Foreign bodies e.g. tea stalks are easily to mix into tea leaves during picking up tea leaves with machine harvester. Tea stems have become a major physical pollutant for finishing tea products, and they cannot be effectively detected with machine vision in the visible region. In the current study, terahertz time domain spectroscopy (THz-TDS) and imaging was employed to rapidly and non-destructively detect tea stalk foreign bodies in tea leaves. With the input variables of THz time-domain signal and frequency-domain absorption coefficients, the K nearest neighbor (KNN) qualitative discriminant models were developed with combination of baseline correction algorithms of adaptive iteratively reweighted penalized least squares (AirPLS), asymmetric least squares (AsLS), background correction (Backcor) and baseline estimation and denoising with sparsity (BEADS). Results showed that the AirPLS-KNN model with the input vector of THz time-domain signals presented the best performance, with the accuracy rate of prediction and recall rate reaching 97.3% and 0.96, respectively. With 0.2 mm step length of both the X and Y directions, the THz transmission imaging was scanned to obtain the THz image of 150X130 pixel2 whose resolution and imaging duration were 1.09 mm and 4.4 min respectively, so the outline of tea stalks could be well identified. In conclusion, the THz-TDS and imaging technology serves as a new means for non-destructive detection of tea stalk foreign bodies in tea leaves.
Read full abstract