A Transfer Learning-Based Multi-Output Framework for Apparel Type and Seam Pattern Classification

Abstract

In today’s rapidly evolving textile and fashion industry, automated classification of garments and stitching styles has become essential for applications such as quality control, inventory tracking, and intelligent sorting systems. The increasing demand for smart manufacturing solutions highlights the need for reliable image-based garment analysis techniques. Conventional methods that rely on manual inspection or basic image processing are no longer adequate, as they struggle to handle large-scale data with consistent accuracy. These traditional approaches often lack scalability, produce inconsistent outcomes, and fail to capture fine-grained stitching details, especially under complex visual conditions. Moreover, the absence of machine learning and advanced feature extraction techniques limits their effectiveness, leading to higher error rates and inefficiency in real-time environments. To address these challenges, the proposed system introduces an intelligent classification framework that leverages transfer learning using EfficientNetB0 for deep feature extraction, combined with machine learning models such as Deep Neural Networks (DNN), Perceptron, and a proposed Multi-Layer Perceptron (MLP) for multi-output classification of both garment types and stitching patterns. The entire system is implemented through a Python-based Tkinter graphical interface, enabling users to upload datasets, preprocess images, train models, and perform real-time predictions seamlessly. By integrating deep learning capabilities with an interactive interface, the system enhances classification accuracy, reduces dependency on manual inspection, and offers a scalable and efficient solution for modern smart textile manufacturing environments.