Corn Plant Disease Detection at Initial Stage Using Deep Learning Models

Zea mays, the scientific name for corn, is the world's third most important plant-based food source, employed in a variety of industries such as food, biofuel, and industrial applications. Despite its crucial role, corn agriculture faces significant challenges particularly its vulnerability to diseases at different stages of growth. Early detection is critical for sustaining crop health, increasing production, and assuring farmer's financial security. This research compares three advanced convolutional neural network (CNN) architectures-ResN etV2, DenseNet-121, and EfficientNetV2B3 for identifying corn leaf disease. With the growing importance of precision agriculture, timely and accurate disease detection for crop health and productivity. We implemented a detailed image processing pipeline to enhance leaf region segmentation in the images that began with clustering techniques to generate a leaf mask, effectively distinguishing the leaf regions from the background. Further refinement involved applying a median blur to smooth the edges and thresholding to delineate the leaf more clearly. To ensure maximum clarity, contrast enhancement applied on the leaf regions to distinguishable from the background. This meticulous preprocessing step was crucial in preparing the images for the CNN models, ensuring that the features related to diseases were prominently highlighted. We tested ResNetV2, DenseNet-121, and EfficientNetV2B3 on their ability to identify various corn leaf diseases, focusing on accuracy and computational efficiency. DenseNet-121 achieved the highest accuracy of 98.47%, outperforming ResNetV2 with 97.63% and EfficientNetV2B3 with 94.61%, while also maintaining fewer parameters. These findings offer valuable insights for agricultural image analysis and precision agriculture, highlighting the effectiveness of different CNN architectures.