CORE TECHNOLOGY EVIDENCE

Livestock Weight Inference AI

RGB-D · 3D Data · Pig Weight Inference · NPU Edge Deployment

ILLUVATION designed its own Weight Inference AI for the purpose of pig weight inference, and redesigned the base model of the server environment into a low-power and lightweight NPU class structure to apply it to actual pig farming field products. The published patent and PCT international application RGB-D··is public evidence that allows you to check the development flow of 3D-based weight inference technology.

Development/Optimization FlowOpen patent technology structure
TECHNOLOGY SUMMARY

Non-contact livestock weight inference based on RGB-D

The model structure is designed according to the purpose of weight inference and the field computing environment, and the product is optimized to operate in a NPU level edge environment.

INPUTRGB-D · Depth · 3DWe utilize 3D field data that combines video and depth information.
DATASET SCALE2M+ Pigs · 50+ Farms · 5+ BreedsCompleted ultra-precision AI model with empirical data from over 2 million pigs collected across 50+ farms and 5+ breeds.
OUTPUTEstimated WeightThe inferred weight of each individual is used for product operation and shipment judgment.
DEPLOYMENTNPU-class Edge AIConsidering both maintaining accuracy and reducing power and weight, we redesign the structure to fit the limited computing environment inside the product or equipment.
DEVELOPMENT CONTEXT

From Server AI to NPU Edge AI

We developed our own model tailored to the purpose of pig weight inference and RGB-D·3D field data, first implemented the base model for the server, and then redesigned the model structure for field application.

01 · DOMAIN REQUIREMENTDevelop purpose-specific own modelWe promoted the development of our own Weight Inference AI tailored to the purpose of pig weight inference and the characteristics of on-site RGB-D·3D data.
02 · SERVER MODELLarge-Scale Data Server AIImplemented the generalization base model and inference architecture using a dataset of over 2 million pigs and sufficient compute resources.
03 · NPU REDESIGNMaintain accuracy, low power, lightweightThe architecture was redesigned to reduce power requirements and model size while maintaining accuracy for operation in field equipment.
04 · FIELD DEPLOYMENTAI inside products and equipmentIn a NPU-level edge environment, Weight Inference AI is implemented to infer within field products and equipment.
PUBLIC PATENT ARCHITECTURE

RGB-D Weight Inference structure shown in published patent

Registered patent KR102340290B1, PCT international application PCT/KR2022/010504, and international publication WO2023008806A1 disclose a purpose-specific architecture that uses Weight Map, Patch, Spatial Volume Information, etc. to infer livestock weight based on RGB-D.

01RGB-D InputConfigure input data including RGB images and depth information.
02Livestock RegionExtract and configure livestock target areas and 3D information.
03Weight MapCreating a spatial representation for weight inference.
04Patch · Spatial VolumePatch unit features and spatial/volume information are used as model input features.
05Weight PredictionInfer livestock body weight through a purpose-specific prediction path.
WO2023008806A1KR102340290B1
TECHNOLOGY EVOLUTION

Technology development flow based on open patents and commercialization

The published patent record shows the accumulation of technology from 3D shape/volume-based estimation to RGB-D-based AI Weight Inference, and in the commercialization stage, the server AI was redesigned into the on-site NPU Edge model.

EARLIER 3D APPROACH

3D Point Cloud · Geometry

In the initial public patent group, a method of obtaining shape, volume, and length information using 3D scanning and point cloud and using this to estimate weight was proposed.

WO2020085597A1 →
RGB-D AI ARCHITECTURE

Weight Map · Patch · Spatial Volume

In the follow-up published patent group, an architecture was revealed that connects RGB-D-based data to AI weight prediction by composing Weight Map, Patch, and Spatial Volume Information.

WO2023008806A1 →
SERVER → NPU EDGE

Weight Inference AI · Architecture Redesign

In order to apply Weight Inference AI for servers to actual products, we redesigned it into a NPU-class model structure that takes into account both maintaining accuracy and reducing power and weight.

Viiew Product Family
TECHNICAL TERMINOLOGYView weight inference technical terms
Livestock Weight Inference AI
ILLUVATIONTechnical expression used to describe this livestock weight inference domain AI
Pig Weight Inference / Estimation
Research and industrial context to infer or estimate the body weight of pig individuals using images, sensors, and AI
RGB-D Pig Weight Estimation
Sensor/AI method to infer pig weight by combining RGB images and depth information
Non-contact Pig Weighing
Application method to obtain weight information without directly weighing the object with a physical scale
3D Point Cloud Weight Estimation
Method of using 3D geometric information such as point cloud shape and volume to estimate weight
PATENT EVIDENCE

Registered patents and PCT international applications

Domestic registered patent KR102340290B1
RGB-D-based livestock weight inference AI system and computer-readable medium on which the computer program for providing the same is recorded
PCT international application PCT/KR2022/010504
International disclosure WO2023008806A1
priority date 2021-07-30
PCT filing date 2022-07-19
International Publication Day 2023-02-02
Previous 3D technology revealed KR102122131B1 · PCT/KR2019/006097 · WO2020085597A1
Weight estimation technology using 3D Scan, Point Cloud, and shape/volume information
PRODUCT APPLICATION

Viiew Product Family

Weight Inference AI does not exist only as a research technology, but is connected to non-contact weight measurement products in actual pig farming fields.

PORTABLE

Viiew Max

A portable product that allows operators to move between pens, verifying target pig weights in approx. 1 second via non-contact sensing to support shipment sorting.

View Viiew Max
FIXED · REAL-TIME

Viiew Point

This is a stationary product that can be installed in a pig pen to check the weight of multiple pigs in a non-contact manner and support market-weight sorting.

View Viiew Point
DISCLOSURE BOUNDARY

Distinction between public patents and current product AI

The detailed structure of this page, including Weight Map, Patch, and Spatial Volume Information, explains the technical structure described in the published patent document. AI models in current commercial products are continually becoming more advanced, and this does not mean that all components of published patents apply equally to current production models. The current product applies its own real-time Weight Inference AI tailored to livestock weight inference purposes and NPU Edge Deployment that takes into account the field computing environment.
TECHNOLOGY · PRODUCT

Livestock weight inference technology and product application

Technology/R&D cooperation is linked at Industrial Physical AI, and actual farm introduction is linked at the Viiew product page.

Industrial Physical AIViiew MaxTechnology/R&D/SI cooperation consultation