Aiming at the problems of low contrast and easy loss of small defects in the traditional phase locked thermal imaging defect feature extraction algorithm,a defect detection algorithm based on the combination of robust principal component analysis (RPCA) and FFT is proposed,and the RPCA model is solved by the inexact augmented Lagrange multiplier method (IALM).The original infrared thermal wave sequence vector is transformed into a two dimensional matrix,and the data is decomposed into two parts by RPCA.The low rank matrix that approximates the extraction of the non uniform background,and the sparse matrix that reflects the defective information,and the magnitude and phase maps of the non uniform background are obtained by using the FFT on the obtained sparse matrix,which is aimed at the problem that IALM needs to artificially introduce the initial value to solve the RPCA model,which affects the optimization results.Tyrannosaurus optimization algorithm (TROA) is used to construct the fitness function by selecting the signal to heterodyne gain and the background suppression factor,and to optimize the initial equilibrium parameters and the penalty factor.The experimental results show that the image obtained by this algorithm has outstanding contrast,obvious information of small defects,and better objective evaluation indexes than other algorithms,in which the entropy value has been greatly reduced,effectively suppressing the non uniform background of the heat wave image.