Related Experiment Video
Updated: Jan 11, 2026
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
Constructing a deep learning-assisted smartphone application for intelligent recognition of steak doneness during
Han Wang1, Yuandong Lin2, Jun-Hu Cheng1
1School of Food Science and Engineering, South China University of Technology, Guangzhou 510641, China; Academy of Contemporary Food Engineering, South China University of Technology, Guangzhou Higher Education Mega Centre, Guangzhou 510006, China; Guangdong Provincial Key Laboratory of Intelligent Food Manufacturing, Foshan University, Foshan 528225, China.
None:
The recognition of steak doneness during cooking poses challenges for consumers. In this study, the framework of deep learning-driven non-destructive steak doneness assessment during cooking was constructed. In addition, physicochemical properties were integrated to ensure the reliability of the doneness processing method and database, thereby improving the accuracy and persuasiveness of the model. Based on this, reliable databases including 1803 images and physicochemical properties of 153 samples were created. These databases were obtained respectively from 601 and 51 different steaks. After comparing 8 models covering the linear model, machine learning models and deep learning models, DenseNet121, a type of convolutional neural network (CNN) model, was found to be the best model with a recognition accuracy of 95.30 ± 1.48 %. Subsequently, gradient weighted class activation mapping (Grad-CAM) and other visualization methods confirmed the feasibility and principles of the model. Finally, a mobile-optimized edge computing custom-built application was developed to demonstrate the real-time performance of the proposed model which achieved an accuracy of 91.93 %. These results demonstrated the reliability of CNN models in predicting steak doneness and provided a theoretical basis for smart cooking devices.
