基於深度學習之線上顧客穿搭風格推薦系統

An Online Outfit Style Recommendation System for Customers Based on Deep Learning

隨箸科技進步及網路的發展，穿著不再只有舒適、保暖的功能，而是更注重整體的搭配、造型。因而產生了許多不同的風格。為了提供消費者能夠更快速、便利的找尋適合自己風格的服飾，我們建立了一個穿搭推薦系統。
本研究使用深度學習演算法YOLOv8s進行訓練及預測，以實現影像辨識功能。透過蒐集三種風格服飾圖片建立資料庫，並使用YOLOv8s對三種風格服飾圖片進行貼標籤及辨識。從研究結果可知，YOLOv8s在預測歐美風的服飾影像有100%的正確率，韓系風為99.97%，古著為99.94%，訓練完成後，驗證測試及辨別風格準璀率可達91%。藉此我們經由YOLOv8s建置服飾風格資料庫，打造穿搭推薦系統，提供給用戶使用。

With the advancement of technology and the development of the internet, clothing is no longer solely focused on comfort and warmth, but has evolved to emphasize overall coordination and personal style. As a result, many distinct fashion styles have emerged. To help consumers more quickly and conveniently find clothing that matches their preferred style, we have developed a fashion outfit recommendation system.
This study employs the deep learning algorithm YOLOv8s for training and prediction to achieve image recognition capabilities. By collecting images of clothing from three different fashion styles, we constructed a database and used YOLOv8s to label and recognize these images. The experimental results show that YOLOv8s achieves a 100% accuracy rate in recognizing European-American style clothing, 99.97% for Korean style, and 99.94% for vintage style. After training, the system achieves an overall accuracy of 91% in validation testing and style classification. Through the implementation of a fashion style database using YOLOv8s, we have built an outfit recommendation system available for user application.