Early detection of tomato leaf spot and wilt diseases based on hyperspectral imaging technology

This study explores the application of hyperspectral imaging technology integrated with machine learning for the early detection of tomato leaf spot and blight diseases. Pathogens causing leaf spot and blight were introduced to young tomato seedlings through foliar sprays and root irrigation. A hyperspectral imaging system was established to capture detailed spectral images of tomato leaves during the initial stages of disease development. The collected hyperspectral data were preprocessed using smoothing algorithms (Savitzky-Golay, SG), multivariate scatter correction (MSC), standard normal variate transformation (SNV), and first derivative (1^st Der). A support vector machine (SVM) detection model was trained to evaluate the effectiveness of these preprocessing methods. Through comprehensive modeling and comparative analysis, the 1^st Der-SG preprocessing approach was identified as the most effective, achieving an overall detection accuracy of 79.3% on the test dataset. Furthermore, feature extraction was performed on the preprocessed data using competitive adaptive weighted sampling (CARS), successive projection algorithm (SPA), uninformative variable elimination (UVE), and principal component analysis (PCA). Subsequently, the dung beetle optimization (DBO) algorithm was employed to enhance the performance of support vector machines (SVM) and bi-directional long short-term memory networks (BiLSTM), resulting in the development of DBO-SVM and DBO-BiLSTM models. The most effective model combination, 1^st Der-SG-UVE-DBO-BiLSTM, achieved an outstanding overall accuracy of 97.3% on the test set. This research highlights the significant potential of hyperspectral imaging technology for the early detection of tomato leaf spot and blight diseases. The findings provide valuable technical insights for tomato disease detection and establish a theoretical foundation for early disease identification in other crops.