Tuberculosis is one of the ten leading causes of death worldwide.To address the complex background and small target size of Mycobacterium tuberculosis in sputum smear images,in this paper,we put forward a YOLOv7 detection network for detecting Mycobac-terium tuberculosis using sputum smear images.In the backbone network,CBAM attention modules are embedded after efficient aggre-gation networks at different scales to extract features at different scales.We introduce separation and merging operations in neck net-works to improve MPConv structure and reduce image feature loss due to convolutional downsampling in deep networks,and introduce Gaussian distribution distances in the head mesh to avoid small target misses due to overlapping bounding boxes.This technique is ex-ceedingly precise and rapid,rendering it ideal for identifying Mycobacterium tuberculosis.Experiments indicate that the mean average accuracy achieved by the improved model is 87.8％,a 5.1％enhancement compared to the baseline network model,and better than similar algorithms,which is important for advancing intelligent detection of Mycobacterium tuberculosis.
关键词
结核杆菌检测/YOLOv7/注意力机制/分离合并操作/高斯分布
Key words
Mycobacterium tuberculosis detection/YOLOv7/attention mechanism/separation and merger operations/Gaussian distribu-tion
维普
万方数据