A multi-source feature stable learning method for rapid identification of cork spot disorder in ‘Akizuki’ pear

Abstract

The quality grading and sorting of ‘Akizuki’ pear, a high-quality fruit, is a vital part of the supply chain. Meanwhile, cork spot disorder, a common physiological issue that affects the healthy development of pear fruit, requires rapid and accurate identification. To further enhance the identification precision of cork spot disorder in ‘Akizuki’ pear, this study proposes a multi-source feature stable learning (MFSL) method based on neural architecture search and sample reweighting techniques, building upon the fusion of near-infrared spectrum and visual image. This method employs secondary reweighted optimization training on a multi-source fusion model, enabling it to fully learn label-related features and thereby enhance generalization. Experimental results show that the optimal modelling performance of the multi-source fusion feature has increased by 26.89 % in accuracy compared to the single spectrum and by 11.19 % compared to the single image. After optimization training, the testing accuracy of the model improved by 1.31 %, reaching 89.47 %, and the F1-score increased by 1.47 %, reaching 89.83 %. The results validate the effectiveness of the method in enhancing the model’s generalization performance. The proposed MFSL method for the precise identification of cork spot disorder in ‘Akizuki’ pear in this study significantly improves the accuracy of the multi-source fusion model in symptom recognition. The research results have important reference value for the efficient grading and sorting of pear fruit quality.