Abstract:
Sorghum is a grain crop that is used for human and animal consumption. In areas that are too
hot, sorghum is grown and a minimum average temperature of 25 ° C is required to ensure
maximum grain production. There are many factors in sorghum production and productivity
enhancement, among them crop diseases are the major ones. The early detection of sorghum
diseases is one of the main reasons that can reduce the yield production loss, and this requires a
huge amount of effort, money, and time. To address these problems, the researcher proposed a
deep learning approach for the classification of sorghum diseases based on their leaves. To do
so, the design science research methodology was followed. To conduct this study, a total of 4000
images were collected from shewarobit werda kobo villages, North Shewa zone, and prepared.
After collecting the necessary images, the researcher applies image preprocessing techniques
such as image resizing, normalizing images, and noise removing were performed. And also,
data augmentation techniques were performed. In feature extraction, the researcher applies
Gabor filter on the raw image for texture feature extraction. It is used for detecting and selecting
important features that account for the symptom of the disease. This research work focuses on
classifying three types of sorghum leaf diseases: Anthracnose, leaf blight, and rust. Based on
this, two Convolutional Neural Network frameworks were proposed namely: train the deep
neural network model from the scratch and transfer learning a pre-trained network model.
Finally, the developed classifier model has been through accuracy, precision, recall, and F measure. Experimental result shows that the accuracy obtained from transfer learning model
VGG19 and VGG16 achieves an accuracy of 91.5%, and 87.75% respectively. Conversely, the
proposed model achieves an accuracy of 94.91%, while after applying Gabor filter the proposed
model achieves an accuracy of 96.75%. As a result, training from the scratch model with Gabor
was selected for developing an effective and robust model for classifying sorghum leaf disease.
