Inaul Textile Pattern Classification
using CNN (Offline Mobile App)
A mobile-based computer vision system that classifies Inaul
textile patterns using a CNN model, designed for offline use to support
cultural preservation and accessibility in local communities.
PROBLEM AND MOTIVATION
Inaul textiles carry cultural identity but are difficult to systematically classify
Existing work is limited to documentation, not intelligent tools
Communities lack accessible technology for recognizing and learning patterns
SOLUTION OVERVIEW
Developed an end-to-end computer vision system that classifies Inaul textile patterns using deep learning, with a focus on both model performance and real-world usability.
Conducted a comparative study of three CNN architectures to evaluate performance, efficiency, and suitability for mobile deployment
Selected the most effective model based on accuracy and computational constraints
Integrated the trained model into a cross-platform mobile application using Flutter
Enabled real-time image classification through camera capture and image input
Optimized the system for offline functionality, ensuring accessibility in low-connectivity environments
DATASET
The dataset used for the Inaul textile classification model was curated through a combination of community collaboration and field data collection, with a focus on representing authentic textile patterns under real-world conditions.
Data Source:
Collected in partnership with community stakeholders, including the Al Qalam
Institute for Islamic Identities and Dialogue in Southeast Asia
Supplemented through field photography of Inaul textiles, captured in
natural environments to reflect realistic variations in lighting and texture
Additional samples were included when available from community-contributed
documentation
Annotation Process:
All images were labeled and validated with the assistance of cultural
and domain experts. Expert input was used to ensure correct identification of
textile patterns and reduce labeling ambiguity
Challenges in Data Collection:
Several challenges were encountered during dataset preparation:
Inconsistent lighting conditions due to field photography environments
High visual similarity between certain patterns, making classification non-trivial
Limited dataset size, requiring augmentation techniques to improve model generalization
Variability in image quality from different sources (community vs. field captures)
METHODOLOGY
The system was developed using a structured deep learning pipeline focused on model comparison, optimization, and mobile deployment constraints.
Pipeline:
Image preprocessing: resizing and normalization
Data augmentation: rotation, flipping, and scaling
Training: supervised learning with validation split
Evaluation: accuracy and loss monitoring across models
Deployment: Optimizing model for mobile use.
RESULTS
Inaul textiles carry cultural identity but are difficult to systematically classify
Existing work is limited to documentation, not intelligent tools
Communities lack accessible technology for recognizing and learning patterns
GITHUB