Apple Leaf Disease Identification: A Hybrid Multi-Scale Deep Learning Model

Abstract

Agricultural diseases have an impact on agricultural production and provide the agricultural sector with a number of significant problems. If any prior knowledge about the diseases and environmental conditions is anticipated, these difficulties can be overcome. Artificial Intelligence and computer vision-based methods are being used to this end in order to enhance the food chain. This study thoroughly examines customized models of deep learning for automating the diagnosis of apple leaf diseases. We evaluated SOTA CNN architectures, including our proprietary CNN, ResNet50, EfficientNetB3, and hybrid models, using the Apple Leaf Disease Dataset. Every model demonstrated remarkable performance, with validation accuracies surpassing 96%. Whereas the custom EfficientNetB3 model attained 100% accuracy, the custom ResNet50 model only managed 99%. These results entail that deep learning could be able to identify cedar apple rust, fire blight, apple scab, and powdery mildew on apple leaves. We assessed the models' disease classification recital using measures including recall, accuracy, and F1-score. This Study results reveal that the models performed exceptionally well on all criteria, indicating that they could correctly identify apple leaves that were healthy or ill. To sum up, our research improves automated agricultural disease detection systems and gives farmers powerful tools to lessen the burden of disease in apple orchards. By facilitating early disease identification and proactive control, the incorporation of deep learning algorithms into agricultural methods has the potential to guarantee sustainable production of apples.