An effective adaptive deep learning method combined with a hyperspectral system to identify the soybeans quality from different regions

Abstract

An effective method that influences classification performance is the extraction of spectral information features. In this study, Global Spatial Spectral Attention Network (GSSA-Net) is designed and integrated with a hyperspectral system to identify soybeans quality across various regions. Initially, a hyperspectral system is utilized to capture spectral information from soybeans in various regions. Subsequently, Global Spatial Spectral Attention (GSSA) is proposed to focus on deep spectral features by calculating attention in multiple subspaces. Ultimately, GSSA-Net is designed to achieve efficient classification of spectral information from various regions. When compared to other state-of-the-art methods for spectral information classification, GSSA-Net demonstrates superior performance with an accuracy of 98.41 %, precision of 98.50 %, and recall of 98.60 %. In summary, the integration of GSSA-Net and the hyperspectral system offers an effective technical approach for identifying soybean quality across different regions.