Automatic Coffee Disease and Pest Damage Identification
No Thumbnail Available
Date
2020-03-03
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Coffee is a major commodity consumed as a beverage by one third of the world’s population, and Ethiopia covers 7-10 % of its production. Around 15 million Ethiopian’s income directly or indirectly depend on coffee production. Coffee plays a major role in the country’s economic growth and stability, yet diseases (like coffee berry disease, coffee leaf rust and brown eye spot) and pests (like coffee leaf miner and green scales) affect its production in quality and yield. An early identification of these diseases and pests is very essential to control the damage caused by diseases and pest infestation. Also, developing an automatic disease and pest identification system that extracts important features from different parts of coffee plant is very crucial and yet lacking. In this research work, we proposed an automatic identification of coffee diseases and pest damages from leaf and berry parts of the plant using image processing and machine learning techniques.
We proposed a segmentation algorithm that separates healthy regions of coffee leaf/berry from damaged ones using L*a*b* color space, YCbCr color space and texture filter. We identified a total of 28 features (i.e. 22 color and 6 texture) to model the classes of coffee diseases and pest damages. Feedforward artificial neural network with backpropagation learning algorithm was used to classify coffee disease and pest types. The network is designed with 28 input and 5 output nodes.
The model is trained using 2,400 sample images of coffee leaves and berries collected from Jimma Agricultural Research Center, Tepi Agricultural Research Center, Mechara Agricultural Research Center, and Sheko and Limmu coffee farms. The training data is randomly split into 80% training and 20% testing. The classifier result was compared with results obtained from SVM and KNN classifiers. The trained classifier model reached an overall classification accuracy of 91.9%. Classification accuracy for coffee berry disease, coffee leaf rust, coffee leaf miner, brown eye spot disease and green scales is 89.6%, 91.5%, 93.3%, 96.6% and 89.0%, respectively.
Description
Keywords
Digital Image Processing, L*A*B* Color Space, Ycbcr Color Space, Local Range Filter, Artificial Neural Network, Coffee Disease