Corn leaf disease detection using deep convolution neural network

Abstract

Detecting corn leaf diseases helps farmers identify and treat impacted crops. Early disease identification reduces crop loss. Manual leaf diagnostic imaging takes time and is prone to mistakes. This thesis proposes a deep convolutional neural network (CNN) model for autonomous corn leaf disease identification. PlantVillage and PlantDoc were utilized. The dataset contains 4,188 photos of healthy maize leaves and three corn leaf illnesses. The photos have disease labels. We rotated, flipped, and scaled images for augmentation. After augmentation, the total number of photos in the dataset is about 12,000. We trained our CNN model using pre-trained ar chitectures like InceptionResNetV2, MobileNetV2, ResNet50, VGG19, InceptionV3, VGG16, and DenseNet201. These architectures were chosen for their image feature extraction and large dataset learning capabilities. We used transfer learning to fine tune a model using a pre-trained model. The model accurately detects corn leaf diseases in new photos. The model is computationally light, making it suited for smartphones and drones. A maize leaf disease detection mobile app was created using the proposed CNN model. The application can detect corn leaves uploaded by anyone. An API analyzes an image using our proposed model from the device’s camera or gallery when a user selects it.