high efficiency in accurately distinguishing COVID-19 images, with accuracy ranging from 96.43% to 98.96%. It is advisable that future works explore our proposed framework for more detection problems.