ML Container Classification via Hidden Layer Rejection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for classifying laboratory containers are inflexible and require multiple images from various angles, as well as reference to labels or machine-readable codes, to accurately determine container categories.

Innovation Solution

A computer-implemented method using a first machine learning model and a rejection algorithm to classify containers based on a single image, where the model is trained to recognize container categories and the rejection algorithm uses the output of a hidden layer to determine if the container belongs to any category.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple images from various angles are used to classify containers, then classification accuracy is improved, but the complexity of the processing system increases

Engineering Contradiction:
Improveclassification accuracyVSAvoidprocessing system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the essential visual features needed for classification from a single image, rather than requiring multiple images. The machine learning model is trained to identify container characteristics (shape, size, closure type) from limited visual information, eliminating the need for complex multi-angle imaging systems while maintaining classification accuracy

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent creates a digital representation (feature vector) of the container from a single image that captures all necessary classification information. This digital copy contains sufficient data for accurate classification without requiring physical manipulation or multiple viewing angles of the actual container

Inventive Principle:
Principle #26Copying

2Reliability

If multiple images and labels are required for container classification, then classification reliability is improved, but the time required for processing increases

Engineering Contradiction:
Improveclassification reliabilityVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The machine learning model is pre-trained on extensive datasets of container images with various characteristics. This preliminary training enables the model to reliably classify containers from single images during operation, eliminating the need for time-consuming multi-image capture and label reading processes while maintaining high classification reliability

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent replaces mechanical/optical systems (multi-angle cameras, label readers) with an intelligent software system. The machine learning model processes single images through computational analysis, substituting physical measurement and recognition systems with algorithms that achieve faster, equally reliable classification

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Loss of information

If conventional classification methods are used, then comprehensive container information is obtained, but the flexibility of the system decreases

Engineering Contradiction:
Improvecontainer information completenessVSAvoidsystem flexibility
Core Design Contradiction:
Loss of informationVSAdaptability or versatility

Solution Approach 1:

The patent changes the approach from fixed-rule classification to adaptive parameter-based classification. The machine learning model dynamically evaluates multiple container parameters (shape, size, closure characteristics) from single images and adapts its classification based on learned patterns, providing both comprehensive information extraction and flexible adaptation to diverse container types

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The machine learning model serves multiple functions: it classifies containers, identifies characteristics, and adapts to different container types using the same single-image input mechanism. This universal approach replaces multiple specialized systems (multi-angle imaging, label reading, manual inspection) with one flexible model that handles all classification needs

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4557231A1Laboratory container processing
Publication Date: 2025.05.21 BECKMAN COULTER INC
  • EP4557231A1 patent drawingFigure 1
  • EP4557231A1 patent drawingFigure 2
  • EP4557231A1 patent drawingFigure 3

AI summary

Aspects relate to a computer implemented method, a computer program, a computer readable medium, a data processing system and a laboratory instrument. The computer-implemented method comprises obtaining a first image depicting at least a portion of a container and determining, using a first machine learning model, the first image and a first rejection algorithm, whether the container depicted in the first image belongs to one container category of a plurality of container categories or the container depicted in the first image does not belong to any container category of the plurality of container categories. The first rejection algorithm uses an output of a hidden layer of the first machine learning model. The first machine learning model is trained to classify containers in the plurality of container categories using training images of containers. Each of the training images of the containers corresponds to exactly one container category of the plurality of container categories.