针对传统烟雾探测器易受船舶机舱环境、布置位置等因素影响导致着火响应不及时的问题，提出一种基于机器视觉的船舶机舱烟雾识别模型。首先，利用通用数据库和实船场景及其烟雾图片建立图形知识库；其次，构建迁移学习和残差网络的融合模型，实现烟雾特征的迁移和学习，并利用验证数据集验证模型的有效性；最后，选取实船工作间火灾实例照片验证模型的有效性。结果表明，相较于传统感烟探测器，该模型提前83 s预警；相较于其他智能算法，该模型可提升烟雾识别准确率。该方法可以更快速识别船舶烟雾，避免重大灾难性火灾事故的发生，可作为船舶智能监控手段。

In order to solve the problem that the traditional smoke detectors were easily affected by the environment and location of the ship engine room, resulting in untimely fire response, a smoke recognition model of the ship engine room based on machine vision was proposed. Firstly, the universal database, real ship scenes and smoke images were used to build a graphic knowledge base. Secondly, the fusion model of migration learning and residual network was constructed to realize the migration and learning of smoke characteristics, and the validity of the model was verified by using the validation data set. Finally, the fire photos of a real ship workshop were selected to verify the validity of the model. The results show that compared with the traditional smoke detector, the proposed model can provide 83 s early warning, and compared with other intelligent algorithms, this model can improve the accuracy of smoke recognition. This method can identify ship smoke more quickly and avoid major catastrophic fire accidents, therefore can be used as an intelligent monitoring method for ships.