This study aims at developing an experimental method to characterise the fluid dynamic stress acting on particles in a lab-scale stirred tank reactor. The method is based on a photo-optical inline particle size measurement technique, which is able to capture the complex morphology of a stress-sensitive clay-floc system. The novel image-based analysis uses a Convolutional Neural Network (CNN), which was trained for particle detection. Its output provides a high-quality particle recognition for subsequent size and shape analysis throughout the flocculation process under various experimental conditions. The flocculation system is used to characterise the stress induced by four impeller geometries in the reactor. The obtained changes of floc size and shape allow to evaluate the particle stress in more detail compared to previous studies. The agreement of main findings of the floc breakage process with known literature assesses the developed method to be applied successfully in further studies.