目前，烃源岩有机显微组分的识别主要依赖于人工，不仅准确率受主观性影响，且效率低下。为实现对烃源岩中有机显微组分的自动识别，了解烃源岩中有机质的分布与地球化学特征，提出一种基于YOLOv8s（you only look once version 8 small）的轻量化烃源岩（hydrocarbon source rock, HSR）有机显微组分识别算法HSR-YOLOv8s。该算法提出将YOLOv8s主干网络更换为GHGNetv2（ghost hourglass net version 2）以减少参数量；在颈部C2f结构中融入高效多尺度注意力（efficient multi-scale attention, EMA）机制，设计C2f-EMA模块，使网络模型关注更多上下文信息，加强对特征的提取；引用Shape-IoU（shape intersection over union）作为边界框的回归损失函数，进一步提升边界框预测的准确性。实验表明，该算法的精确率、mAP50、mAP50：95分别达到90%、88.4%、64.4%，参数量为8.36 MB。该算法不仅有较高的检测精度，还具备轻量化的网络结构，适合部署到边缘端实践中。

At present, the identification of organic macerals in hydrocarbon source rocks is mainly done manually. It is not only affected by subjectivity in terms of accuracy but also has low efficiency. In order to achieve automatic identification of organic macerals in hydrocarbon source rocks and help understand the distribution and geochemical characteristics of organic matter in hydrocarbon source rocks, a lightweight source rock organic maceral identification algorithm hydrocarbon source rock-YOLOv8s (HSR-YOLOv8s) was proposed. The backbone network of YOLOv8s was replaced by ghost hourglass net version 2 (GHGNetv2) to reduce the number of parameters. The efficient multi-scale attention (EMA) was incorporated into the neck C2f structure and the C2f-EMA module was designed to make the network model pay more attention to context information and strengthen the extraction of features. Shape intersection over union (Shape-IoU) was quoted as the regression loss function of the bounding box to further improve the accuracy of bounding box prediction. Experiments show that the precision, mAP50, and mAP50:95 of the algorithm reach 90%, 88.4%, and 64.4% respectively, and the number of parameters is 8.36 MB. This algorithm not only has high detection accuracy but also has a lightweight network structure and is suitable for deployment in edge-end practice.