As global average fertility rates decline annually, a crisis brought by declining birthrates gradually emerges, further impacting the existing education system. In response to the impact of declining birthrates on education, countries have recently accelerated the development of distance learning. Distance learning provides guaranteed access to comprehensive and quality education while encouraging continuous learning opportunities for everyone. However, the identity recognition technology used in the current distance learning platform seldom addresses the problem of inconsistent image quality in actual scenarios, making it difficult to ensure the practical application of the model. To solve these issues, we propose a lightweight dual-attention convolutional neural network (LDA-FV) constructed through a dual attention-based inverted residual block (DA-IRB) and implemented using an adaptive margin loss (AML) function for finger-vein recognition. This method not only extracts effective finger-vein features through DA-IRB but also adjusts the training difficulty in accordance with the image quality. Furthermore, owing to its lightweight design, the model can be more flexibly deployed on existing hardware devices (e.g., mouse) in distance education scenarios. The experimental results of this study indicate that the proposed method effectively enhanced the model’s recognition capability using the finger-vein database of the University Sains Malaysia (FV-USM) and the PLUSVein dorsal-palmar finger-vein (PLUSVein-FV3) public database, achieving correct identification rates (CIRs) of 99.90% and 97.50%, respectively.