Publication: Corn leaf disease detection system based on deep learning
Loading...
Date
2022-07-01
Authors
Fu, Cheau Pyng
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Corn, one of the essential commodities in Malaysia, was given increased attention as one of the new sources of national wealth. However, because corn plants are susceptible to plant diseases, current corn production is insufficient to meet consumption demand. Implementing earlier detection of corn leaf disease is critical for increasing the quantity and quality of corn yield. In the past, the traditional methods for early detection of plant diseases were time-consuming and inefficient since the resources of knowledge and plant experts were limited. Over the last few years, deep learning approaches have demonstrated notable performance in plant disease identification using digital images. Hence, an approach to corn leaf disease detection with various convolutional neural networks (CNNs) is proposed in this research. The proposed framework is mainly focused on transfer learning with three different CNN-based pre-trained models such as ResNet50, DenseNet121, and EfficientNetB0. In this project, the models are trained, validated, and tested with a total of 3849 corn leaf images from PlantVillage to classify the four different categories of corn leaf images, which are gray leaf spot, common rust, northern leaf blight, and healthy leaves. The hyperparameters are varied and optimized to enhance the accuracy of each model. The results of these models in detecting corn leaf disease are compared using multiple standard evaluation metrics. Based on the results, it was discovered that the best recognition accuracy achieved by ResNet50, DenseNet121, and EfficientNetB0 models was 98.97%, 98.45%, and 93.56%, respectively. The user interface is developed by deploying the ResNet50 model into the web application, which allowed farmers to upload a snapshot of infected corn leaves. The disease category along with the confidence percentage is then displayed on the web page. In short, it is expected that the system would give farmers a better opportunity to keep their crops healthy and minimize the number of times they use the incorrect fertilizers, which could hurt the plants.