Investigation into maize seed disease identification based on deep learning and multi-source spectral information fusion techniques

Detection of diseases in maize seeds is crucial for their quality evaluation and disease control. This study uses hyperspectral imaging (HSI) and deep learning methods for analysis and identification. Successive projections algorithm (SPA) and principal component analysis (PCA) were applied to extract feature variables, and data-level fusion, feature-level fusion, and decision-level fusion were employed to process different types of feature data. Classification models with different fusion strategies were built using partial least squares discriminant analysis (PLS-DA), random forest (RF), support vector machine (SVM), and convolutional neural network (CNN-RB). The results show that the modeling performance based on spectral features outperforms that based on color and texture features. Among them, the accuracy of CNN-RB based on feature variable modeling was 94.44 %, which was better than RF (93.89 %) and SVM (92.78 %), and only second to PLS-DA (97.78 %). Different fusion strategies had different performances, among which the decision-level fusion had the best effect, with an accuracy of 98.12 %. The study shows that the proposed CNN-RB model can effectively enhance the feature extraction ability of the network, and the multi-source information fusion technique can improve the recognition performance of the model. The method has great potential for application in seed disease detection.