A web-based clothing online store via similar image features with neural network approach

Abstract

The thesis provides the final output as a web application platform with various features based on the role of each participant on the website. Users can view, search as well as do the purchase and payment process. Admin post clothing items to the website, perform order tracking and confirmation. This application will run on an Express platform, connect to a database using MySQL, and use web building tools, techniques, and frameworks to provide a User Interface as a Web Application to visually represent how the web application works on the whole project. Moreover, a model is run below the web application system by using different features of classification work with CNN and bounding box recognition with YOLO, focusing on identifying different and most characteristic features from the image and using k NN to produce the same image results from the web data by calculating the difference between the embeddings. Streamlit is used to build a website applying model and establish the link to the main website. Moreover, different hyperparameters as well as pre-trained models including MobinetV3, Resnet50, VGG16, InceptionV3 are performed to evaluate to find the most suitable and accurate settings aiming to the goal of improving quality and accuracy.