Robusta Coffee Leaf Detection based on YOLOv3- MobileNetv2 model
Dann Paulo P. Javierto, John Dannielle Z. Martin, Jocelyn F. Villaverde
Abstract
Coffee is one of the Philippines’ main crops, and Robusta Coffee is the most produced type. However, crop yield faces many challenges because of pervading diseases that cause it to lose both its quality and quantity of production. The current method of identifying these insect infestations, pathogens, and spots resulted in significant losses to farmers because of inappropriate diagnosis and treatment. Therefore, the purpose of this research is to apply deep learning algorithms to alleviate the problem. YOLOv3, a kind of Convolutional Neural Network (CNN), was trained and optimized to accomplish the task of identifying the biotic agents present on robusta coffee leaves to assist farmers in applying the appropriate treatment to their crop. YOLOv3 paired with MobileNetv2 offers a broader range of possibilities as its intermediate expansion layer uses lightweight depth-wise convolutions to filter source features of non-linearity. This model will make image processing work well even with low graphics processing units (GPU). The accuracy of the model was verified using a confusion matrix and it resulted that the system has 90% accuracy in leaf disease detection. This accuracy is achieved through sufficient lighting and uniformity in the background of the training dataset.