Application of Principal Component Analysis (PCA) for Dimensionality Reduction in Watermelon Leaf Classification

Machine learning, along with digital image processing is one of the most efficient ways to identify and classify plants. However, this can be expensive and lead to overfitting if a lot of features are used for classification. The research process begins with image preprocessing, including cropping, resizing, separating foreground and background images, and noise removal. Next, image segmentation is performed using Canny edge detection to separate leaf objects from the background. Feature extraction is conducted based on two categories: shape and texture features. Shape features include basic geometric features such as diameter, physiological length, physiological width, leaf area, and leaf perimeter, as well as digital morphology features like aspect ratio, rectangularity, narrow factor, and the ratio of perimeter to diameter, length, and width. For texture features, extraction is done using the Gray Level Co-occurrence Matrix (GLCM) method. After extracting all features, PCA is applied to reduce the dimensionality from 14 original features to 5 principal components while retaining 85% of the data variance. The classification process is then performed using SVM with a quadratic kernel. The results show that the SVM classification method, after dimensionality reduction with PCA, achieved an accuracy of 83.68%, which is higher than the accuracy without PCA, which was only 74.54%. Thus, the use of PCA has proven effective in improving classification accuracy and speeding up computation by reducing the number of features without losing significant information.