Fig. 1. Schematic of softsign weighting function

Fig. 2. Flow chart of LSRPLS baseline correction algorithm

Fig. 3. Simulated Raman spectra with different types of baselines (signal to noise ratio of 15). (a) Linear baseline; (b) sinusoidal baseline; (c) Gaussian baseline; (d) exponential baseline

Fig. 4. Spectrum correction results with combination baseline. (a) Root-mean-square error for various λ when signal to noise ratio is 15; (b) Raman spectra before and after baseline correction when signal to noise ratio is 15; (c) Raman spectra before and after baseline correction when signal to noise ratio is 20; (d) Raman spectra before and after baseline correction when signal to noise ratio is 30

Fig. 5. Experimental results of simulated Raman spectra. (a) Estimated error curves for linear baseline; (b) estimated error curves for sinusoidal baseline; (c) estimated error curves for Gaussian baseline; (d) estimated error curves for exponential baseline; (e) vertically enlarged error curves for Gaussian baseline; (f) vertically enlarged error curves for exponential baseline

Fig. 6. SERS spectra estimated baselines and correction results of R6G. (a) Original R6G Raman spectrum and estimated baselines with different correction algorithms; (b) baseline correction result using LSRPLS algorithm

Fig. 7. SERS spectra of two samples before and after baseline correction. (a) Original spectra of 4-MBA; (b) corrected spectra of 4-MBA; (c) original spectra of 4-MPY; (d) corrected spectra of 4-MPY

Fig. 8. Plots of PCA scores for Raman spectra of two samples. (a) PCA score of original Raman spectra; (b) PCA score of baseline corrected spectra

Table 1. Comparison of peak heights before and after simulated spectrum baseline correction

Table 2. Mean root-mean-square error using different baseline correction methods

Table 3. Classification results for 4-MBA and 4-MPY by PCA-LDA