Abstract:The novel coronavirus variant, which is also widely known as COVID-19, is currently a common threat to all humans across the world. Effective recognition of COVID-19 using advanced machine learning methods is a timely need. Although many sophisticated approaches have been proposed in the recent past, they still struggle to achieve expected performances in recognizing COVID-19 using chest X-ray images. In addition, the majority of them are involved with the complex pre-processing task, which is often challenging and time-consuming. Meanwhile, deep networks are end-to-end and have shown promising results in image-based recognition tasks during the last decade. Hence, in this work, some widely used state-of-the-art deep networks are evaluated for COVID-19 recognition with chest X-ray images. All the deep networks are evaluated on a publicly available chest X-ray image dataset. The evaluation results show that the deep networks can effectively recognize COVID-19 from chest X-ray images. Further, the comparison results reveal that the EfficientNetB7 network outperformed other existing state-of-the-art techniques.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to effectively identify COVID - 19 by using advanced machine - learning methods, especially through chest X - ray images. Although many sophisticated identification methods have been proposed in recent years, they still have difficulty achieving the expected performance in identifying COVID - 19 using chest X - ray images, and most of these methods involve complex pre - processing tasks, which are often challenging and time - consuming. Therefore, this paper evaluates the performance of several widely - used state - of - the - art deep networks in identifying COVID - 19 using chest X - ray images, in the hope of finding a more effective method.
Specifically, the main problems mentioned in the paper are:
1. **Limitations of existing methods**: Existing COVID - 19 identification algorithms usually require complex pre - processing steps, which not only increase the time and computational costs but also may lead to data distortion.
2. **Advantages of deep networks**: Deep networks are selected as research objects because of their end - to - end training ability and excellent performance in image classification tasks.
3. **Evaluating the performance of different deep networks**: By comparing the performance of multiple deep networks on the same data set, find the model most suitable for COVID - 19 identification.
The main contributions of the paper include:
- Evaluating the performance of multiple state - of - the - art deep networks (such as VGG16, VGG19, DenseNet121, DenseNet201, InceptionV3, ResNet101, ResNet152, Xception, EfficientNetB0, EfficientNetB7, NASNetLarge, NASNetMobile, MobileNetV2, MobileNetV3 Small and MobileNetV3 Large) on chest X - ray images.
- Through the analysis of experimental results, it is found that the EfficientNetB7 network performs excellently in identifying COVID - 19, achieving a classification accuracy rate of 95.54%.
In conclusion, this paper aims to find an efficient, accurate and easy - to - implement method to identify COVID - 19 by comparing different deep - learning models, thereby providing technical support for timely diagnosis and epidemic control.