An affective computing framework for preschool music education using MoE architecture and neural radiance fields

The key content of this paper is the digital construction of preschool music education scenarios and affective computing. It adopts an innovative approach, integrating the MoE architecture and neural radiance field technology together. The Edu-NeRF reconstruction framework is relied on to achieve high-fidelity reconstruction of dynamic scenes, and its PSNR value is 28.5. The SSIM value is 0.92, which is 10.8% higher than that of the traditional method. This paper also constructed an affective computing model. After training this model with 60 h of multimodal data from 120 preschool children, its accuracy rate for emotional state recognition reached 89.7%, with a macroscopic F1 value of 0.881. By combining cross-modal fusion algorithms, we have successfully broken through the technical bottlenecks in dynamic scene reconstruction and affective computing. In terms of real-world applicability, this approach significantly outperforms conventional 2D music education products. It enables real-time, high-fidelity interaction on standard low-power devices, thereby reducing hardware deployment costs and lowering the barrier for widespread school adoption. Furthermore, experimental verification demonstrates that this model increases children’s learning engagement and the efficiency of personalized interaction by 30%. By enhancing both the immersive learning experience and teacher acceptance, this framework provides a highly practical and scalable technical path for the digital transformation of early childhood education.