Deep Learning in Maize Disease Classification

As a strategic global crop, maize productivity is directly threatened by leaf diseases such as Southern Leaf Blight andGray Leaf Spot, making early and accurate detection crucial for food security. Artificial intelligence, particularly deep learning, provides apowerful solution for the automated classification of plant diseases from images. This study developed an intelligent system to address thischallenge, utilizing the publicly available PlantVillage dataset to evaluate five leading Convolutional Neural Network (CNN) architectures:DenseNet121, InceptionV3, MobileNetV2, ResNet-50, and VGG16. The models were optimized with established techniques, includingtransfer learning, data augmentation, and hyper-parameter tuning, while a Soft Voting Ensemble strategy was used to enhance combinedperformance. Evaluation across multiple metrics showed that InceptionV3 achieved the highest test accuracy at 94.47%. However,MobileNetV2 demonstrated the strongest performance across all metrics with a 95% cumulative accuracy and proved highly efficient,making it ideal for deployment on mobile devices. These findings confirm the significant potential of deep learning for building cost-effectiveand efficient diagnostic systems in agriculture, ultimately contributing to the reduction of crop losses and the promotion of sustainablefarming practices.