Abstract:
Maize is an important crop, but it is vulnerable to plant diseases such as maize turcicum leaf
blight and common rust. Traditional methods to identify and classify maize leaf diseases have
drawbacks. Therefore, this study focused on developing a maize leaf disease detection and
classification algorithm using an artificial neural network technique. The maize leaf image
samples were taken from Haramaya University's glass house "Raree" Research Station using
random sampling techniques over three rounds. A total of 396 images were acquired using a
digital camera with high resolution and imported into a computer. The RGB images were
converted into CIELAB colour space, and the detection of turcicum leaf blight and common
rust was performed using k-means clustering to estimate the disease of the leaves. Then,
Otsu’s thresholding was used to segment the maize diseased leaf images, and morphological
features were extracted from each image. For the recognition and classification analysis, 4
morphological and 6 color features, totaling 10 features, were extracted from each image.To
evaluate the recognition and classification accuracy, from the total data set of 396 images,
70% were used for training, 15% were used for validation, and 15% were used for testing the
model. Based on morphological and color features, an artificial neural network (ANN) was
used to classify the maize leaf diseases.The performance measures of the classifier, such as
sensitivity, specificity, and accuracy, were computed. It was observed that the proposed
scheme with an ANN classifier outperformed the competition, giving 94.19% accuracy, 96.6%
sensitivity, and 92.2% specificity for morphological features, whereas the performance of the
classifier for color features were 93.5% accuracy, 92.6% sensitivity, and
94.2% specificity.Based on the experimental results using morphology and color features, the
morphological feature performed better than the color feature. All the techniques were
implemented through the MATLAB R2018a platform, and the design and implementation of
these image processing techniques help with the easy detection of plant diseases.
