CLASSIFICATION OF CASSAVA LEAF DISEASES USING DEEP LEARNING MODELS
No Thumbnail Available
Files
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Description
Agricultural production, both qualitative and quantitative, yields economic benefits, which can be realized through crop monitoring, disease detection, and prevention. The cassava
plant is renowned for being a rich source of carbohydrates, yet it is susceptible to a number of diseases that threaten food security in sub-Saharan Africa. Traditional methods of
identifying plant diseases involve manual inspection of the plants which becomes impractical with a vast expanse of farmlands; thereby, necessitating the need for automation. Disease detection through image classification and recognition is known to be the best and most cost-effective method for early detection and prevention of diseases to
prevent further damage to a plant. However, some researchers concentrated on identifying just one form of cassava leaf disease while some classified just two forms of cassava leaf
disease. Also, numerous methods that were proposed in the literature were trained and tested using the Makerere University AI Lab Kaggle dataset or Scholar sphere dataset. In this study however, three datasets were combined – Makerere University AI Lab dataset from Kaggle, Scholar Sphere Penn State University libraries dataset, and Oba-Ile Akure (OBA) cassava leaves dataset. Three convolutional neural network models were trained to classify cassava leaf diseases and healthy plants. The four types of diseases are Cassava Bacterial Blight (CBB), Cassava Brown Streak Disease (CBSD), Cassava Green Mottle (CGM), and Cassava Mosaic Disease (CMD). Pre-trained ResNet50 model was used on
the OBA cassava leaves dataset for cassava leaf detection, which yielded a precision of 97%. Classification of cassava leaf diseases was done using MobileNetV2, VGG16, and
Inception-ResNet-v2 models yielding accuracies of 88.47%, 96.58%, and 96.58% respectively. In addition, VGG16 yielded a precision of 96.58%, a recall of 94.8%, and F1 of 96.6% thereby outperforming other models used. The result obtained in this work presents an excellent way for classifying cassava leaf diseases quickly thereby helping farmers on the field to take actions quickly, consequently improving food security
Keywords
T Technology (General), TK Electrical engineering. Electronics Nuclear engineering