Deep Learning Food Detection Reducing Labeling Labor
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current food detection and identification methods in dining halls face challenges such as high costs and labor-intensive manual labeling for deep learning solutions, and high costs and environmental limitations of RFID-based systems, which hinder large-scale adoption.
Innovation Solution
A method utilizing a general multi-target positioning network and a classification network trained through deep convolutional neural networks, where the multi-target positioning network is trained once and can be applied across all restaurants, and the classification network identifies food without the need for extensive manual labeling, reducing labor and costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling of food pictures is performed to train deep learning models, then identification accuracy is improved, but labor cost and time consumption increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-training a general multi-target positioning network on a large-scale food picture database before deployment. This pre-trained network can automatically locate and identify food items without requiring manual labeling at each restaurant, thus improving identification accuracy while reducing the time and labor needed for data preparation.
Solution Approach 2:
The patent uses copying by transferring the knowledge learned from the general multi-target positioning network to specific restaurant scenarios. The pre-trained network serves as a template that can be adapted to different restaurants without requiring complete re-labeling, thereby maintaining high identification accuracy while minimizing manual labeling efforts.
2Measurement precision
If manual labeling of food pictures is performed to train deep learning models, then identification accuracy is improved, but labor cost increases
Solution Approach 1:
The patent applies preliminary action by pre-training a general multi-target positioning network on a large-scale food picture database before deployment. This pre-trained network can automatically locate and identify food items without requiring manual labeling at each restaurant, thus improving identification accuracy while reducing the time and labor needed for data preparation.
Solution Approach 2:
The patent uses copying by transferring the knowledge learned from the general multi-target positioning network to specific restaurant scenarios. The pre-trained network serves as a template that can be adapted to different restaurants without requiring complete re-labeling, thereby maintaining high identification accuracy while minimizing manual labeling efforts.
3Measurement precision
If RFID tags are used for food identification, then identification accuracy is improved, but cost increases and the system is not suitable for high temperature environments
Solution Approach 1:
The patent replaces the RFID mechanical/electronic system with a computer vision-based deep learning system. Instead of using RFID tags that are sensitive to high temperatures, the system uses image processing and neural networks to identify food items, which are not affected by temperature conditions. This substitution maintains identification accuracy while eliminating temperature-related limitations.
4Adaptability or versatility
If a general deep learning model is trained for all restaurants, then large-scale promotion is facilitated, but the model requires extensive manual labeling data
Solution Approach 1:
The patent applies preliminary action by pre-training a general multi-target positioning network on a large-scale food picture database before deployment. This pre-trained network can automatically locate and identify food items without requiring manual labeling at each restaurant, thus improving identification accuracy while reducing the time and labor needed for data preparation.
Solution Approach 2:
The patent achieves universality by creating a general multi-target positioning network that can be applied across multiple restaurants with different food menus and styles. The pre-trained network serves multiple functions: it can identify various types of food, adapt to different restaurant environments, and reduce the need for restaurant-specific data collection and labeling, thereby facilitating large-scale promotion.
Data Source
AI summary
The present invention discloses a food detection and identification method based on deep learning, which realizes food positioning and identification by a deep convolutional network. The method comprises: firstly, training a general multi target positioning network and a classification network by using food pictures; secondly, inputting the results of the positioning network into the classification network; finally, providing a classification result by the classification network. The method uses two deep convolutional networks with different functions to respectively detect and identify the food, which can effectively reduce the labeling cost of the food and improve the accuracy of positioning and identification.