Color change recognition system based on hydrogen-sensitive color change film and leakage quantitative analysis method

Abstract

The invention provides an intelligent identification system based on a hydrogen-sensitive color-changing film and a leakage quantification method, and belongs to the crossing field of intelligent sensing and computer vision. The system adopts a multispectral imaging and artificial intelligence algorithm, and realizes high-precision quantification of optical response of the thin film through a three-level architecture of acquisition, analysis and edge calculation. The core comprises a 450 / 530 / 650 nm LED light source, + / -0.1 DEG C temperature control, a multispectral imaging unit and an RK3568 + NPU embedded system, and algorithms cover wavelet noise reduction, 14-dimensional feature extraction and OfficientNet-B2 analysis. The advantages are that: delta E = 0.5 sensitivity, self-adaptive light compensation is + / -0.3%, and response is lt; and 50ms. Experiments show that the delta R / R of 10ppm hydrogen is larger than or equal to 5%, the uncertainty of the full scale is 1.8%, the temperature adapts to-30 DEG C to 85 DEG C, circulation is stable, and 128-node industrial monitoring is supported.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of intelligent sensing and computer vision, specifically relating to an optical recognition system based on multispectral imaging and artificial intelligence, used for real-time monitoring and quantitative analysis of optical response changes in hydrogen-sensitive color-changing films. This system innovatively integrates spectral analysis technology, machine vision algorithms, and edge computing architecture to construct a complete scheme for the acquisition and analysis of hydrogen-sensitive color-changing signals. Background Technology

[0002] Hydrogen, as a clean energy carrier, is colorless, odorless, flammable, and explosive, making reliable leak detection technology crucial. While traditional hydrogen-sensitive color-changing films achieve visual detection in principle, they still face significant technical bottlenecks in practical industrial applications. These limitations are mainly reflected in four aspects: First, in terms of detection accuracy, existing visual detection methods heavily rely on the professional experience and subjective judgment of operators. Studies show that the human visual perception has a success rate of less than 60% in identifying subtle color changes with ΔE < 3, directly leading to concentration quantification errors generally reaching ±20% in practical applications. Second, there is the issue of environmental interference. Complex lighting conditions in industrial environments (including dynamic illuminance ranges of 300-20000 lx and the mixed use of various color temperature light sources) can cause significant measurement deviations. Laboratory simulation data shows that the color difference value deviation caused by the same hydrogen concentration under different lighting environments can be as high as ΔE = 5.7, which is the main reason why the false alarm rate of existing systems reaches 15-30%. Thirdly, there is a lack of dynamic response capability. Most commercially available detection equipment currently uses a static image acquisition frequency of 1-5Hz, which is completely unable to accurately capture the rapid transient response characteristics of hydrogen-sensitive materials within 500ms after contact with hydrogen gas. Finally, there is a serious lack of intelligence. Over 95% of existing products are still at the basic image acquisition stage, lacking both advanced data analysis capabilities and the ability to coordinate multiple sensors. In recent years, the development of multispectral imaging technology and breakthroughs in deep learning algorithms have provided new means for accurately detecting the optical characteristics of thin films, while the widespread adoption of edge computing devices has made real-time intelligent analysis possible. These technological advancements have propelled hydrogen detection systems towards higher precision, interference resistance, and intelligence, opening up effective paths to solve existing technical challenges. Summary of the Invention

[0003] This invention provides a hydrogen-sensitive color-changing thin-film optical recognition system based on multispectral imaging technology and artificial intelligence algorithms, aiming to solve key technical problems in traditional hydrogen leak detection methods, such as insufficient sensitivity, large environmental interference, and slow response speed. The system adopts a modular three-level architecture design of "optical acquisition - intelligent analysis - edge computing," and through the collaborative optimization of innovative multispectral dynamic illumination technology, high-precision color feature extraction algorithms, and an embedded AI processing platform, it achieves accurate quantitative analysis of the optical response characteristics of hydrogen-sensitive thin films. The optical acquisition module employs a three-wavelength tunable LED light source array of 450nm / 530nm / 650nm, combined with a Sony IMX342 back-illuminated CMOS sensor and an aspherical collimating lens with NA=0.65, to construct a multispectral imaging system with a reflectivity measurement accuracy of ±0.3%. The intelligent analysis module innovatively developed the HYD-AI algorithm framework, which integrates wavelet denoising, spatiotemporal feature extraction, and deep learning. Through multimodal analysis of 14-dimensional feature vectors and end-to-end processing of the EfficientNet-B2 optimized network, it achieved a color change detection sensitivity of ΔE=0.5. The edge computing module is built on an RK3568 processor and a dedicated NPU accelerator, with a low latency processing capability of less than 50ms. This system features a specially designed adaptive ambient light compensation mechanism and dynamic temperature correction algorithm, enabling it to maintain stable performance under a wide illuminance range of 3000-20000 lx and harsh temperature conditions of -30℃ to 85℃. It also supports distributed deployment of up to 128 monitoring nodes and LoRaWAN wireless networking, effectively solving the applicability issues of existing technologies in industrial settings.

[0004] The specific implementation scheme of the hydrogen-sensitive color-changing thin-film optical recognition system described in this invention is designed according to the systems engineering method, mainly including five key links: optical detection module construction, intelligent signal processing system implementation, mechanical packaging and environmental verification, field deployment scheme and performance verification. Regarding the optical detection module, a three-band LED array of 450nm (350mA), 530nm (300mA), and 650nm (400mA) is fixed on a high thermal conductivity copper substrate using Au80Sn20 eutectic solder. Each LED is equipped with an independent Peltier temperature control module (TEC1-12705) to achieve precise temperature control of ±0.1℃, forming an illumination light field with a uniformity >90%. The imaging unit uses a Sony IMX342 CMOS sensor paired with a 16mm focal length industrial lens (MTF>80%@100lp / mm), and the optical path parallelism is calibrated to <0.02° using a PI-624.2CD six-axis fine-tuning platform. The signal processing system adopts the RK3568+NPU hardware platform. The processing flow includes three levels: the preprocessing stage acquires 1024 frames of dark field images to establish a noise basis and uses the NIST SRM-2025 standard whiteboard to construct a three-dimensional lookup table; the feature extraction stage simultaneously performs 3×3 Sobel gradient operation (kernel weights [1 0-1; 2 0 -2; 1 0 -1]), 560-570nm / 645-655nm dual-band reflectivity ratio analysis, and ARIMA(2,1,1) time series modeling (autoregressive coefficient 0.32±0.05); the intelligent analysis stage uses an improved EfficientNet-B2 network architecture to process 14×60×60 feature tensors, and the output is optimized by Sigmoid activation (threshold 0.87) and 5-point moving average filtering. The system integration utilizes a Φ50×15mm 316L stainless steel casing (2mm wall thickness). Internal optical components are fixed with LOCTITE 4305 UV adhesive and have undergone rigorous environmental testing, including random vibration at 20-2000Hz (20Grms), temperature cycling at -30℃↔85℃ (50 cycles), and damp heat aging at 85℃ / 85%RH (1000 hours). During field deployment, nodes are arranged at 15m intervals indoors and 10m intervals outdoors, using the TDMA protocol (1s time slot, 37.5kbps rate) to construct a monitoring network of 128 nodes / LoRaWAN gateways. Performance verification was conducted using a GST-100F gas mixing system to generate standard gases ranging from 0.1 to 10000 ppm. Under conditions of 25℃±0.5℃ and RH 50%±3%, the system demonstrated a sensitivity of ΔR / R0≥5% at 10 ppm hydrogen, with a sensitivity decay of <2.9% after 8000 cycles. The full-range (0.1-10%) measurement uncertainty was 1.8% (k=2). Process control of the entire system relied on high-precision equipment such as the AJA Orion-8 magnetron sputtering system and the JAWoollam M-2000U ellipsometer to ensure the repeatability and reliability of all technical indicators.

Claims

1. An optical hydrogen sensing system based on a smart color-changing thin film, characterized in that... include: (1) Multispectral light source module, consisting of three-band LED arrays of 450nm, 530nm and 650nm, each LED is fixed by copper substrate eutectic bonding and equipped with an independent Peltier temperature control module to achieve temperature stability control within ±0.1℃; (2) High-precision imaging unit, including CMOS sensor and industrial lens combination with MTF value >80%@100lp / mm, the optical path parallelism is calibrated to <0.02° through fine adjustment platform; (3) Embedded signal processing system, performing three-level processing flow: a) Preprocessing stage to establish nonlinear compensation lookup table b) Feature extraction stage to synchronously calculate 3×3 Sobel spatial gradient and 560-570nm / 645-655nm band reflectivity ratio c) Intelligent analysis stage to adopt improved neural network processing with 14×60×60 feature tensor input; (4) Environmentally adaptable packaging design, using 316L stainless steel shell and verified by 20-2000Hz random vibration and -30℃↔85℃ temperature cycle.

2. The optical hydrogen sensing system according to claim 1, characterized in that: (1) A dynamic gas distribution system was used for calibration. At 25℃±0.5℃, it was verified that ΔR / R0≥5% when the hydrogen concentration was 10ppm and the measurement uncertainty of the full range (0.1-10%) was 1.8% (k=2). (2) Monitoring nodes were arranged at a spacing of 15m indoors and 10m outdoors to build a wireless sensor network based on the TDMA protocol.