A Convolutional Neural Networks Approach to Detecting Foliar Diseases in Zea Mays

Abstract

The presence of diseases in Zea Mays (Maize) pose a big threat to food security. This could cause big losses in output to farmers, hence subjecting them to economic burdens. Traditional methods for identifying maize diseases depend on skilled visual identification and expertise, which are not very effective and can lead to wrong diagnosis. This study hereby proposes an effective deep learning approach for the automated identification of foliar diseases in maize plants. The study specifically utilized Convolutional Neural Networks (CNN), incorporating two pre-trained models: EfficientNet and VGG16. Varied dataset comprising images of maize leaves afflicted with several foliar diseases were ensembled to train and test the CNN models. EfficientNet and VGG16 were pretrained models that were fine-tuned to work better for the specific goal of finding the diseases in maize. The experimental findings showed that both VGG16 and EfficientNet demonstrated promising performance, while VGG16 remains a robust baseline with 91% accuracy, the structural efficiency of EfficientNet (96% accuracy) provides a more viable solution for deployment in mobile devices for real-time agricultural diagnosis. The higher accuracy of EfficientNet suggests its suitability for this specific agricultural application and indicates the effectiveness of CNN models in accurately identifying foliar diseases in maize.