The invention discloses a nonlinear laser fluorescence spectrum real-time identification method, which comprises the following steps: learning a sample spectrum, testing sample spectral classification, extracting ROI in an interested region, preprocessing the spectrum, extracting the fluorescence spectrum characteristics by discrete curvelet transform, forming feature vectors, constructing i classes of support vector machines, and distinguishing the test results by classes. The invention adopts the classification method of the support vector machines, and does not depend on large sample training, the input vector is the low-frequency coefficient part after curvelet decomposition, the number of training samples is small, the number of the support vectors is greatly reduced, so the operation time is shortened and the method has instantaneity. The second-generation curvelet transform adopted by the invention is based on a new support frame, and can provide high-efficient, stable and nearly-optimal sparse representation for the curve function with strangeness. Compared with the traditional method, the method is more effective and has higher identification rate. The invention can identify the spectrum samples with data format and image format, and has better adaptability.