Pulse-compression thermography is an emerging non-destructive technique whose effectiveness strictly depends on the choice of the coded excitations used to modulate the heating stimulus. In this paper, the features of frequency-modulated coded signals, i.e., chirps, have been tested for imaging thin Teflon defects embedded within a carbon fiber composite specimen. With the aim of maximizing the heat transferred within the sample, the use of several optimized non-linear chirp signals has been also investigated and their defect detection capability compared in terms of the maximum achievable signal-to-noise ratio.