The paper presents the results of the analysis of the thermal cracking patterns and fracture zone of microsilica-modified cement matrix made of the low-alkali cement. The material was subjected to a two-stage thermal load. To identify cracks an original, the image double-segmentation procedure, using machine learning algorithms was developed. To characterize the cracks structure, the fractal geometry was applied, measuring the fractal dimension of the cracking patterns. Moreover, the total crack area and the crack density were analyzed. In the case of the fracture line analysis the fractal dimension was examined in two variants. The applied parameters allowed to describe the complexity degree of the surface cracks structure. The process of development of thermal cracks caused by repeated thermal loading was also characterized. The obtained results showed strong correlation between the tensile strength, fractal dimension of the cracking pattern and crack density. Knowledge in this area allows to estimate mechanical properties of the cement matrix on the basis of measurement and evaluation of morphology of the thermal cracking patterns.