流体包裹体对油气资源评价、油藏地球化学、流体类型、流体来源与勘探等都具有重要的指导意义。然而流体包裹体识别主要依赖于人工寻找，这种方法费时费力。为了解决这个问题，提出了一种改进的YOLOv5s流体包裹体目标检测算法。将原始的YOLOv5s模型中的特征提取网络部分和特征融合网络部分进行改进，提高模型的检测能力，使模型更适用于流体包裹体的检测。在特征提取网络部分加入了坐标注意力机制提高定位和识别能力；在特征融合网络部分将原始模型中的路径聚合网络换成双向特征金字塔网络，改进后的网络具有更强大的特征融合能力，可提升小目标的检测能力。通过实验结果表明，与原YOLOv5s模型比较，改进后的YOLOv5s平均精度由75.3%提升到77.3%，比原算法的平均精度提高了2%，检测速度由58.14帧/s帧提升到62.89帧/s，提升了4.75帧/s，实现了更准确高效的流体包裹体检测。

Fluid inclusions have important guiding significance for oil and gas resource evaluation, reservoir geochemistry, fluid types, fluid sources, and exploration. However, The identification of fluid inclusions primarily relies on manual searching, a process that is time-consuming and labor-intensive. To address this issue, an improved YOLOv5s fluid inclusion object detection algorithm was proposed. The feature extraction and feature fusion components of the original YOLOv5s model were enhanced to improve the model's detection capability, making it more suitable for fluid inclusion detection. A coordinate attention mechanism was introduced in the feature extraction component to enhance localization and recognition capabilities. Additionally, the original path aggregation network in the feature fusion component was replaced with a bidirectional feature pyramid network. The upgraded network possesses stronger feature fusion capabilities, thereby enhancing the detection capability of small targets. Experimental results demonstrate that compared to the original YOLOv5s model, the average precision of the improved YOLOv5s increases from 75.3% to 77.3%, representing a 2% improvement over the original algorithm. The detection speed also improves from 58.14 frames/s to 62.89 frames/s, resulting in a 4.75 frames/s improvement, thus achieving more accurate and efficient fluid inclusion detection.