Klasifikasi Spesies Jamur Menggunakan Convolutional Neural Network Dan Transfer Learning

Bramantika, Nagara Gusti (2026) Klasifikasi Spesies Jamur Menggunakan Convolutional Neural Network Dan Transfer Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221083-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Identifikasi spesies jamur secara manual sering kali sulit dilakukan karena disebabkan oleh kemiripan karakteristik morfologi antar spesies, terutama antara jamur konsumsi dan jamur beracun. Kesalahan identifikasi dapat menimbulkan risiko kesehatan yang serius sehingga diperlukan metode klasifikasi citra otomatis. Penelitian ini bertujuan untuk membandingkan kinerja model Convolutional Neural Network (CNN) dan transfer learning arsitektur ResNet-50 V2 dalam melakukan tugas klasifikasi enam spesies jamur. Adapun objek penelitian terdiri dari tiga spesies konsumsi (Pleurotus ostreatus, Flammulina velutipes, dan Coprinellus micaceus) serta tiga spesies beracun (chlorophyllum molybdites, amanita phalloides, dan galerina marginata). Evaluasi model dilakukan dengan menggunakan metrik akurasi, presisi, recall, F1-Score, weighted F1-Score, Area Under Curve (AUC), dan Receiver Operating Characteristic (ROC). Untuk mengatasi masalah ketidakseimbangan distribusi data, diterapkan teknik augmentasi dan class weight. Selanjutnya, dilakukan pengujian ketahanan model terhadap gangguan rotasi, perubahan tingkat kecerahan, dan gaussian noise. Lalu, dilakukan visualisasi integrated gradients untuk menginterpretasikan fokus perhatian model dalam melakukan klasifikasi. Berdasarkan berbagai metrik evaluasi, pengujian ketahanan terhadap gangguan citra, dan interpretabilitas model, ResNet-50 V2 dengan data augmentasi ditetapkan sebagai model dengan performa terbaik dalam mengklasifikasikan keenam spesies jamur. Hasil penelitian menunjukkan bahwa model ResNet-50 V2 dengan data augmentasi menunjukkan performa terbaik dengan nilai akurasi sebesar 88%, weighted F-1 Score sebesar 88%, serta nilai macro dan micro AUC masing-masing sebesar 0,98. Model ini juga menunjukkan ketahanan yang baik terhadap gangguan rotasi dan perubahan tingkat kecerahan dibandingkan dengan model lainnya. Sementara itu, hasil visualisasi integrated gradients menunjukkan bahwa model ResNet-50 V2 memiliki pola interpretasi yang berbeda dengan CNN, dimana ResNet-50 V2 menghasilkan area atribusi yang lebih terpusat pada bagian citra yang paling berkontribusi terhadap proses klasifikasi.
====================================================================================================================================
Manual identification of mushroom species is often challenging due to the similarity of morphological characteristics among species, particularly between edible and poisonous mushrooms. Misidentification may lead to serious health risks; therefore, an automated image classification method is required. This study aims to comparing the performance of a Convolutional Neural Network (CNN) model and transfer learning using the ResNet-50 V2 architecture for classifying six mushroom species. The research objects consist of three edible species (Pleurotus ostreatus, Flammulina velutipes, and Coprinellus micaceus) and three poisonous species (Chlorophyllum molybdites, Amanita phalloides, and Galerina marginata). Model performance was evaluated using accuracy, precision, recall, F1-Score, weighted F1-Score, Area Under the Curve (AUC), and Receiver Operating Characteristic (ROC) metrics. To address the problem of class imbalance, data augmentation and class weight techniques were applied. Furthermore, robustness testing was conducted under image rotation, brightness variation, and gaussian noise disturbances. Integrated Gradients visualization was also performed to interpret the model's focus of attention during the classification process. Based on various evaluation metrics, robustness testing, and model interpretability analysis, the ResNet-50 V2 model with data augmentation was identified as the best-performing model for classifying the six mushroom species. The results indicate that the ResNet-50 V2 model with data augmentation achieved the best performance, with an accuracy of 88%, a weighted F1-Score of 88%, and macro and micro AUC values of 0,98. This model also demonstrated better robustness against rotation and brightness disturbances compared to the other models. Meanwhile, the integrated gradients visualization revealed that the ResNet-50 V2 model exhibited a different attribution pattern from the CNN model, with attribution regions more concetrated on the image areas that contributed most to the classification process.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Convolutional Neural Network, Klasifikasi Citra, ResNet-50 V2, Spesies Jamur, Transfer Learning, Convolutional Neural Network, Image Classification, Mushroom Species, ResNet-50 V2, Transfer Learning
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Nagara Gusti Bramantika
Date Deposited:	04 Aug 2026 08:44
Last Modified:	04 Aug 2026 08:44
URI:	http://repository.its.ac.id/id/eprint/143548

Actions (login required)

View Item