1：Experimental Design and Method Selection:

The study used near-infrared spectroscopy (NIRS) with a portable spectrometer to collect absorption spectra from wood samples. Three pre-processing methods (MSC, SNV, derivative combined with Savitzky-Golay) were applied to eliminate noise and improve data quality. Classification models (PLS-DA, LS-SVM, BPNN) were developed and compared for defect detection and classification.

2：Sample Selection and Data Sources:

Pinus koraiensis wood boards (400 mm x 200 mm x 20 mm) were prepared with four types of surface defects (live knots, dead knots, cracks, defect-free) based on national standards (GB/T 4823-2013). A total of 360 samples were used, with 240 for calibration and 120 for prediction sets.

3：3). A total of 360 samples were used, with 240 for calibration and 120 for prediction sets. List of Experimental Equipment and Materials:

3. List of Experimental Equipment and Materials: A portable NIR fiber optic spectrometer (Insion Co., Gmbh, Heilbronn, Germany) with wavelength range 900-1900 nm and resolution 9 nm; SPEC view 7.1 software for data collection; MATLAB R2012b for data analysis; wood samples of Pinus koraiensis.

4：1 software for data collection; MATLAB R2012b for data analysis; wood samples of Pinus koraiensis. Experimental Procedures and Operational Workflow:

4. Experimental Procedures and Operational Workflow: NIR spectra were collected using the spectrometer with fiber optic probes at 90° angle to the sample surface under controlled laboratory conditions (20 ± 2 °C, 40-50% relative humidity). Thirty spectra per sample were averaged. Data were pre-processed and used to build and validate classification models.

5：Data Analysis Methods:

Statistical analysis included PLS, PLS-DA, LS-SVM, and BPNN models. Performance was assessed using coefficients of determination (Rc2, Rp2) and root mean square errors (RMSEC, RMSEP).