YOLO lung CT disease rapid detection classification with fused attention mechanism

Currently, although the use of convolutional neural networks (CNN) for detecting lung infection has improved the detection performance and efficiency, it still has certain shortcomings, low feature utilization for images or difficulty in focusing key features. An effective YOLO algorithm with fused attention mechanism is proposed for lung CT images to detect normal, common pneumonia and COVID-19 images to address the above problems. The YOLO with fused attention mechanism is mainly divided into two parts for model training and experiments: the first step performs lung segmentation of chest CT images and data cleaning of CT images based on physician diagnostic image values; the second step uses the cleaned lung CT images for training and model evaluation of the Yolov5 model with fused attention mechanism (CBAM). We use a series of operations such as binarization, expansion erosion and connected domain segmentation for initial segmentation and filtering of lung images, and incorporate the attention mechanism into the YOLO model, which enables the model to better focus on key features and avoid interference from erroneous data. The results on the COVID-19x dataset show that the YOLO model with the fused attention mechanism detects classification with an accuracy rate of 0.85 and a recall rate of 0.88. In summary, the fused attention mechanism YOLO outperforms the original YOLO model by 6.5% in accuracy and 8.8% in recall, which helps clinicians diagnose lung inflammatory infections in a timely manner type.