Gas Leakage Image Processing Using LSTM Background Fluctuation Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gas leakage detection technologies, such as those using infrared cameras, struggle with false detections due to non-stationary background fluctuations like sunlight, clouds, wind, and dust, making it difficult to accurately separate airflow from background noise.
Innovation Solution
An information processing device and method utilizing a prediction model that incorporates temporal and spatial change rates in inspection images, trained using LSTM, to distinguish between gas presence and background fluctuations, enhancing detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If time-series frequency analysis is used to detect gas flow, then gas detection capability is provided, but false detection occurs due to background fluctuation
Solution Approach 1:
The patent segments the detection task by separating gas detection from general temperature fluctuation analysis. It uses a dedicated detection unit that specifically analyzes temperature changes in the gas absorption wavelength range, isolating the gas detection function from other thermal variations in the background.
Solution Approach 2:
The patent introduces an intermediary mechanism - a prediction model trained on temporal and spatial change rates - that acts as a filter between the raw infrared data and the final detection result. This intermediary learns to distinguish gas-related patterns from background noise through training data, reducing false detections.
2Duration of action of moving object
If infrared camera is fixedly installed for continuous monitoring, then monitoring coverage is improved, but detection accuracy decreases due to non-stationary background
Solution Approach 1:
The patent applies dynamics by making the detection system adaptive through the prediction model that learns from historical data. The system dynamically adjusts its detection thresholds and patterns based on trained temporal and spatial change rates, allowing it to maintain precision despite continuous operation and varying background conditions.
Solution Approach 2:
The patent changes the detection parameters from static threshold-based methods to dynamic parameters derived from trained models. The prediction model outputs adjusted detection criteria based on learned temporal and spatial patterns, enabling the system to adapt to non-stationary background conditions while maintaining continuous monitoring.
3Loss of information
If time-series frequency analysis is applied to temperature data, then airflow visualization is achieved, but separation of gas flow from background noise becomes difficult
Solution Approach 1:
The patent extracts only the relevant information for gas detection by focusing specifically on temperature changes within the gas absorption wavelength range. The detection unit isolates gas-related thermal signatures from the broader temperature field, removing irrelevant background information while retaining critical gas detection data.
Solution Approach 2:
The patent replaces traditional mechanical frequency analysis methods with a machine learning-based prediction model. This substitution enables the system to automatically learn and distinguish gas flow patterns from background noise through training, overcoming the limitations of conventional signal processing techniques.
Data Source
AI summary
An information processing device includes: an acquisition unit that acquires an inspection image obtained by capturing an inspection target; and a detection unit that detects presence of a gas in a vicinity of the inspection target by using a prediction model that is trained by an image captured in a case where the gas is not present to learn a fluctuation of a background, and in which a temporal and spatial change rate in the inspection image is set as an input parameter and information where whether or not the gas is present in the inspection image from which the fluctuation of the background is removed is detectable is set as an output parameter.