Defect characterization from its non-defective counterpart from the raw thermal response plays a vital role in Quadratic frequency modulated thermal wave imaging (QFMTWI). The strength of the bone reduces due to the skeletal disorder as the age of the person grows, Early diagnosis corresponding to disease is necessary to provide good bone strength. By detecting bone density variations the disease can be managed effectively. A non-stationary thermal wave imaging method, Quadratic frequency modulated thermal wave imaging (QFMTWI) is used to characterize strictness of the human bone, as well as experimentation also carried on Carbon fiber reinforced polymers (CFRP) sample and are extended to unsupervised machine learning algorithms like k-means clustering and fuzzy c-means clustering algorithms. In case of an observer with less expertise, a perfect unsupervised clustering approach is necessary to fulfill this requirement. In present article, we applied k-means and fuzzy c-means based unsupervised clustering techniques for subsurface defect detection in QFMTWI. The applicability of these algorithms is tested on a numerical simulated biomedical bone sample having various density variations and an experimental Carbon fiber reinforced polymers (CFRP) sample with flat bottom holes of different depths with same size. Signal to noise ratio (SNR) is taken as performance merit and on comparison, we conclude Fuzzy c-means provides better detection and characterization of defects compared to K-means clustering for QFMTWI.