3D Food Weight Estimation Using AI for Density Variations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the weight of food objects using 3D surface profiles are inaccurate due to air-pockets and variations in density, leading to incorrect volume and weight calculations.
Innovation Solution
Employing an artificial neural network software module trained on 3D image data and weight data of food objects to correlate specific shapes with density variations, allowing for accurate weight estimation by identifying air-pockets and non-uniform densities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3D surface profiles are used to calculate volume and weight of food objects, then weight estimation can be performed without direct weighing, but air-pockets and shape variations cause measurement inaccuracy
Solution Approach 1:
The patent replaces the traditional mechanical assumption-based volume calculation system with an AI-based image analysis system. Instead of using geometric formulas that assume uniform shapes, the system uses deep learning models trained on actual food object images to directly estimate weight, automatically accounting for air-pockets, irregular shapes, and density variations without requiring explicit volume measurement.
Solution Approach 2:
The patent changes the estimation approach from direct volume measurement to direct weight estimation using AI. The system transforms the problem from measuring physical dimensions (length, width, height) and calculating volume through mathematical formulas to using image pixel analysis and neural network predictions that directly output weight estimates, bypassing the intermediate volume calculation step that introduces errors.
2Device complexity
If fixed density assumption is made for weight calculation, then calculation process is simplified, but density variations within and between food objects lead to additional inaccuracy
Solution Approach 1:
The patent replaces the simple but inaccurate fixed-density assumption with an AI-based system that learns density variations from training data. The neural network automatically adapts to different food types, shapes, and density distributions without requiring explicit density measurements or complex multi-step calculations, achieving both accuracy and computational efficiency.
3Ease of operation
If line laser and camera system is used to capture 3D profiles, then non-contact measurement is achieved, but air-pockets between food object and support surface are misidentified as food material
Solution Approach 1:
The patent replaces the optical 3D scanning system that creates geometric models with an AI-based image analysis system. Instead of using laser line projection and camera-based surface reconstruction that are sensitive to air-pockets, the system uses standard imaging to capture food objects and applies deep learning to directly estimate weight from image features, making the measurement robust to support surface contact conditions.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Significantly improves weight estimation accuracy by compensating for air-pockets and density variations, reducing estimation errors by approximately 44% compared to traditional methods.
Implementation Method 1
the reflection of the light from the surface of the food object is detected e.g.by a camera 103 and used to make the 3D surface profile
Data Source
Figure 1(a)~3
Figure 4~6
Figure 7~10
AI summary
This disclosure relates to a method of estimating weight of food objects, comprising training an artificial neural network software module, and using the trained artificial neural network software module to provide a weight correlated data estimate for said food object based on a three dimensional (3D) image of the food object.