Article Details

Abstract

Aiming at the problems of object detection and classification in canteen dish recognition, this paper proposes a fusion algorithm based on computer vision. The model is trained on a self-built canteen dish dataset under the YOLOv7 environment, and various data augmentation and combination augmentation methods are used to evaluate model performance. Subsequently, Real-ESRGAN is introduced to perform denoising and super-resolution reconstruction on image data to improve the clarity and feature quality of input images. Experimental results on the self-built dataset containing 10 000 images show that compared with the original YOLOv7 method without Real-ESRGAN preprocessing, the improved model integrating Real-ESRGAN increases by approximately 5% in the mAP@0.5 index and by approximately 3% in the stricter mAP@0.5:0.95 index. Among them, the performance is optimal under the high-definition ×2 condition, where mAP@0.5 reaches 0.98 and mAP@0.5:0.95 is approximately 0.70. The results indicate that this method effectively improves the accuracy of dish recognition and has good generalization performance and application potential.

Keywords

YOLOv7; Real-ESRGAN; image recognition; algorithm optimization

References

1. [1] Xu Yue Research on Chinese cuisine image recognition algorithm based on deep learning [D]. Xi'an: Xijing University, 2022
2. [2] Li Lei Research on Chinese Cuisine Recognition Method Based on Deep Learning [D]. Nanjing: Nanjing University of Information Science and Technology, 2021
3. [3] Zhang Lulu, He Wenxue. A deep learning based automatic recognition and weighing system for fruits and vegetables [J]. Automation and Instrumentation, 2022 (6): 103-107
4. [4] Min Weiqing, Liu Linhu, Liu Yuxin, et al. Review of Food Image Recognition Methods [J]. Journal of Computer Science, 2022, 45 (3): 542-566
5. [5] Li Weizhe Supermarket Self checkout Fruit and Vegetable Recognition System Based on Neural Network Model [D]. Xi'an: Xi'an University of Architecture and Technology, 2021
6. [6] Jiang Ming Research on Lightweight Algorithm for Image Super Resolution Based on Deep Learning [D]. Ganzhou: Jiangxi University of Science and Technology, 2022
7. [7] Ren Jingjing Image Blurring Removal Based on Convolutional Neural Networks [D]. Hefei: Anhui University, 2017
8. [8] Ma Wenqi Research on Super Resolution Reconstruction of Plankton Images Based on Deep Learning [D]. Jinan: Shandong Normal University, 2022
9. [9] Song Zhiwei, Huang Xinbo, Ji Chao, et al. Defect detection and fault warning method for transmission line insulators based on Flexible YOLOv7 [J]. High Voltage Technology: 2023, 49 (12): 5084-5094
10. [10] Zhao Yuanlong, Dan Yugang, Yuan Jie. Improved Mask Wearing Pedestrian Tracking with YOLOv7 and DeepSORT [J]. Computer Engineering and Applications, 2023, 59 (6): 221-230
11. [11] Tan Xiao, Diao Yichao, Chen Xinjian, et al. Deep Supervised Feature Aggregation Network for High Myopia Stripe Damage [J]. Chinese Journal of Imaging and Graphics, 2022, 27 (3): 961-972

Funding

Key Project of Natural Science Research in Anhui Province (2024AH050526); Natural Science Research General Project of Anhui Vocational and Technical College of Commerce (2025KYZ01)

About the authors

Hou Tingting (1988-), female, Han ethnicity, born in Jiaozuo, Henan Province, China, teaching assistant, master's student, research direction: mathematical modeling, stochastic processes.