Combination of Feature Extraction Methods for Identification of Diseases in Corn Leaves

This research develops a classification model that is effective in detecting diseases on corn leaves using the Corn or Maize Leaf Disease dataset from Kaggle. The preprocessing process includes image augmentation, such as horizontal rotation and flipping, increasing data variation and preventing overfitting. The image size is standardized to $150 \times 150$ pixels, and color segmentation with HSV color space is used to separate relevant objects. The feature extraction methods used are Hu Moments, Haralick, and Histogram, which collectively provide a rich and informative feature representation. Of the various classification algorithms tested, Support Vector Machine (SVM) showed the best performance with an accuracy of 98.53% and an error rate of $\mathbf{0. 1 0 \%}$. Logistic Regression (LR) and Linear Discriminant Analysis (LDA) also performed well, but were not as superior as SVM, while Naïve Bayes (NB) had the lowest performance. Evaluation with the confusion matrix shows a very low error rate in the classification of various disease classes. With an overall accuracy of $\mathbf{9 8 \%}$, this model shows a good balance between precision and recall, indicating its ability to consistently identify corn leaf diseases and healthy conditions with high accuracy. This research is expected to increase the productivity and health of corn plants through faster and more accurate disease detection.