Research Article
Pneumonia Detection Algorithm Based on Improved YOLOv3
@INPROCEEDINGS{10.1007/978-3-030-94182-6_22, author={Hailong Liu and Jinrong Cui and Chaoda Peng}, title={Pneumonia Detection Algorithm Based on Improved YOLOv3}, proceedings={IoT and Big Data Technologies for Health Care. Second EAI International Conference, IoTCare 2021, Virtual Event, October 18-19, 2021, Proceedings, Part II}, proceedings_a={IOTCARE PART 2}, year={2022}, month={6}, keywords={Object detection Pneumonia Deep learning YOLOv3}, doi={10.1007/978-3-030-94182-6_22} }- Hailong Liu
Jinrong Cui
Chaoda Peng
Year: 2022
Pneumonia Detection Algorithm Based on Improved YOLOv3
IOTCARE PART 2
Springer
DOI: 10.1007/978-3-030-94182-6_22
Abstract
Pneumonia is a kind of disease caused by bacteria, viruses and other pathogens, which can seriously endanger human health, and has strong infectivity. Timely and accurate detection of pneumonia symptoms can not only make patients receive timely treatment, but also prevent the disease from spreading to others. This paper proposes an improved object detection algorithm YOLOv3-P which was based on YOLOv3. Using the idea of Path Aggregation Network (PANet) for reference, after feature fusion, the location information is enhanced through a bottom-up path, which makes full use of the feature information of each layer. And the better backbone network CSPDarkNet53 was used to replace DarkNet53 of YOLOv3, to better extract features from the pneumonia images. Experiments on the lung X-ray image data set provided by the North American Society of Radiology show that the average precision of the algorithm reaches 50.43%, which was improved compared with the YOLOv3 algorithm, and has good performance compared with other common object detection algorithms. YOLOv3-P can help doctors judge the location of pneumonia tissue faster and more accurately.
