Cassava leaf disease classification using deep learning and convolutional neural network ensemble

Abstract

Cassava is a high-protein and nutrient-dense plant, notably inside the leaves. Cassava is often used as a rice alternative. Pests, viruses, bacteria, and fungus may cause a variety of illnesses on cassava leaves. This study consists of four main diseases that commonly affect cassava leaves: Cassava Bacterial Blight (CBB), Cassava Brown Streak Disease (CBSD), Cassava Green Mite (CGM), and Cassava Mosaic Disease (CMD) and we took these four diseases as labels in our research. Furthermore, we took 22000 infected images from Kaggle and we have transformed our dataset into four different image transformation to ensure the accuracy of our model. These four different augmentations are Random Crop Augmentation, Random Flip Augmentation, Random Rotation Augmentation and Random Contrast Augmentation. Finally, we used six algorithms to detect the diseases of cassava leaves. These six algorithms are Xception, EfficientNetB0 Resnet50, VGG16 Densenet121, InceptionV3. While we operated these algorithms on our trained dataset, it gave diverse precision. For the Xception, it gave 91.3% accuracy, EfficientNetB0:91.1%, ResNet50: 85.0 %, VGG16: 68.0 %, DenseNet121: 87.0 % and for the InceptionV3, it gave 86.4 % precision respectively. Here, not every one of the algorithms performed well. Xception and EfficientNetB0 have the most noteworthy accuracy among these.