To reduce the impact of temperature drift on sensor measurement results in high temperature and low humidity environments, a humidity sensor based on quartz crystal microbalance (QCM) was developed by using a quartz crystal with a fundamental frequency of 4 MHz as a substrate and depositing graphene oxide (GO) on the substrate using a drop?on?demand method. The temperature drift phenomenon of AT?cut quartz crystal wafers and graphene oxide materials in high temperature environments is significant, resulting in frequency output drift of the sensor. Therefore, a deep?learning method was used to correct the temperature drift. The adaptability of the back propagation (BP) neural network correction model to the QCM humidity sensor was tested under different absolute humidity conditions. The experimental results show that the correction model obtained through deep learning can effectively improve the sensitivity, stability, and response speed of the QCM humidity sensor. It is of great significance for studying the frequency correction technology of QCM humidity sensors under temperature and humidity coupling conditions.