Abstract: Image recognition-based coal-gangue sorting robots have become a research hotspot in the field of coal-gangue separation. To address the issue of low recognition accuracy caused by complex real-world conditions—such as dust adhesion, lighting variation, water stains, and coal slurry coverage—a dynamic weighing method for coal and gangue was proposed, integrating a tension sensor and an acceleration sensor to enable secondary recognition. By analyzing the influence mechanism of triaxial acceleration on the tension sensor during the high-speed motion of the robotic arm in a coal-gangue sorting robot, a dynamic weighing model for coal and gangue based on triaxial acceleration compensation was established. Furthermore, an outlier elimination mechanism based on the interquartile range (IQR) algorithm was introduced to suppress random noise in the dynamic weighing model. An experimental platform for dynamic weighing of coal and gangue in a coal-gangue sorting robot was constructed to conduct experiments. Experimental results showed that the weighing error reached 66.43% without triaxial acceleration compensation. After introducing z-axis, and x-and y-axis acceleration compensation, the errors were reduced to 12.97% and 8.69%, respectively. With the addition of the IQR algorithm, the weighing error of the dynamic weighing model was further reduced to 4.69%, representing a 61.74% reduction compared to the case without triaxial acceleration compensation and the IQR algorithm. The model was able to achieve secondary recognition between coal and gangue when their density difference exceeded 0.35 g/cm³, effectively solving the problem of low recognition accuracy under complex real-world conditions.