Klasifikasi Citra Daun Anggur Menggunakan SVM Kernel Linear

Abstract

The use of artificial intelligence for the image recognition process has been carried out by many researchers. One of its fields is to recognize diseases of grape leaves. Modeling has been carried out using augmentation before support vector machine classification with kernel cubic, with 97.6% accuracy. Improved performance of image prediction accuracy through modeling can still be improved through various means. Some techniques that can be used include using feature selection, initial processing to find and discard outliers, or selecting classifier algorithms that are specifically able to handle datasets with certain characteristics. Another technique is to pass images in the feature extraction process to obtain models with relatively higher accuracy than previous studies. This study aims to improve the acquisition of accuracy figures using the help of the feature extraction process, as well as comparing the performance of several classifiers, namely k-Nearest Neighbor, Random Forest, Naïve Bayes, Neural Network, and Support Vector Machine. The method used starts from the feature extraction process utilizing the SqueezNet algorithm to obtain a dataset with a specific composition. Furthermore, the division of training and test data was carried out with a ratio of 60:40. Data training uses a variety of validated classifiers using 2-fold cross-validation. The data used is a secondary dataset of grape leaves, consisting of 7222 leaf images, divided into four validated classes from related studies. The results obtained outperformed the previous study, namely 98.1% on the Support Vector Machine classifier using linear kernels.