On the basis of physiological and physical properties of rice seeds with different aging time, an infrared thermal model for testing rice seed germination rate by multi-scale wavelet transform and grey neural network is proposed to realize fast and non-destructive detection of rice seed germination rate, and to solve the problems of long experimental period, complex operation resulted from traditional germination rate test methods. 144 samples are extracted from germ section of different rice seeds. Detail signal of the third layer wavelet decomposition (d3) is the greatest contribution by analyzing approximation signal and detail signal through a multi-scale wavelet transform. So the detail signal of the third layer wavelet decomposition is used as the model input, and the samples are randomly divided into a calibration set (96 samples) and a prediction set (48 samples). The infrared thermal difference of rice seeds with different aging time is analyzed and compared through partial least squares (PLS), back propagation (BP) neural network, radial basis function neural network (RBFNN) and grey neural network (GNN) to establish infrared thermal prediction models of rice seed germination rate. The results show that the optimal model of germination rate is constructed by GNN artificial neural network, by which the correlation coefficient (RC) and standard deviation (SEC) of the calibration set are 0.9619 and 2.5013 respectively, and the correlation coefficient (RP) and standard deviation (SEP) of the prediction set are 0.9554 and 2.4172 respectively, the relevance reaches a higher level and the error is small. The experimental results show that adopting wavelet decomposition and GNN to establish the infrared thermal prediction model of rice seed germination rate is feasible.