Abstract

Info

Author(s):: YUE Zhen1; SUN Zhen2; JU Ruiwen2; LI Qingdang3; 1. School of Information Science and Technology, Qingdao University of Science and Technology, Qingdao 266061, China 2. School of Data Science, Qingdao University of Science and Technology, Qingdao 266061, China 3. College of Sino-German Science and Technology, Qingdao University of Science and Technology, Qingdao 266061, China

Abstract:: Abstract:To address the challenges of detecting tuberculosis pathogens in sputum smear images, such as complex backgrounds, small targets, and high costs of manual screening, a detection method based on YOLOv8s is presented. The structure is improved through spatial and channel reconstruction convolutions to limit feature redundancy, and a coordinate attention is introduced to expand the receptive field of the model. Furthermore, a spatial pyramid pooling cross-stage partial network is used to extract feature information at different scales, and a normalized attention mechanism is embedded to suppress less significant features. The experimental results on a public dataset show that compared with the original YOLOv8s model, the improved network model enhances precision and recall rates by 2.7% and 1.5%, respectively, and improved mean average precision at a confidence level of 0.5 by 2.3%, demonstrating that the improved model can effectively assist radiologists in diagnosis.