A Method for Identifying Women's Sleeves Based on Improved YOLOv5 and ResNet50

Abstract

Abstract: In response to the problems of numerous classifications of women's clothing sleeve shapes, difficulty in feature recognition, and unsatisfactory detection results, on the basis of fully utilizing the correlation information between different women's clothing sleeve shapes, an improved YOLOv5 object detection and ResNet50 object classification deep learning method was used to achieve automatic recognition of women's clothing sleeve shapes.
Firstly, two methods of sleeve type classification were combined based on the length and shape of women's clothing sleeves. The sleeves were divided into four primary classifications based on the length of women's clothing: sleeveless, ultra short, short, and long sleeves. On the basis of the primary classifications, they were further divided into 15 secondary classifications based on their morphological characteristics, including bra, neck hanging, raglan sleeves, bubble sleeves, and leg sleeves. Secondly, clothing sample images were collected from e-commerce platforms, taking into account factors such as different angles, lighting, and backgrounds. On the basis of balancing different sleeve types, 3600 sleeve type images of women's clothing were collected, screened, and labeled. A clothing sleeve type dataset containing approximately 6500 sleeve type samples was obtained, and sleeve type labeling was performed using Labelimg software. Once again, based on the analysis of the YOLOv5 object detection network, CBAM attention mechanism, ResNet50 residual network principle, and network features, an improved YOLOv5 and ResNet50 combined deep learning method based on CBAM attention mechanism is proposed for women's sleeve automatic recognition. Among them, YOLOv5 model gradually adjusts the parameters of the network model through the back propagation and gradient descent characteristics of the Convolutional neural network on the self labeled garment sleeve shape data set to obtain the network parameters suitable for the detection of women's sleeve shape, thus realizing the target detection of women's sleeve shape in the primary level classification. The convolutional attention module CBAM, which combines channel attention mechanism and spatial attention mechanism, is beneficial for solving the problem of no attention preference in the original network, thereby enhancing the effectiveness of sleeve detection. Four independent ResNet50 residual networks were used to carry out sleeve type secondary classification recognition based on the improved YOLOv5 network detection of four sleeve types: sleeveless, ultra short sleeved, short sleeved, and long sleeved, respectively, in order to obtain the final results of women's sleeve type recognition. Finally, based on the Python language and Pytorch framework, the proposed deep learning algorithm for women's clothing sleeve recognition was designed and implemented, and the model was trained on the sleeve dataset to verify the effectiveness of sleeve recognition through experiments.
The results indicate that ① compared to the YOLOv5 method and the CBAM improved YOLOv5 method, the CBAM improved YOLOv5 and ResNet50 combined method, which introduces the correlation information between women's sleeve shapes, has more advantages in the accuracy of women's sleeve shape recognition. The overall recognition accuracy is about 93.3, and its overall accuracy is 12.2 and 8 percentage points higher than the YOLOv5 model improved by YOLOv5 and CBAM, respectively; ② in the task of identifying women's sleeve type by YOLOv5, improved YOLOv5, YOLOv5 and ResNet50 combined methods, compared with sleeveless and long sleeves, the identification of ultra-short sleeves and short sleeves is more difficult, and the overall accuracy is more difficult to improve.

Key words: women's sleeve shape, deep learning, YOLOv5；attention mechanism, ResNet50

摘要： 针对女装袖型分类繁多、特征识别困难、检测效果不理想等问题，根据不同女装袖型的关联信息，结合注意力机制改进的YOLOv5目标检测网络和ResNet50残差网络，提出了一种女装袖子造型的自动识别方法。首先，从电商平台收集服装样本图像，按照长短大类和形态小类标记对女装袖型进行归类，建立了包含3600张图像的袖型数据集；其次，结合注意力机制改进的YOLOv5目标检测网络和ResNet50残差网络，设计了女装袖型识别方法；最后，在袖型数据集上开展模型训练，并通过实验验证袖型识别的效果。结果表明：改进YOLOv5和ResNet50相结合的深度学习方法可以有效地对女装袖型进行识别，整体识别准确率约93.3%。该女装袖型识别方法准确、便捷，可以实现大量服装款式的分类快速检测，提高服装设计效率，促进人工智能技术在服装设计领域的应用，助力我国智能制造和电子商务的发展。

关键词: 女装袖型, 深度学习, YOLOv5, 注意力机制, ResNet50

CLC Number:

TS941.26

CAO Hanying, TUO Jiying. A Method for Identifying Women's Sleeves Based on Improved YOLOv5 and ResNet50[J]. Advanced Textile Technology, 2024, 32(1): 45-53.

曹涵颖, 妥吉英. 基于改进YOLOv5和ResNet50的女装袖型识别方法[J]. 现代纺织技术, 2024, 32(1): 45-53.

References

[1] 陈亚亚, 孟朝晖. 基于目标检测算法的FashionAI服装属性识别[J]. 计算机系统应用, 2019, 28(8):170-175.
CHEN Y Y, MENG Z H. FashionAI clothes recognition based on object detection algorithm[J]. Computer Systems & Applications, 2019 28(8):170-175.
[2] 周正中. 基于感知与推理模型的服装图像检索和推荐系统[D]. 上海：上海交通大学, 2019:12-25.
ZHOU Z Z. Clothing image retrieval and recommendation system based on perception and reasoning model [D].Shanghai：Shanghai Jiao Tong University, 2019:12-25.
[3] JING J, ZHUO D, ZHANG H, et al. Fabric defect detection using the improved YOLOv3 model[J]. Journal of Engineered Fibers and Fabrics, 2020, 15:155892502090826.
[4] KIAPOUR M H, HAN X F, LAZEBNIK S, et al. Where to buy it: matching street clothing photos in online shops[C]// IEEE International Conference on Computer Vision. Santiago, Chile. IEEE, 2016: 3343-3351.
[5] CHEN L, YE J T, JIANG L G, et al. Synthesizing cloth wrinkles by CNN-based geometry image superresolution: Synthesizing cloth wrinkles by CNN-based geometry image SR[J]. Computer Animation and Virtual Worlds, 2018, 29(3-4):e1810.
[6] 刘正东, 刘以涵, 王首人. 西装识别的深度学习方法[J]. 纺织学报, 2019, 40(4):158-164.
LIU Z D, LIU Y H, WANG S R. Depth learning method for suit detection in images [J]. Journal of Textile Research, 2019, 40 (4):158-164.
[7] 赵宏伟，刘晓涵，张媛，等. 基于关键点注意力和通道注意力的服装分类算法[J]. 吉林大学学报(工学版), 2020, 50(5):1765-1770.
ZHAO H W, LIU X H, ZHANG Y, et al. Clothing classification algorithm based on landmark attention and channel attention [J]. Journal of Jilin University (Engineering and Technology Edition), 2020, 50(5) :1765-1770.
[8] 邓莹洁, 罗戎蕾. 基于卷积神经网络的半身裙款式特征分类识别[J]. 现代纺织技术, 2021,29(6):98-105.
DENG Y J, LUO R L. Classification and recognition of bust skirt style and common features based on convolutional neural networks [J]. Advanced Textile Technology, 2021,29(6):98-105.
[9] 陈金广，李雪，邵景峰，等. 改进YOLOv5网络的轻量级服装目标检测方法[J].纺织学报，2022,43(10):155-160.
CHEN J G, LI X, SHAO J F, Ma Lili, et al. Lightweight clothing detection method based on an improved YOLOv5 network [J]. Journal of Textile Research, 2022, 43 (10) :155-160.
[10] 尹光灿，罗戎蕾．基于卷积神经网络的服装领型识别与分类研究[J]. 现代纺织技术, 2020, 28(3):48-53.
YIN G C, LUO R L. Research on recognition and classification of garment collar type based on convolutional neural networks [J]. Advanced Textile Technology, 2020, 28 (3): 48-53.
[11] 庹武，郭鑫，张启泽，等. 基于SE－Inception v3与迁移学习的服装袖型识别与分类[J]. 毛纺科技, 2022，50(10): 99-106.
TUO W, GUO X, ZHANG Q Z, et al. Clothing sleeve recognition and classification based on SE-Inception v3 and transfer learning [J]. Wool Textile Journal, 2022, 50 (10): 99-106.
[12] CHEN Y, ZHANG C, QIAO T, et al. Ship detection in optical sensing images based on YOLOv5[C]// Twelfth International Conference on Graphics and Image Processing (ICGIP 2020). SPIE, 2021, 11720: 102-106.
[13] WANG Q, CHENG M, HUANG S, et al. A deep learning approach incorporating YOLO v5 and attention mechanisms for field real-time detection of the invasive weed Solanum rostratum Dunal seedlings[J]. Computers and Electronics in Agriculture, 2022, 199: 107194.
[14] ZHU X K, LYU S C, WANG X, et al. TPH-YOLOv5: Improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision, Montreal, BC, Canada. IEEE,2021: 2778-2788.
[15] 陆建波, 谢小红, 李文韬. 一种改进残差网络的服装图像识别模型[J]. 计算机工程与应用, 2020, 56(20):206-211.
LU J B, XIE X H, LI W T. Improved clothing image recognition model based on residual network [J]. Computer Engineering and Applications, 2020, 56 (20): 206-211.