Detection of Tubercle Bacilli by Fusion with YOLOv5s and ESRGAN

In order to solve the problem of small target and complex background in target detection, this paper proposes a fusion algorithm based on YOLOv5s target detection method and ESRGAN image super-resolution method -ESRGAN-YOLOv5s algorithm. The algorithm adds ESRGAN network to the input layer of YOLOv5s network, and then performs super-resolution data enhancement on the Mycobacterium tuberculosis data set input into the model; Then add channel and spatial attention module CBAM (Convolutional Block Attention Module) to the FPN (Feature Pyramid Networks) feature pyramid network of the model to improve the ability of the model to extract the characteristics of Mycobacterium tuberculosis; Finally, a large-scale feature map is added to the detection head of the model for the detection of tuberculosis bacilli. The experimental results show that the average accuracy (mAP) of tuberculosis detection based on the fusion of YOLOv5s and ESRGAN is 85.9%, which is 2.5% higher than the original YOLOv5s model. The improved tuberculosis detection algorithm has a better recognition effect on tuberculosis targets.