Fiber is effective in restricting cracks and improving the toughness of geopolymer composites, but few studies have focused on the surface crack characteristics of fiber‐reinforced geopolymer composites. In this paper, after flexural tests of polypropylene fiber‐reinforced geopolymer mortar, the surface cracking image was collected by a digital camera and cracking information was extract by deep learning. Finally, the cracking and fractal characteristics were specifically discussed. The semantic segmentation network can accurately extract surface cracks for calculating various parameters. The results showed that the mean intersection over union (mIoU) and mean pixel accuracy (mPA) of the cracks are 0.8451 and 0.9213, respectively. Generally, the crack length, width, area, and fractal dimension of the specimen are all increased with the increase in the fiber volume fraction. These crack parameters grow rapidly when the fiber content is small, and the growth of the crack parameters gradually slows down as the fiber volume fraction increases to approximately 1.5%. The highest crack parameter values were found in the geopolymer mortar, with a 0.48 water–binder ratio and 12 mm fiber length. The variation of the bottom crack length and the side crack fractal dimension can be used to represent the overall crack variation patterns. Meanwhile, the crack parameters increase with the increased fiber factor in a quadratic function. Based on these crack parameters, the critical fiber factor and dense fiber factor of polypropylene fiber‐reinforced geopolymer mortar were 200 and 550, respectively. They are greater than those of fiber‐reinforced Portland cementitious composites. The influence of various crack parameters on the flexural strength is in the order of the crack area, width, length, and fractal dimension.