<p>Accurate identification of microalgae species is a critical component in maintaining the health of aquatic ecosystems. In polar waters, phytoplankton, as the foundation of the food web, undergoes compositional changes that have profound impacts on the fragile polar ecosystems, making it a key indicator for monitoring environmental changes in polar regions. However, the spectral data of aquatic microalgae is difficult to directly analyze due to its insufficient amount, lack of representativeness, and high dimensionality. Additionally, the complex structure and low timeliness of conventional processing algorithms make real-time detection of the algae difficult. In order to do this, we suggested a deep learning-based categorization technique for spectral data on aquatic microalgae. First, we recorded the fluorescence spectrum data of microalgae in a one-dimensional linear array using a hyperspectral microscopy imaging method based on multimodal illumination. After that, we used the center value of the gathered one-dimensional light intensity sequence as the location of the final image’s center. To improve the dimensionality of the data, the data on both sides was then arranged outward in a spiral pattern, yielding two-dimensional center-sorted spectral data that is appropriate for deep learning algorithms. Ultimately, using deep learning techniques, we were able to attain a spectral classification accuracy of 93.67% for microalgae fluorescence microscopy cells. The proposed approach is quite beneficial for increasing the efficiency of spectral image analysis. The current findings will significantly boost future studies involving more microorganisms. The proposed method provides robust technical support for efficient and accurate in situ monitoring of polar microalgal community dynamics and assessment of climate change impacts on polar ecosystems.</p>