Segmentasi Citra Daun Tomat untuk Klasifikasi Penyakit Tanaman Menggunakan Support Vector Machine (SVM)

Abstract

Tomatoes are one of the most widely grown crops worldwide. In Indonesia, particularly in West Sumatra, tomato production has declined. This is due to extreme weather conditions and plant disease outbreaks. One solution to help with early identification of tomato plant diseases is through digital image-based classification. This process involves several important stages, starting from image acquisition, preprocessing, segmentation, feature extraction, and classification. However, the quality of classification is highly dependent on the effectiveness of segmentation in separating leaf objects from the background. This study proposes a method for segmenting tomato leaf images based on a combination of color thresholding techniques, morphological operations, contour filtering, and bitwise masking to ensure that only the leaf parts are processed further. After undergoing the segmentation process, images are extracted based on color characteristics in HSV space and GLCM texture, then further processed using an SVM algorithm with an RBF kernel. The dataset used consists of 4000 tomato leaf images with an 80% training and 20% testing data division scheme, accompanied by 5-fold cross validation. The model achieved an accuracy of 96.97% on the training data and 93.75% on the testing data. The results show that segmentation methods using color thresholding, morphology, contours, and bitwise masking can help improve the consistency of extracted features, thereby potentially supporting more stable classification performance.