Abstract: In agricultural production management, the agricultural Internet of Things can be used to efficiently obtain the growth environment information of crops. However, the distribution distance between terminal nodes in farmland is relatively long. It is limited by the access capacity and equipment power consumption, which brings difficulties to crop environment monitoring. To address the problems exiting in smart agriculture, this paper uses NB-IoT technology to build a wireless sensor network to collect and monitor the crop growth environment in real-time. Furthermore, a method based on multi-sensor data fusion that uses adaptive weighted average fusion and neural network methods to fuse the sensor data is studied. This approach yields a comprehensive assessment of the crop environment. This method can effectively improve the prediction accuracy of crop status. Compared with the traditional method, the prediction error is reduced by more than 45%, which verifies the effectiveness and superiority of this method.