基于改进的YOLOv5肺结节检测方法

Based on the Improved YOLOv5 Lung Nodule Detection Method

摘要: 近年来随着深度学习的不断发展，基于YOLO的算法成为当前在目标检测的主流方法，但针对肺结节这种小目标的检测，检测精度和检出率有待进一步提升。因此对YOLOv5网络模型进行改进，在骨干网络中加入注意力机制，提高肺结节在浅层网络的特征提取能力，并在颈部网络中使用加权双向的特征金字塔网络进行特征融合，将不同层次的特征进行融合，实现整体算法的提升。最后通过实验证明改进后的方法在肺结节检测上获得了较好的检测精度的查全率。

Abstract: In recent years, with the continuous development of deep learning, the YOL-based algorithm has become the mainstream method in target detection. However, for the detection of small targets such as pulmonary nodules, the detection accuracy and detection rate need to be further improved. Therefore, the YOLOv5 network model was improved, the attention mechanism was added into the backbone network to improve the feature extraction ability of pulmonary nodules in the shallow network, and the weighted bidirectional feature pyramid network was used in the neck network for feature fusion to integrate features at different levels, thus achieving the improvement of the overall algorithm. Finally, experiments show that the improved method can obtain a better recall rate in the detection of pulmonary nodules.