Sensor Behavior Monitoring for Unrecognized Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning-based behavior recognition models struggle to recognize certain behaviors, such as abnormal behaviors in objects like calves, leading to difficulties in monitoring and managing these behaviors effectively.
Innovation Solution
A system and method that utilizes sensor data to identify pattern data on unidentified behaviors not recognized by machine learning models, determining target data based on measurement frequency and magnitude, and providing information to users to assist in monitoring and managing these behaviors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning-based behavior recognition models are used to monitor object behaviors, then common behaviors can be recognized, but abnormal or unidentified behaviors cannot be detected
Solution Approach 1:
The patent segments behavior monitoring into two independent modules: a machine learning-based recognizer for common behaviors and an anomaly detector for unidentified behaviors. This segmentation allows each module to specialize in its respective function, resolving the contradiction between recognition accuracy for known behaviors and coverage of all behavior types including abnormal ones.
Solution Approach 2:
The patent introduces an intermediary anomaly detection module that bridges the gap between machine learning recognition and complete behavior monitoring. This intermediary component captures behaviors that escape the primary ML model, thereby expanding behavior type coverage without compromising the reliability of common behavior recognition.
2Loss of information
If all sensor data is collected and analyzed, then complete behavior information is obtained, but data processing complexity and time increase
Solution Approach 1:
The patent extracts only the essential features needed for anomaly detection from the full sensor data stream. By taking out and focusing on specific relevant features rather than processing all raw data, the system maintains behavioral information completeness while significantly reducing data processing time and computational complexity.
Solution Approach 2:
The patent applies partial action by selectively processing data only when anomalies are detected or when specific conditions are met. Instead of continuously analyzing all sensor data, the system performs targeted analysis on relevant portions, thereby maintaining information completeness for critical behaviors while minimizing unnecessary processing time.
Data Source
AI summary
A method for monitoring an object is provided. The method includes the steps of: specifying, from sensor data measured by a sensor for at least one object, pattern data on an unidentified behavior of the at least one object that is not recognized by a machine learning-based behavior recognition model; and determining at least a part of the pattern data as target data on the basis of at least one of a measurement frequency and a measurement magnitude of the pattern data, and providing information on the target data to a user.