The invention discloses an ICSO-SVM based benzenoid fluorescence spectrum detection method. The ICSO-SVM based benzenoid fluorescence spectrum detection method comprises the following steps: analyzingthe fluorescence characteristic of mixed solutions of two of three solutions of benzo[k]fluoranthene (BkF), benzo[b]fluoranthene (BbF) and benzo[a]pyrene (BaP), and selecting the spectrum data rangecorresponding to the main fluorescence peak with most possibility, wherein the excitation wavelength is within the range of 300 to 330 nm when the emitting wavelength is 406 nm; and establishing an optimal SVM model again to classify spectrum data, introducing two learning factors on the basis of CSO, not enabling the algorithm to be caught in local optimum as much as possible, and optimizing theSVM algorithm by ICSO to classify 21 prediction samples, wherein the classification accuracy result operated by the ICSO finally is 100 percent, the accuracy is highest in the three algorithms compared with the results of the SVM optimized by CSO, GA and PSO, the operating time, the convergence rate and the fitting degree are optimal, and the aim of identifying similar substances is fulfilled.