An Image-Based Cocoa Diseases Classification Based on an Improved Vgg19 Model

Purpose: The focus of this study is to provide accurate detection of cocoa diseases based on image analysis using a deep learning model. Design/Methodology/Approach: Transfer learning based on a convolutional neural network such as VGG19 provides significant accurate results in image classification. This paper proposes an image-based cocoa diseases classification based on an improved VGG19 model. A comparison is made with other pre-trained models such as VGG16 and ResNet50. Findings: The results indicate that VGG19 outperforms the other pre-trained models. Research Limitations/ Implications: The income obtained from cocoa production is one of the bedrock of the economies in some west African countries. Cocoa production has been increasing steadily globally in recent years; however, there are high disease pathogens in most areas of production. It is estimated that the black pod disease causes 30% to 90% losses in annual cocoa production. Consequently, the global losses of cocoa production due to diseases are estimated at 20% to 25%, which is about 700,000 metric tons of global production. Researchers have proposed varying methods for the classification of cocoa diseases; however, the identification and classification of cocoa diseases still remain a challenge. Deep learning has been very promising in its application in various fields. Practical Implications: Accurate prediction of cocoa disease will provide stakeholders, especially farmers to provide appropriate remedies and improve productivity. Originality/Value: The model is very effective and performs better than the state-of-the-art techniques employed on the public dataset of cocoa diseases.