Anomaly Detection Encoder Using Semantic and Visual Latent Vectors

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing anomaly detection methods are hindered by complex backgrounds and content variants, leading to reduced performance in distinguishing normal and anomalous data, as they fail to effectively separate semantic and visual appearance information.

Innovation Solution

A computing device comprising an encoder that generates semantic and visual appearance latent vectors, and a training module that updates parameters to improve the separation of these features, using a variational autoencoder-generative adversarial network (VAE-GAN) to enhance anomaly detection by focusing on semantic information for confidence scoring.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If existing anomaly detection methods are used, then the detection process is simple, but performance is reduced by complex backgrounds and content variants

Engineering Contradiction:

Improveanomaly detection performanceVSAvoidfeature separation mechanism

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the feature representation into two distinct latent vectors: semantic latent vectors and visual appearance latent vectors. This segmentation allows the system to separately process and analyze semantic content versus visual appearance, thereby improving anomaly detection performance by focusing on semantic consistency while being invariant to appearance variations caused by complex backgrounds or content variants.

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If semantic and visual appearance information are separated, then anomaly detection accuracy is improved, but the processing complexity increases

Engineering Contradiction:

Improveanomaly detection accuracyVSAvoidencoder and training module structure

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary training module that acts as a mediator between the encoder and the loss function. This training module orchestrates the joint training process by coordinating the optimization of semantic and visual appearance latent vectors through a composite loss function, thereby managing the processing complexity while achieving improved anomaly detection accuracy through structured feature separation.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Reliability

If feature separation is implemented, then the impact of complex backgrounds is reduced, but the model training becomes more challenging

Engineering Contradiction:

Improverobustness to complex backgroundsVSAvoidtraining process complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent employs parameter changes by introducing a composite loss function that dynamically adjusts the training objective to balance semantic accuracy and appearance reconstruction. The loss function incorporates multiple components (semantic loss, reconstruction loss, and regularization terms) that adaptively guide the training process, making it more manageable despite the increased complexity of training the feature separation mechanism.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS11748629B2Device and method of handling anomaly detection

Publication Date: 2023.09.05 MOXA INC

AI summary

A computing device for handling anomaly detection, comprises an encoder, for receiving an input image, to generate a first latent vector comprising a semantic latent vector and a visual appearance latent vector according to the input image and at least one first parameter of the encoder; and a training module, coupled to the encoder, for receiving the input image and the first latent vector, to update the at least one first parameter according to the input image and the first latent vector and a loss function.