Design and Implementation of a Web-Based Visual Search System for MSME E-Commerce Using the Flask Framework

This research presents the design and implementation of an end-to-end web-based visual search system for MSME e-commerce using the Flask framework and a VGG16-based convolutional neural network. The system addresses two critical challenges commonly faced by MSME digital platforms: product tagging errors during product uploads by sellers and limitations of text-based search for customers. A dual-model architecture is implemented, consisting of a visual search module for similarity-based image retrieval and a backend classification module for automatic product categorization. The system is evaluated using a locally collected MSME product image dataset from the Tapal Kuda region, achieving a classification accuracy of 89.17% and visual search performance with a macro precision of 0.85, macro recall of 1.0, and macro F1-score of 0.91. To support real-time deployment, visual features are pre-extracted and stored, enabling efficient query processing with response times under 2 seconds during concurrent usage testing. The results demonstrate that the proposed system provides effective and practical visual search functionality within a localized MSME context while maintaining feasible computational requirements, making it suitable for deployment in resource-constrained MSME environments.