Patch Feature Learning for Local Image Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection systems face inefficiencies when dealing with high-dimensional image data, particularly in identifying local patterns of anomalies due to resource-intensive data processing and limited labeled data, leading to reduced accuracy and task processing performance.
Innovation Solution
A patch feature learning method that performs feature-based learning on a pretrained model, reducing variance among similar patch features and enhancing differences between heterogeneous features through techniques like semi-supervised concept learning and neural network similarity calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all data for the entire image is used at once to detect the anomaly, then the anomaly detection can be performed on the complete image, but the resource for data processing and computation becomes inefficient
Solution Approach 1:
The patent divides the high-dimensional image data into multiple patches, processing each patch separately through the pretrained model. This segmentation approach reduces the computational burden of processing the entire image at once while maintaining anomaly detection accuracy by focusing on local patterns in each patch.
2Measurement precision
If conventional methods are used for anomaly detection, then the system can process images, but the ability to identify local patterns in small portions of the image is insufficient
Solution Approach 1:
The patent extracts multiple patches from the input image and processes each patch independently through the pretrained model to capture local anomaly patterns. This segmentation enables focused analysis of small regions while maintaining overall system simplicity by reusing the pretrained model architecture.
Solution Approach 2:
The patent applies local quality by treating each image patch as a distinct unit with its own feature extraction process. The model learns to identify anomaly patterns specific to local regions, enhancing the ability to detect localized defects without requiring complex global analysis.
3Measurement precision
If labeled data is limited or additional training is required on large amounts of unlabeled data, then the model can be trained, but the task processing performance for anomaly detection is limited
Solution Approach 1:
The patent uses a pretrained model that has already been trained on large datasets before being applied to the specific anomaly detection task. This preliminary training action allows the model to leverage pre-learned features and patterns, reducing the need for additional labeled data and training computations while maintaining high detection performance.
Data Source
AI summary
A patch feature learning method and a patch feature learning system for anomaly detection perform patch feature-based learning on a predetermined pretrained model based on an image data set for an anomaly detection target. The method and the system may acquire a feature map according to a first image data set; acquire a plurality of patch features based on local data in a predetermined image, based on the acquired feature map; perform feature representation learning based on the plurality of acquired patch features; acquire a reconstructing patch feature based on the performed feature representation learning; and perform anomaly detection based on the acquired reconstructing patch feature.