Early identification of strawberry leaves disease utilizing hyperspectral imaging combing with spectral features, multiple vegetation indices and textural features

Abstract

Gray mold is a devastating disease during the growth of strawberries, which markedly affects strawberry yield and quality. Accurate, rapid, and nondestructive recognition in the early phase of the disease is important for strawberry production management. This study focused on the potential of using hyperspectral imaging (HSI) combined with spectral features, vegetation indices (VIs), and textural features (TFs) for the detection of gray mold on strawberry leaves. First, hyperspectral images of healthy and 24-h infected leaves were collected using a HSI system. Subsequently, the preprocessed hyperspectral images were utilized to extract the spectral features and VIs. TFs were acquired from the images using a grey-level co-occurrence matrix (GLCM). Third, competitive adaptive reweighted sampling (CARS) was performed to select the optimum wavelengths (OWs), ReliefF was employed to select significant VIs, and correlation-based feature selection was used to select the effective TFs. Finally, three machine learning models (extreme learning machine (ELM), support vector machine (SVM), and K-nearest Neighbor (KNN)) of strawberry gray mold were developed based on OWs, significant VIs, effective TFs, and fusion features. The results demonstrated that the models based on OWs and significant VIs performed well, with their highest classification accuracy reaching 93.33%. Although the model based on selected TFs performed slightly worse, the results presented on disease detection by TFs are encouraging for further studies. The performance of the models with combined features was better than those based on single features, with an accuracy range of 93.33–96.67%. Overall, the combined feature-based method significantly improved the recognition accuracy of strawberry gray mold and could accurately identify infected leaves in the early stages.