Spectral estimation has received extensive research in the field of spectral imaging. A spectral estimation method based on adaptive weighted linear regression is proposed. First, a global training model is developed based on pseudo-inverse algorithm, homogeneous polynomial expansion of digital response, and Tikhonov regularization constraint. Second, based on the influence of global training sample on the accuracy of spectral estimation, the adaptive weighted training model is developed to further improve the spectral estimation accuracy by introducing the Gaussian weighted method. The sensitivity functions of 28 digital cameras is used to build a simulation imaging system, the Munsell Matte color sample and the X-rite ColorChecker SG color chart are used as experimental samples, and the metrics of spectral root-mean-square error and color difference are used to evaluate the spectral estimation accuracy. The results show that for the global training model, the new method can achieve almost the same accuracy as the existing methods, and can overcome the exposure sensitivity problems of the existing methods. The new method achieves better results than the existing methods in both spectral accuracy and colorimetric accuracy under the adaptive weighted training model.