1. Foliar symptom-based disease detection in black pepper using convolutional neural network.
- Author
-
Sreethu, P. T., Paul, Manju Mary, Gopinath, Pratheesh P., Shahana, I. L., and Radhika, N. S.
- Abstract
Black pepper is the most important and widely consumed spice in the world. Insects and diseases are the major concerns for black pepper production, among the many variables causing a decline in black pepper productivity. The major diseases that affect black pepper are foot rot (Phytophthora capsica) and anthracnose (Colletotrichum gloeosporioides). Early and precise diagnosis of diseases is crucial as it will enable the farmers to make timely interventions. In the current scenario, the application of image processing and deep learning techniques for the automatic detection of plant diseases emerges as a solution capable of promptly delivering interventions in time-sensitive scenarios, given its capacity to deliver performance approaching expert levels. Through this study, a deep learning-based approach has been developed to classify black pepper diseases based on leaf images. A model has been developed to detect the two major diseases of black pepper, i.e., anthracnose and foot rot diseases, using a Convolutional Neural Network (CNN) in Kerala, India. We have collected 2786 leaf images from different black pepper farms in Kerala, belonging to three classes of pepper diseases and one healthy leaf class in total. The classes of leaf diseases considered include an early and advanced stage of anthracnose, and Phytophthora foot rot. As the accuracy of the model increases with the number of images, different image augmentation techniques are performed on the originally captured images to generate a total of 18,234 images. The developed CNN model has been compared with eight other pre-trained state-of-the-art models, such as VGG16, VGG19, ResNet50, ResNet50V2, MobileNet V2, DenseNet121, InceptionV3, and Xception. The result shows that the developed CNN model attained a higher classification accuracy, precision, recall, and F1-score of 98.72%, 99.28%, 97.65%, and 98.66% respectively, on the unseen test dataset. A web application named "Black pepper Disease Identification App" for demonstrating the proposed model is developed. According to an overall performance assessment, deep learning is an effective technique for classifying black pepper diseases based on leaf images and identifying them in their early stages. Based on the overall performance, the newly developed model is found to be efficient in classifying the selected pepper diseases. The proposed model holds significant promise for enabling the timely identification of diseases with minimal human intervention. Its deployment benefits both researchers and farmers by facilitating prompt disease detection directly in the field. [ABSTRACT FROM AUTHOR] more...
- Published
- 2025
- Full Text
- View/download PDF