With the continuous expansion of the application range of gob–side entry retaining technology, the depth, height, and advancing speed of coal seams also increase, which brings great problems to the stability control of surrounding rock structures of gob–side entry retaining. As one of the main bearing structures of the surrounding rock, the stability of the roadway–side support body is a key factor for the success of gob–side entry retaining. In order to study the deformation characteristics and instability mechanism of roadway-side support body, based on the roadway–side support materials of gob-side entry retaining, the dynamic expansion test of back–filling concrete cracks under uniaxial compression was carried out. The YOLOv5 algorithm was applied to establish the fine identification and quantitative characterization method of macroscopic cracks of the samples, and the dynamic expansion rule of roadway-side support body cracks and its dimensional effect were revealed by combining the fractal theory. The results show that the F1 value and average precision mean of the intelligent dynamic crack identification model reached 75% and 71%, respectively, the GIoU loss value tends to fit around 0.038, and the model reached the overall optimal solution. During the uniaxial compression process, micro cracks on the surface of the back–filling concrete first initiated at the end, and after reaching the yield stress, the macroscopic cracks developed significantly. Moreover, several secondary cracks expanded, pooled, and connected from the middle of the specimen to the two ends, inducing the overall instability of the specimen. The surface crack expansion rate, density, and fractal dimension all show stage change characteristics with the increase in stress, and the main crack expansion rate has obvious precursor characteristics. With the increase in the size, the decrease in crack density after back–filling concrete failures gradually decreases from 93.19% to 4.08%, the surface crack network develops from complex to simple, and the failure mode transits from tensile failure to shear failure. The above research results provide a basic experimental basis for design optimization and instability prediction of a roadway–side support body for engineering-scale applications.