Ery Hartati
Faculty of Computer Science and Engineering, Universitas Multi Data, Palembang, South Sumatra, Indonesia

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

Classification of Diseases in Oyster Mushroom Baglogs Using the MobileNetV3-Small Architecture Muhammad Vihir Mujahid; Ery Hartati
Green Intelligent Systems and Applications Volume 6 - Issue 2 - 2026
Publisher : Tecno Scientifica Publishing

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.53623/gisa.v6i2.1447

Abstract

Oyster mushroom baglogs are highly susceptible to contamination by pathogens such as Trichoderma, Mucor, Neurospora, and Penicillium, while disease identification still relies primarily on subjective manual observation by farmers. This study evaluated the effect of data augmentation on the classification performance of the MobileNetV3-Small architecture for detecting diseases in oyster mushroom baglogs. A primary dataset comprising 900 independently collected images was organized into five classes: Healthy, Trichoderma, Mucor, Neurospora, and Penicillium. The dataset was divided at the physical baglog level into 600 training images (66.67%), 150 validation images (16.67%), and 150 testing images (16.67%) to prevent data leakage between subsets. Two experimental scenarios were compared under identical conditions: a baseline model without augmentation and an augmented model in which data augmentation was applied exclusively to the training set. Model training employed transfer learning using ImageNet-pretrained weights, beginning with feature extraction and followed by progressive fine-tuning. Classification performance was evaluated using accuracy, precision, recall, F1-score, and confusion matrices. The results demonstrated that the baseline model achieved a test accuracy of 91.33% and an F1-score of 0.9127, whereas the augmented model achieved a higher test accuracy of 94.67% and an F1-score of 0.9466. All classification errors occurred between the visually similar Penicillium and Trichoderma classes, with data augmentation reducing the number of misclassified images from 13 to 8. Furthermore, both models correctly classified all test baglogs using majority voting across multiple image views. These findings demonstrate that data augmentation improves the classification performance of MobileNetV3-Small, particularly in distinguishing visually similar disease classes. The study provides a foundation for developing lightweight, augmentation-enhanced disease classification systems for oyster mushroom cultivation.