Microplastic pollution has become a global environmental problem. Strengthening the supervision of microplastic pollution in urban waters is the key to solving the problem. Therefore, this paper carried out research on a rapid and real-time microplastic identification method in urban rivers. In this paper, a genetic algorithm optimized generalized regression neural network (GA-GRNN) algorithm combined with micro Raman spectroscopy was proposed. The experimental detection and theoretical calculation of microplastic powder were carried out, and the fitting interpretation of Raman spectrum characteristic peaks and hidden peaks of microplastic powder were analyzed. The Raman spectra of microplastic suspensions with different concentrations were evaluated. The GA-GRNN algorithm established the microplastic recognition and classification model. The model's classification accuracy was 100%, which realized the accurate recognition of microplastic particles separated from the river. This paper presents a very practical technical method for combining the GA-GRNN algorithm with micro Raman spectroscopy, which has a good reference significance in the supervision and guidance of micro plastic pollution in urban waters in the future.