Abstract: The free face serves as the core controlling condition for blast energy release, stress-wave reflection, and rock fragmentation. The number of free faces directly determines the distribution of the explosion energy, dynamic evolution of the fracture field, and final fragmentation characteristics, which play critical roles in blasting efficiency, mining safety, resource recovery rate, and subsequent shoveling and transportation efficiency in deep mines. With the increasing depth of underground mining, the triaxial stress of the rock mass increases, and the clamping effect strengthens. Traditional blasting often suffers from uneven fragmentation, high boulder yield, and low energy utilization efficiency. To reveal the influence mechanism of the number of free faces on the evolution of the explosion fracture field and fragmentation distribution, this study used a digital laser dynamic caustics experimental system to conduct explosion-loading model tests on 0–4 free faces. High-speed photography was used to capture the entire process of crack initiation, propagation, branching, and coalescence. Fractal theory and the MATLAB box-counting dimension program were used to quantitatively describe the complexity of fracture networks. In addition, an intelligent blasting fragmentation identification system was developed based on the segment anything model (SAM) deep learning model to achieve high-precision automatic calculation of fragmentation distribution, uniformity, boulder yield, and other indicators. Finally, field tests of side-hole dual-energy presplit blasting were conducted in the thin vein of the Gongchangling underground iron mine to verify the improvement effect of the multi-free-face control technology on fragmentation performance. The test results showed that the number of free faces was positively correlated with the crack propagation velocity, dynamic stress intensity factor, and fracture fractal dimension. The specimen without a free face had sparse, short, disordered, and unconnected cracks. With the increase in free face, the directionality of crack propagation was enhanced, branch density was increased, and the penetration range was expanded. The specimen with four free faces formed a fully penetrated and dense fracture network, leading to significantly improved fragmentation. The peak crack propagation velocity increased from 624 to 865 m/s, an increase of 38.6%. The peak dynamic stress intensity factor rose from 2.8 to 4.8, an increase of 71.4%. The fractal dimension of the fracture network increased from 1.11503 to 1.42025, indicating continuous improvement in fragmentation sufficiency and fracture complexity. Engineering application results demonstrate that side-hole dual-energy presplit blasting can actively create a controllable free face, optimize the release path of explosion energy, and significantly improve the uniformity of fragmentation distribution. The boulder yield decreased from 19% to 4%. The nonuniformity coefficient dropped from 18.3 to 9.4, a reduction of 48.6%. The intelligent fragmentation identification system based on the SAM model achieved a segmentation accuracy of over 95% and a dimension correction error of less than 1%, which can quickly complete muck pile image collection, ore segmentation, parameter extraction, and effect evaluation to provide reliable data support for the intelligent optimization of blasting parameters. The research findings clarify the control law of the number of free faces on the evolution of the explosion fracture field and propose a side-hole presplit blasting technology suitable for thin-vein mines. This study provides a theoretical basis and engineering reference for precise blasting, intelligent rock breaking, and safe and efficient mining in deep mines.