dc.description.abstract |
Ethiopia produces diversified types and grades of coffee. In this research work, digital image
analysis techniques were used to identify, classify and count defects of coffee beans into five
classes (pressed, under dried, floater, elephant and faded coffee beans) based on their
morphological, color and textural features extracted from 50 images. For each defect, 10
images were captured, thus in total 50 images were used to extract morphology, color and
texture features. Artificial Neural Network was employed to classify defects of coffee beans to
their respective category by using the extracted features. In total 31 features (12 colors, 7
morphological and 12 textural features), were extracted from 50 images of defects of coffee
beans to be classified into five classes. Setups are designed for classifications of morphology,
color and texture features. From these datasets 60% (30), 20% (10) and 20% (10) were used
in the network for training, testing and validating, respectively. The accuracy of classification
using morphology, color and texture features were 94%, 96%, and 86%, respectively. The
results of classification show that the performance of network in terms of morphological
(shape and size) and color features are relatively better than the result found using texture
feature |
en_US |