Abstract:
Agriculture is a vital economic sector for any country. Therefore, improving the agriculture
sector properly is much needed; especially computer technology has been widely
used for improving the productivity of the cultivation. Potato is one of the mostly
consumed vegetables, but it is susceptible to a variety of diseases. The commonly used
way to identify them is by looking at the leaves of the plant. Therefore, if artificial
intelligence can be utilized to recognize them, it can be used to provide immediate
responses. Accordingly, the main aim of this study is to identify a machine learning
algorithm that is fast, and more accurate to identify diseases that affect the potato leaf
such as late blight and early blight, using multiclass image classification. Image data
for the study was acquired from the Potato disease library of the Kaggle repository. It
contains three subclasses named early blight, healthy, and late blight which contains
2152 images in the above three classes. VGG16, efficientNetB0, and Support Vector
Machine (SVM), Extreme Gradient Boost (XGBoost) were chosen as deep learningbased
algorithms and machine learning algorithms, respectively. Initially, models were
trained and a test split was done with Keras preprocessing library. Then a comparison
was done among the classifiers, considering the accuracy, validation, and loss, and
then the best one was selected for the potato disease identification. The factors and
parameters affected to increase the power of the models were then considered. The test
accuracies achieved by the classifiers were 92%, 95% for CNN models, 83% for SVM,
and 86% for XGBoost as average model accuracy approximately in testing phase. These
findings would lead to the development of a model that is best suited for detecting
potato leaf diseases. Future researchers will be able to program a flying drone using the
aforementioned model and computer vision to identify plant diseases immediately.
