At present, high-precision digital thermometers based on industrial platinum resistance have become a popular research direction and are widely used in environmental monitoring, medical health, industrial automation and other fields. However, due to the influence of materials and manufacturing processes, the measurement accuracy is average. With the increase of service life, it is inevitable that the temperature measurement deviation will be caused by the drift of the resistance value. The algorithm of correcting temperature is an effective method to improve the measurement accuracy of digital thermometers. Traditional compensation function correction algorithms such as polynomial fitting and B-spline fitting have good correction effect, but the problems of resistance drift cannot be solved. The segmented linear correction algorithm is simple and easy to implemente, but it requires multi-point temperature measurements. Because of the nonlinear changes of the temperature correction curve, the correction effect is average, which limits its correction accuracy and universality. Therefore, we propose n algorithm of reconstructing temperature correction curve based on the pseudo inverse method. Firstly, the reconstruction matrix is built by using the original data and multiple characteristic temperature points. Then, the complete temperature correction curve is reconstructed by the characteristic temperature points to be reconstructed and the reconstruction matrix. Finally, the reconstructed temperature correction curve is automatically included in the sample database, which improves the diversity of samples and the correction accuracy of the algorithm. Experimental results show that the proposed algorithm has a better correction effect on nonlinear changes and drifts of the temperature correction curve. And the proposed algorithm is less affected by the number of characteristic temperature points and the selection combination. The complete temperature correction curve is well reconstructed by collecting only 4 characteristic temperature points. Therefore, the proposed algorithm can provide the effective support for improving the measurement accuracy of digital thermometer.