When Mobilenetv2 Meets Transformer: A Balanced Sheep Face Recognition Model

Sheep face recognition models deployed on edge devices require a good trade-off between model size and accuracy, but the existing cannot do so. To solve above problems, this paper combines Mobilenetv2 with Vision Transformer to propose balanced sheep called MobileViTFace. MobileViTFace enhances model’s ability extract fine-grained features suppress interference of background information through distinguish different faces more effectively. Thus, it can The accuracy 96.94% is obtained self-built dataset containing 5490 photos 105 sheep, which 9.79% improvement compared MobilenetV2, only small increase in Params (the number parameters) FLOPs (floating-point operations). Compared such as Swin-small, currently performs SOTA, are reduced by nearly ten times, whereas 0.64% lower. Deploying Jetson Nano-based computing platform, real-time accurate results obtained, has implications for practical production.