基于深度学习的院区冷水机组指针式仪表读数研究

Research on pointer instrument reading of hospital’s water chilling unit based on deep learning

目的:解决医院后勤部门在巡检时对指针式仪表读数存在的精度低、频率低、效率低等问题。方法:基于开源深度学习框架PaddlePaddle(飞浆),使用目标检测模型和语义分割模型构建适用于医院动力机房实地场景的指针式仪表读数模型,搭建了一套基于深度学习的全自动指针式仪表读数系统,实现对图像内指针式仪表的自动化识别、读数和记录。结果:通过基于深度学习的指针式仪表读数模型可以实现对目标仪表实时识别与读数,模型的预测平均绝对误差(MAE)为0.04,满足了实际工作需求。结论:基于深度学习的指针式仪表识别技术能够有效提高后勤部门对指针式仪表的数据采集效率,进而提高后勤部门的巡检效率。

Objective To solve the problems of low precision, low frequency, and low efficiency of pointer instrument reading in the inspection of the hospital logistics department. Methods Based on the open-source deep learning framework PaddlePaddle, a pointer instrument reading model applicable to the field scenario of hospital power room was constructed by using the target detection model and semantic segmentation model, this study builds a fully automatic pointer instrument reading system based on deep learning, to realize the automatic recognition, reading and recording of the pointer instrument in the images. Results The real-time recognition and reading of the target instrument were realized through the pointer instrument reading model based on deep learning, and the predicted mean absolute error(MAE) of the model was 0.04, which meets the actual work requirements. Conclusion The recognition technology of pointer instruments based on deep learning can effectively improve the data collection efficiency of pointer instruments for the logistics department, and then improve its inspection efficiency.