Denoising Model Training for Accurate Target Signal Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing detection systems face challenges in accurately detecting detection targets due to noise interference, necessitating effective noise cancellation methods.

Innovation Solution

A detection method involving a denoising model trained using noise signals to produce denoised signals, utilizing unsupervised machine learning models like autoencoders, particularly variational autoencoders, to enhance detection accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If noise cancellation is performed using conventional methods, then detection accuracy is improved, but the complexity of the detection system increases

Engineering Contradiction:
Improvedetection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces conventional mechanical or algorithmic noise filtering methods with a deep learning-based denoising model. The model uses neural networks to automatically learn and remove noise patterns from detection signals, substituting complex signal processing algorithms with a trained artificial intelligence system that achieves better noise cancellation performance while simplifying the overall detection pipeline.

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

2Ease of manufacture

If a denoising model is trained using unsupervised machine learning, then the training process becomes simpler, but the model requires more data and computational resources

Engineering Contradiction:
Improvetraining easeVSAvoiddata quantity
Core Design Contradiction:
Ease of manufactureVSQuantity of substance

Solution Approach 1:

The patent implements self-service by using unsupervised learning where the denoising model automatically learns noise patterns and cleaning strategies from raw data without requiring manual labeling or supervision. The system serves itself by autonomously identifying noise characteristics and optimizing its denoising capabilities through iterative training on unlabeled datasets, eliminating the need for expensive annotated training data.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4707776A1Detection method, detection system, noise removal model training method, and noise removal model training system
Publication Date: 2026.03.11 THE UNIV OF TOKYO
  • EP4707776A1 patent drawingFigure 1
  • EP4707776A1 patent drawingFigure 2
  • EP4707776A1 patent drawingFigure 3

AI summary

A detection system includes: a detection signal obtainer 306 configured to obtain a signal from a detection target as a detection signal; and a denoised signal obtainer 307 configured to input the detection signal into a denoising model 304 trained using a noise signal in the absence of the detection target to obtain at least one denoised signal that has undergone denoising.