The Raman spectroscopy analysis technique has found extensive applications across various disciplines due to its exceptional convenience and efficiency, facilitating the analysis and identification of diverse substances. In recent years, owing to the escalating demand for high-efficiency analytical methods, deep learning models have progressively been introduced into the realm of Raman spectroscopy. However, the application of these models to portable Raman spectrometers has posed a series of challenges due to the computational intensity inherent to deep learning approaches. This research addresses this challenge by introducing a lightweight Raman spectroscopy classification model named RepDwNet. This model not only maintains high accuracy but also exhibits relatively lower computational overhead. Its design integrates advanced techniques, including multi-scale convolutional kernels, depthwise separable convolutions, and residual connections. These modules effectively capture features across varying scales while maximizing the coherence between them.To validate the feasibility and efficacy of the proposed model, a series of experiments were conducted utilizing an augmented Raman spectroscopy dataset from blood samples for species identification tasks. The experimental outcomes demonstrate that the model achieved remarkable accuracy rates of 98.24% and 97.37% on the reflectance and transmittance Raman blood data subsets, respectively. Furthermore, structural reparameterization techniques were employed to compress the model, thereby significantly enhancing inference speed and reducing model parameter size while preserving high classification accuracy. This optimization renders the model more suitable for deployment in portable device scenarios.