Research on Extracting Clothing Style Images Based on Computer Vision Technology

Deep learning-based image recognition and generation technology is a key aspect of visual artificial intelligence, and the rapid development of AI technology is driving the clothing industry towards digitization. How to automatically and accurately obtain relevant clothing demand information to meet the severe market challenges has become a focal point for major clothing e-commerce platforms. This paper employs deep learning techniques to construct CNN classification models and object detection models for the automatic recognition of multiple features, including clothing color, style, and patterns. The study utilizes an improved lightweight VGGNet network for recognizing colors and styles, combined with YOLOv3, Faster R-CNN, and SSD object detection algorithms to achieve pattern recognition and localization. Through comparative experiments, the recognition accuracy of the improved lightweight VGGNet in identifying clothing styles and colors reached 96.49%; in clothing pattern detection, YOLOv3 achieved the highest mean Average Precision (mAP) of 86.66%. Among the three categories of patterns, texture patterns demonstrated the best detection performance with a mAP of 96.14%, followed by animal patterns with an mAP of 83.69%, and text patterns with an mAP of 79.80%. The results indicate that the graphic recognition algorithm proposed in this paper can effectively extract and identify patterns on clothing, thereby promoting the digital transformation of the clothing industry.