Objective In the process of accurate monitoring of health status changes of the elderly， the basic error back propagation （BP） neural network is difficult to accurately calibrate the measurement signals of flexible sensors， resulting in easy overfitting phenomenon， which makes the calibrated signals have a large root mean square error （RMSE）. Therefore， a calibration method based on improved genetic GA-BP neural network is designed for flexible sensors for elderly health monitoring.Method The Kalman filtering algorithm and sliding average filtering algorithm were combined to perform hybrid filtering processing on the real-time measurement signals of the flexible sensors to obtain the effective signals with noise interference removed. The pre-processed signal was converted into multiple signal sub-sequences by segmentation operation， and the signal root mean square value and fluctuation coefficient were calculated to complete the signal feature vector extraction. With BP neural network as the core， the calibration model of flexible sensor measurement signal was constructed， and the improved genetic algorithm（GA） was applied to optimize the model parameters to improve the performance of the network model， and the feature vectors were input to automatically predict the changes of the health monitoring signals in the future moments， and the calibration can be completed by comparing with real-time measurement signals.Result The experimental results show that after applying the method to calibrate the health monitoring signals of the elderly given by the flexible sensors， the RMSE values of the measured signals are lower than 0.07.Conclusion The proposed improved GA-BP neural network calibration method for measurement signals satisfies the signal error calibration requirements.