An Integrated Framework of Two-Stream Deep Learning Models Optimal Information Fusion for Fruits Disease Recognition

Diseases impact the rates of production of many agricultural goods. These diseases require detection, which is difficult to do manually. Therefore, the creation of some automated illness detection systems is urgently required. Deep learning showed significant success in the area of precision agriculture for the recognition of plant disease. Compared with the traditional techniques, the deep learning architecture automatically extracts deep features from the deeper layer. In this work, we proposed a new automated method for classifying apple and grapefruit leaf disease recognition utilizing two-stream deep learning architecture. The proposed framework entails several steps. The first phase is picture contrast enhancement, which combines the information from DnCNN and top–bottom hat filtering to create a better image. Then, the augmentation process uses horizontal and vertical flips to increase the dataset's original size. The Inception-ResNet-V2 deep learning model is then adjusted and trained using deep transfer learning on the expanded dataset. After being extracted from the training model, the best features are chosen using two techniques—an entropy-based strategy and tree growth optimization. Finally, a new effective method combines the chosen features, and machine learning classifiers are used to complete the classification. On the augmented dataset, the proposed framework correctly classified apple and leaf diseases with the accuracy rates of 99.4% and 99.9%, respectively.