Identification of tea quality at different picking periods: A hyperspectral system coupled with a multibranch kernel attention network

The material content and nutritional composition of tea vary during different picking periods, leading to variations in tea quality. The absence of rapid evaluation methods for identifying tea quality at different picking periods hinders the smooth operation and maintenance of agricultural production and market sales. In this work, hyperspectral technology combined with the multibranch kernel attention network (MBKA-Net) is proposed to identify the overall quality of tea during different picking periods. First, spectral information of six different tea picking periods is obtained using a hyperspectral system. Second, the multibranch kernel attention (MBKA) method is proposed, which effectively mines spectral features through multiscale adaptive extraction and achieves classification of tea at different picking periods. Finally, MBKA-Net achieves outstanding performance with 96.18% accuracy, 97.14% precision, and 97.18% recall. In conclusion, MBKA-Net combined with a hyperspectral system provides an effective detection method for identifying the quality of tea at different picking periods.