Ulfa ., , Agung Ramadhanu

Implementasi Pengolahan Citra untuk Klasifikasi Jenis Bunga Matahari, Mawar, dan Tulip Menggunakan Algoritma K-Means Clustering

Introduction

Implementasi pengolahan citra untuk klasifikasi jenis bunga matahari, mawar, dan tulip menggunakan algoritma k-means clustering. Otomatisasi klasifikasi bunga matahari, mawar, dan tulip menggunakan pengolahan citra & K-Means Clustering. Tingkatkan akurasi identifikasi & efisiensi industri hortikultura.

Abstract

Manual identification and classification of ornamental flower varieties is time-consuming and highly dependent onindividual expertise, resulting in identification errors that impact the production value chain and operational efficiency ofthe horticulture industry. This research aims to implement an automated classification system for three types ofornamental flowers (sunflower, rose, and tulip) using K-Means Clustering method with visual feature analysis to improveidentification accuracy and computational efficiency. The research methodology includes acquisition of 210 high-qualitybalanced flower images (70 samples per class), preprocessing with RGB to HSV color space transformation, segmentationusing K-Means with k=3, and extraction of 10 multi-dimensional features encompassing morphology, color, and GLCMtexture. The dataset was divided into 80% training and 20% testing using stratified sampling with K-Fold CrossValidation. Performance evaluation was conducted using accuracy, precision, recall, and F1-score metrics. The researchresults demonstrate overall accuracy of 88.89% with sunflower achieving F1-score of 0.98 (0% error), rose 0.86 (14.3%error), and tulip 0.85 (19% error). Aspect ratio, solidity, and mean red channel intensity proved to be the mostdiscriminative features. Misclassification predominantly occurred in the rose-tulip pair (71.4%) due to red spectrum coloroverlap and morphological variation. K-Means algorithm demonstrated optimal balance between accuracy,computational efficiency (0.3s/image), and interpretability, although it has limitations on low feature separability. Thisstudy is limited to a small dataset (210 images) and controlled conditions, requiring real-world validation for bettergeneralization

Review

This paper presents a commendable effort to address the practical challenge of manual flower identification by proposing an automated classification system for sunflowers, roses, and tulips. The authors clearly articulate the problem of time-consuming and expertise-dependent manual methods, which lead to errors impacting the horticulture industry. The research aims to improve identification accuracy and computational efficiency through the implementation of K-Means Clustering combined with visual feature analysis. This objective is well-defined and holds significant relevance for industrial applications, offering a tangible solution to a real-world problem. The abstract provides a well-structured overview of the study's purpose and its potential contribution to the field of automated horticultural classification. The methodology outlined demonstrates a solid understanding of image processing and machine learning principles. The acquisition of a balanced dataset of 210 high-quality images, followed by systematic preprocessing steps like RGB to HSV transformation and K-Means segmentation (k=3), indicates a robust approach. The extraction of 10 multi-dimensional features covering morphology, color, and GLCM texture is particularly strong, as it leverages a comprehensive set of visual descriptors. The use of stratified sampling with K-Fold Cross-Validation for dataset splitting and the evaluation with standard metrics (accuracy, precision, recall, F1-score) further reinforces the methodological rigor. The results, an overall accuracy of 88.89%, with sunflowers showing excellent performance (F1-score 0.98) and identified discriminative features (aspect ratio, solidity, mean red channel intensity), are promising. The detailed analysis of misclassification, particularly the rose-tulip pair due to color overlap and morphological variation, adds valuable insight into the model's performance and challenges. While the study presents a compelling and efficient classification system (0.3s/image), it acknowledges its primary limitations: a relatively small dataset of 210 images and controlled conditions. This is an important caveat, as the generalization of these findings to real-world scenarios with varying backgrounds, lighting conditions, and a broader range of flower variations would require extensive validation. Future work should prioritize expanding the dataset significantly and incorporating images captured under more diverse, uncontrolled environments to assess the model's robustness. Exploring advanced deep learning techniques or hybrid models that combine clustering with supervised classification could also be beneficial, especially for mitigating misclassifications in closely related classes like roses and tulips. Despite these limitations, the research offers a strong foundational contribution to automated flower classification and provides clear directions for future advancements.

Full Text

Comments

You need to be logged in to post a comment.