Claim Missing Document
Check
Articles

Found 36 Documents
Search

Classification Of Ulos Fabric Motifs Using MobileNetV3-Small Architecture Sihombing, Mecha Bella Permata; Devella, Siska
INOVTEK Polbeng - Seri Informatika Vol. 11 No. 1 (2026): February
Publisher : P3M Politeknik Negeri Bengkalis

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35314/d2sfn245

Abstract

Ulos fabric is an important cultural heritage of the Batak people in North Sumatra, characterized by diverse motifs and philosophical meanings that support social and ritual life. Public knowledge of Ulos motifs is declining due to lifestyle changes and limited use of digital technology for cultural education, so accurate image-based motif classification is needed for preservation and wider utilization. This study evaluates the performance of the lightweight MobileNetV3-Small architecture with a transfer learning approach for classifying Ulos motif images, positioning it as one of the earliest uses of MobileNetV3-Small for Ulos motif classification compared to previous Ulos studies that relied on heavier CNN or earlier MobileNet variants. The dataset consists of 906 images split into 80% training, 10% validation, and 10% testing, and the model is trained using the Adam optimizer with a batch size of 32 and learning rates of 0.001 and 0.0001. On the test data, the model achieves accuracies of 98.96% and 97.92%, with consistently high precision, recall, and F1-scores, demonstrating the effectiveness of MobileNetV3-Small for Ulos motif classification as a digital educational medium to support Batak cultural heritage preservation.
Performance Analysis of MobileNetV2 and GhostNetV2 in Classifying Cervical Cancer Images in the SIPaKMeD Dataset Shela; Siska Devella
INOVTEK Polbeng - Seri Informatika Vol. 11 No. 1 (2026): February
Publisher : P3M Politeknik Negeri Bengkalis

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35314/62samp73

Abstract

Cervical cancer remains a significant global health burden, largely due to limited screening coverage and the reliance on manual cytological interpretation. The intrinsic complexity of cervical cell morphology and constraints in clinical resources necessitate automated classification systems that are both accurate and computationally efficient. This study aims to evaluate and compare the performance of two lightweight CNN architectures, MobileNetV2 and GhostNetV2, for cervical cell image classification using the SIPaKMeD dataset. The dataset comprises 4,049 cell images, which were preprocessed through normalization, augmentation, and partitioning into training, validation, and testing sets. Both models were implemented using transfer learning and trained under comparable hyperparameter settings with basic data augmentation. Model performance was assessed using confusion matrices and standard evaluation metrics, including accuracy, precision, recall, and F1-score. Experimental results demonstrate that MobileNetV2 achieved superior performance with an accuracy of 98.50%, outperforming GhostNetV2, which attained a maximum accuracy of 97.60%. The consistent performance across metrics indicates robust and balanced classification capability. These findings suggest that MobileNetV2 offers an optimal trade-off between accuracy and computational efficiency, making it a promising candidate for deployment in resource-constrained and edge-based cervical cancer screening systems. Nevertheless, further external validation and clinical evaluation are required prior to real-world implementation.
Classification of Herbal Plant Images Using Transfer Learning EfficientNetV2-B3 Ambarwati, Rizki; Devella, Siska
INOVTEK Polbeng - Seri Informatika Vol. 11 No. 1 (2026): February
Publisher : P3M Politeknik Negeri Bengkalis

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35314/fz4jy549

Abstract

Herbal plants are natural resources that have high economic and health value, but the identification process is still done manually, making it prone to errors due to morphological similarities between species. This study aims to develop a leaf image classification model for herbal plants using a Convolutional Neural Network (CNN) with the EfficientNetV2-B3 transfer learning approach and AdamW optimizer. The dataset used is the Indonesian Herb Leaf Dataset 3500, which consists of 3,500 leaf images from 10 types of Indonesian herbal plants, namely Belimbing Wuluh, Jambu Biji, Jeruk Nipis, Kemangi, Lidah Buaya, Nangka, Pandan, Pepaya, Seledri, and Sirih. The research stages included preprocessing, dataset division, and augmentation such as flipping, rotation, zooming, contrast and brightness changes, translation, and the addition of Gaussian noise and salt-and-pepper noise to increase data variation and test model robustness. Evaluation based on accuracy, precision, recall, and F1-score shows that the model without augmentation achieved 98.57% accuracy, 98.63% precision, 98.57% recall, and 98.58% F1-score, while the model with augmentation and noise addition achieved an accuracy of 97.71%, precision of 97.83%, recall of 97.71%, and an F1-score of 97.72%. These results prove that EfficientNetV2-B3 is capable of effectively classifying herbal plant leaves with good performance.  
Comparison of MobileNetV3-Small and EfficientNetV2-Small for Low-Resolution X-ray Image Classification Andhika Rizky Cahya Putra; Siska Devella
INOVTEK Polbeng - Seri Informatika Vol. 11 No. 1 (2026): February
Publisher : P3M Politeknik Negeri Bengkalis

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35314/7j5twc37

Abstract

Lung diseases are a global health concern that require accurate and efficient automated diagnostic systems, particularly in healthcare facilities with limited resources. This study evaluates the performance and computational efficiency of two lightweight convolutional neural network architectures, namely MobileNetV3-Small and EfficientNetV2-Small, on the multi-label classification task of low-resolution ChestMNIST chest X-ray images. Experiments were conducted across eight testing scenarios with and without light spatial data augmentation. The evaluation encompassed predictive performance using accuracy and Area Under the Curve (AUC-ROC) metrics, as well as computational efficiency based on the number of parameters, FLOPs, model size, training time, and inference time. Results indicated that although both models achieved high accuracy (0.93–0.95), MobileNetV3-Small consistently produced higher and more stable AUC-ROC values compared to EfficientNetV2-Small, while being significantly more computationally efficient. Moreover, the application of light spatial data augmentation on low-resolution datasets such as ChestMNIST did not provide consistent performance improvements and instead increased training costs, indicating the limited effectiveness of simple geometric variations when spatial information in the images is highly constrained. These findings provide insight that, in low-resolution medical image multi-label classification, the suitability of an efficient CNN architecture design has a greater impact on overall performance than increasing model complexity or applying light spatial augmentation.
Classification of Chili Plant Pests Using the ConvNeXt Architecture Jocelyn, Jennifer; Devella, Siska
Jutisi : Jurnal Ilmiah Teknik Informatika dan Sistem Informasi Vol 15, No 1 (2026): Februari 2026
Publisher : STMIK Banjarbaru

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35889/jutisi.v15i1.3461

Abstract

Chili (Capsicum annuum L.) is a high-value horticultural commodity in Indonesia; however, its productivity often declines due to pest attacks that cause significant economic losses. This study aims to compare the performance of several ConvNeXt variants (V1 and V2) for chili pest classification using the Red Chili Pepper Pest dataset, which consists of four pest classes annotated with bounding boxes. The data were divided into training and testing sets, and a cropping process was applied to the object regions to ensure that the model focuses on pest images. The preprocessing stages included resizing, normalization, and data augmentation to improve model robustness against variations in image conditions. Model training was conducted using the timm library with uniform hyperparameter settings across all variants to ensure a fair comparison. Performance evaluation was carried out using accuracy, precision, recall, F1-score, and Area Under the Curve (AUC). In addition, computational complexity was analyzed in terms of the number of parameters, FLOPs, and inference latency. The results indicate that ConvNeXt V2 variants, particularly Nano and Tiny, achieve very high classification performance (macro-AUC > 0.99) with fewer parameters and lower latency compared to larger models. Robustness evaluation under various image degradations shows that Gaussian noise has the most significant negative impact on performance. Overall, ConvNeXt V2-Nano and V2-Tiny are recommended as the most efficient and stable models for implementing chili pest detection systems on resource-constrained devices within precision agriculture applications.Keywords: Chili Pest Classification; ConvNeXt; Deep Learning; Image Processing; Smart Agriculture.AbstrakCabai (Capsicum annuum L.) merupakan komoditas hortikultura bernilai tinggi di Indonesia, namun produktivitasnya sering menurun akibat serangan hama yang menyebabkan kerugian ekonomi. Penelitian ini bertujuan membandingkan kinerja varian ConvNeXt (V1 dan V2) dalam klasifikasi hama cabai menggunakan dataset Red Chili Pepper Pest yang terdiri atas empat kelas hama dengan anotasi bounding box. Data dibagi menjadi data pelatihan dan pengujian, kemudian dilakukan proses cropping pada objek untuk memastikan model berfokus pada citra hama. Tahapan prapemrosesan meliputi resizing, normalisasi, dan augmentasi untuk meningkatkan ketahanan model terhadap variasi citra. Pelatihan model dilakukan menggunakan pustaka timm dengan pengaturan hiperparameter pada seluruh varian untuk menjamin perbandingan adil. Evaluasi dilakukan menggunakan akurasi, presisi, recall, F1-score, dan AUC, serta analisis kompleksitas melalui jumlah parameter, FLOPs, dan latensi inferensi. Hasil penelitian menunjukkan ConvNeXt V2, khususnya Nano dan Tiny, mencapai performa tinggi (macro-AUC > 0,99) dengan kompleksitas komputasi lebih rendah. Uji robustness menunjukkan Gaussian noise memberikan penurunan performa paling signifikan. 
A Hybrid Deep Feature Based VGG19 and Support Vector Machine Approach for Durian Leaf Classification Santoti, Jennifer Velensia; Devella, Siska
Indonesian Journal of Artificial Intelligence and Data Mining Vol 9, No 1 (2026): March 2026
Publisher : Universitas Islam Negeri Sultan Syarif Kasim Riau

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24014/ijaidm.v9i1.38831

Abstract

Durian leaf classification has remained challenging due to high visual similarity among superior durian varieties and the limited robustness of conventional convolutional neural network models that rely on Softmax classifiers. This study aimed to address this limitation by investigating a deep feature-based classification framework that combined VGG19 as a feature extractor with a Support Vector Machine classifier. The experiments were conducted on a dataset of 1,530 durian leaf images representing four varieties: Bawor, Duri Hitam, Musang King, and Super Tembaga. Four experimental scenarios were designed to evaluate classification performance using Support Vector Machine and Softmax classifiers under both imbalanced and balanced data conditions through the application of Synthetic Minority Over-sampling Technique. The research gap addressed in this study lay in the absence of prior investigations that systematically evaluated the integration of VGG19 and Support Vector Machine for durian leaf variety classification under varying data distributions. Experimental results showed that the proposed VGG19–Support Vector Machine framework consistently achieved higher accuracy and more stable performance than Softmax-based models. This study demonstrated that replacing the conventional Softmax classifier with a Support Vector Machine significantly improved classification robustness compared to previous approaches that employed end-to-end convolutional neural network architectures.