Anomaly Detection Score Map Correction for False-Positive Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional anomaly detection systems struggle to accurately distinguish between normal and anomalous sections in input data, often misidentifying normal sections as anomalies due to fluctuations in data structure, leading to overdetection.
Innovation Solution
An anomaly detection device that includes an anomaly score map calculating unit, a correction score map calculating unit, and a correcting unit, which utilize feature maps of normal data to correct anomaly scores, reducing overdetection by accounting for data structure fluctuations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional anomaly detection methods are used to identify anomalous sections in input data, then anomaly detection can be performed, but normal sections resembling anomalous ones are misidentified as anomalies, reducing detection accuracy
Solution Approach 1:
The patent segments the anomaly detection process into two distinct components: (1) an anomaly score map that identifies potential anomalies, and (2) a correction score map that refines the detection by accounting for data structure fluctuations. This segmentation allows the system to separately handle anomaly identification and false positive reduction, thereby improving both detection accuracy and reliability.
Solution Approach 2:
The correction score map acts as an intermediary between the raw anomaly score map and the final anomaly detection result. By introducing this intermediate component that specifically addresses data structure fluctuations, the system can adjust and refine anomaly scores before making final detection decisions, reducing false positives while maintaining detection accuracy.
2Measurement precision
If anomaly detection is performed using difference processing or feature space subtraction, then anomalous sections can be identified, but the system cannot distinguish between anomalous sections and normal sections with similar features
Solution Approach 1:
The patent applies local quality by creating a correction score map that specifically addresses local data structure characteristics. Instead of applying a uniform detection threshold across the entire dataset, the system locally adjusts anomaly scores based on position-specific fluctuations in the data structure, enabling better distinction between normal and anomalous sections with similar features.
Solution Approach 2:
The system changes the parameter used for anomaly assessment by introducing a correction factor derived from data structure fluctuations. Rather than relying solely on the raw anomaly score, the system modifies the assessment parameter by incorporating positional information about normal fluctuations, thereby improving the ability to distinguish between normal and anomalous sections.
Data Source
AI summary
An anomaly detection device according to an embodiment includes an anomaly score map calculating unit, a correction score map calculating unit, and a correcting unit. The anomaly score map calculating unit calculates an anomaly score map indicating the degree of anomaly of the input data. The correction score map calculating unit uses the difference in the feature of one or more pieces of normal data included in a normal dataset, and calculates one or more correction score maps for correcting the anomaly score map. The correcting unit corrects the anomaly score map using the correction score maps.