Material identification method in high-temperature dust multi-condensation environment

Abstract

The invention discloses a material identification method in a high-temperature dust multi-condensation environment. The method comprises the following steps: S1, multi-modal data synchronous acquisition: acquiring data by adopting a three-light fusion sensor; s2, environmental parameter real-time monitoring: acquiring environmental data through a dust sensor and a hygrothermograph; s3, dynamic data fusion: fusing function infrared features, radar features and spectral features by using an adaptive weighted fusion algorithm; s4, condensation interference elimination: performing polarization hyperspectral processing, calculating Stokes vector separation specular reflection and diffuse reflection, and reconstructing a spectral curve without water film interference; s5, material matching identification: inputting the processed features into a pre-training model; and S6, triggering self-cleaning. According to the method, thermal infrared, millimeter wave radar and hyperspectral three-waveband data are fused, the identification bottleneck of a single mode in an extreme environment is broken through, the problem that material identification is difficult in the extreme environments of high temperature, high dust, multiple condensation and the like is solved, and the method has outstanding industrial popularization prospects.

Description

Technical Field

[0001] This invention relates to the field of material identification technology, specifically a method for material identification in high-temperature, dusty, and condensing environments. Background Technology

[0002] In industrial settings such as metallurgy, cement, and thermal power generation, real-time identification of material composition is crucial for process control, product quality improvement, and energy efficiency optimization. Especially in extreme environments such as high temperatures (≥80℃), high dust levels (concentration ≥50mg / m³), and high condensation levels (humidity ≥90%RH), achieving stable and accurate identification of material types and compositions has become a key technical challenge for the development of intelligent industrial processes.

[0003] Currently, commonly used material identification methods in industrial settings include traditional visual recognition, laser ranging, infrared thermal imaging, and near-infrared spectral analysis. However, these technologies generally suffer from poor adaptability and low recognition accuracy under complex working conditions. For example, in high-dust environments, traditional visual recognition systems typically have a false judgment rate exceeding 40% due to image blurring and reduced contrast; laser ranging systems can have measurement errors exceeding 15cm due to the scattering effect of dust particles; single RGB or infrared cameras suffer severe image quality degradation under high-temperature radiation and humidity interference, making it difficult to extract effective features; furthermore, water films formed on material surfaces due to condensation significantly alter their optical properties, leading to distortion in near-infrared spectral analysis results and a decrease in recognition accuracy of over 60%.

[0004] In specific industrial applications, the steel industry has extremely high requirements for the accuracy of identifying the FeO content (45%-55%) in sintered ore, while hazardous waste treatment lines need to accurately determine the mixing ratio of metal slag and combustibles (error ≤3%). Existing technologies often fail to meet these key indicators. In addition, traditional identification systems lack a dynamic response mechanism to environmental changes and cannot adaptively adjust to sudden increases in dust, temperature and humidity fluctuations, etc., resulting in unstable identification performance and high system maintenance frequency. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the existing defects and provide a material identification method in high-temperature, dusty, and condensing environments. This method integrates thermal infrared, millimeter-wave radar, and hyperspectral data to break through the identification bottleneck of a single mode in extreme environments. It solves the problem of difficult material identification in extreme environments such as high temperature, high dust, and high condensation, and has outstanding prospects for industry promotion. It can effectively solve the problems in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for material identification in a high-temperature, dusty, and condensing environment, comprising the following steps:

[0007] S1. Multimodal data synchronous acquisition: Data is acquired using a three-light fusion sensor, with the three sensors working together: a thermal infrared sensor, a millimeter-wave radar sensor, and a hyperspectral sensor.

[0008] S2. Real-time monitoring of environmental parameters: Environmental data is collected through dust sensors and temperature and humidity meters. When T>80℃ and RH>90%, the anti-interference mode of the three-light fusion sensor is activated.

[0009] S3. Dynamic Data Fusion: An adaptive weighted fusion algorithm is adopted, and the fusion function is expressed as follows:

[0010] Fusion result = w1 × infrared feature + w2 × radar feature + w3 × spectral feature;

[0011] Among them, weight = (1 / ), For each sensor, the signal-to-noise ratio is used.

[0012] Dynamic weight calculation:

[0013] , where σ is the noise variance and A is the signal amplitude;

[0014] Uncertainty in fusion results: ;

[0015] S4. Condensation interference elimination: Polarization hyperspectral processing, extracting 0°, 45°, and 90° polarization images, calculating Stokes vectors to separate specular reflection and diffuse reflection, and reconstructing spectral curves without water film interference;

[0016] Stokes vector calculation:

[0017]

[0018] Reflectance reconstruction after defogging:

[0019] ;

[0020] S5. Material Matching and Recognition: Input the processed features into the pre-trained model:

[0021] Primary classification: Material categories are determined based on ResNet50;

[0022] Secondary identification: Specific components are determined by matching the material feature database using SVM;

[0023] The recognition accuracy function is:

[0024] ;

[0025] in, = ;

[0026] S6. Self-cleaning trigger: Optical window contamination monitoring. If the light transmittance drops by more than 10% for 3 consecutive frames, a 0.3s pulse backflushing will be initiated. If the humidity is >95%RH for 5 minutes, the electric heating decondensation will be activated.

[0027] Preferably, in step S1, the thermal infrared sensor acquires the material temperature distribution with a resolution of 0.5℃; the millimeter-wave radar sensor generates a three-dimensional point cloud of the surface with an accuracy of ±2mm; and the hyperspectral sensor collects the reflectance in the 900-1700nm band with a 5nm interval.

[0028] Preferably, in step S2, when the anti-interference mode of the three-light fusion sensor is activated, the sensor sampling rate enhancement function is expressed as follows:

[0029] .

[0030] Preferably, in step S3, the sensor weights are adjusted in real time based on the prediction of dust concentration change trend using an LSTM network, and the infrared weight w1 is increased to 0.7 when dust suddenly increases.

[0031] Preferably, in step S5, the material characteristic database includes spectral fingerprints of more than 200 industrial materials.

[0032] Preferably, in step S6, the optical window is provided with a 0.5MPa pulsed airflow nozzle and a 50W semiconductor heating film.

[0033] Compared with existing technologies, the beneficial effects of this invention are: it integrates thermal infrared, millimeter-wave radar, and hyperspectral data to construct a redundant sensing system, breaking through the identification bottleneck of a single mode in extreme environments; the three sensors work together: thermal infrared provides strong dust resistance and acquires temperature distribution; millimeter-wave radar penetrates dust and acquires the three-dimensional surface structure; and hyperspectral imaging provides material composition characteristics, combined with polarization technology to solve condensation interference; the softmax function dynamically adjusts sensor weights according to the signal-to-noise ratio, improving the reliability of the fusion results; combining polarization imaging with hyperspectral imaging for industrial material identification, and using Stokes vector separation to separate specular reflection and diffuse reflection, solves the spectral distortion problem caused by condensation on the material surface, demonstrating significant technological breakthroughs and solving the problem of difficult material identification in extreme environments such as high temperature, high dust, and high condensation, with outstanding prospects for industry promotion. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0035] The present invention can be explained in detail through the following embodiments. The purpose of disclosing the present invention is to protect all technical improvements within the scope of the present invention. In the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "front", "rear", "left", "right" indicating the orientation or positional relationship, they are only corresponding to the drawings of this application for the convenience of describing the present invention, and are not intended to indicate or imply that the device or element referred to must have a specific orientation.

[0036] Please see Figure 1 This invention provides a technical solution: a method for material identification in a high-temperature, dusty, and condensing environment, comprising the following steps:

[0037] S1. Multimodal data synchronous acquisition: Data is acquired using a three-light fusion sensor, with the three sensors working together: a thermal infrared sensor, a millimeter-wave radar sensor, and a hyperspectral sensor.

[0038] Specifically, the thermal infrared sensor acquires the material temperature distribution with a resolution of 0.5℃; the millimeter-wave radar sensor generates a three-dimensional point cloud of the surface with an accuracy of ±2mm; and the hyperspectral sensor collects reflectivity in the 900-1700nm band with a 5nm interval.

[0039] S2. Real-time monitoring of environmental parameters: Environmental data is collected through dust sensors and temperature and humidity meters. When T>80℃ and RH>90%, the anti-interference mode of the three-light fusion sensor is activated.

[0040] The sensor sampling rate has been increased to: .

[0041] S3. Dynamic Data Fusion: An adaptive weighted fusion algorithm is adopted, and the fusion function is expressed as follows:

[0042] Fusion result = w1 × infrared feature + w2 × radar feature + w3 × spectral feature;

[0043] Among them, weight = (1 / ), For each sensor, the signal-to-noise ratio is used.

[0044] Dynamic weight calculation:

[0045] , where σ is the noise variance and A is the signal amplitude;

[0046] Uncertainty in fusion results: ;

[0047] Based on the prediction of dust concentration change trend using LSTM network, the sensor weights are adjusted in real time. When dust suddenly increases, the infrared weight w1 is increased to 0.7.

[0048] By integrating data from three bands—thermal infrared (8-14μm), millimeter-wave radar (77GHz), and hyperspectral (900-1700nm)—a redundant sensing system is constructed.

[0049] S4. Condensation interference elimination: Polarization hyperspectral processing, extracting 0°, 45°, and 90° polarization images, calculating Stokes vectors to separate specular reflection and diffuse reflection, and reconstructing spectral curves without water film interference;

[0050] Stokes vector calculation:

[0051]

[0052] Reflectance reconstruction after defogging:

[0053] ;

[0054] By separating the surface water film from the reflected light from the bulk using polarization hyperspectral imaging, the characteristic spectral lines of the real substance can be restored.

[0055] S5. Material Matching and Recognition: Input the processed features into the pre-trained model:

[0056] Primary classification: Based on ResNet50, determine the major category of materials, such as metals or non-metals;

[0057] Secondary identification: The specific components are determined by matching the material feature database with SVM. The material feature database includes spectral fingerprints of more than 200 industrial materials.

[0058] The recognition accuracy function is:

[0059] ;

[0060] in, = ;

[0061] Propose an accuracy function, when >90%, conduct pollution testing, such as If the data acquisition rate is less than 90%, the three-light fusion sensor will re-acquire data.

[0062] S6. Self-cleaning trigger: Optical window contamination monitoring. If the light transmittance drops by more than 10% for 3 consecutive frames, a 0.3s pulse backflushing will be initiated; if the humidity is >95%RH for 5 minutes, the electric heating decondensation will be activated.

[0063] The optical window is equipped with a 0.5MPa pulsed airflow nozzle and a 50W semiconductor heating film. It is designed with a pulse backflushing + electrothermal decondensation composite system to maintain the light transmittance of the optical window ≥85%.

[0064] Example 1: Identification of FeO content in sintered ore of steel plants under high temperature and high dust conditions;

[0065] 1. Implementation scenario:

[0066] This embodiment is applied to a sinter conveyor belt line in a steel plant. The ambient temperature is 85℃, the air humidity is 92%RH, and the dust concentration reaches 60mg / m³. The system is required to identify in real time whether the FeO content in the sinter is within the range of 45%-55% during operation, with an identification error not exceeding ±1.5%.

[0067] 2. System Deployment:

[0068] Installation location: Install 3 meters above the belt, away from direct exposure to high-temperature heat sources;

[0069] Hardware configuration: Multimodal sensor: integrating thermal infrared (8-14μm), millimeter-wave radar (77GHz), and hyperspectral (900-1700nm);

[0070] Self-cleaning unit: 0.5MPa pulse airflow nozzle + 50W heating film;

[0071] Edge computing device: NVIDIA Jetson AGX Orin, runtime latency <50ms;

[0072] Adaptive fusion algorithm: dynamically adjusts weights based on the softmax function;

[0073] Classification and recognition model: ResNet50 + SVM spectral matching model;

[0074] 3. Operation process:

[0075] Environmental monitoring: After the system is started, it collects ambient temperature, humidity and dust concentration data in real time;

[0076] Level 3 anti-interference mode activated: When T>80℃ and RH>90% are detected, the sampling rate is automatically increased to 10Hz and the polarization hyperspectral processing module is enabled;

[0077] Multimodal data acquisition:

[0078] Thermal infrared: Acquires the surface temperature distribution of materials (resolution 0.5℃);

[0079] Millimeter-wave radar: generates 3D point clouds of surfaces (accuracy ±2mm);

[0080] Hyperspectral: Reflectance was collected in the 900-1700nm band (5nm intervals);

[0081] Data fusion and processing: Dynamic weighted fusion, with infrared weight of 0.4, radar weight of 0.3, and spectral weight of 0.3. When dust increases suddenly, the infrared weight increases to 0.7.

[0082] Polarization hyperspectral processing: separating the water film and bulk reflection, and reconstructing the spectral curve;

[0083] Recognition output: Primary classification, ResNet50 identifies it as "metallic"; Secondary recognition, SVM matches the spectral library, outputting FeO content as 51.2%;

[0084] Self-cleaning triggered:

[0085] If the transmittance drops by more than 10% for three consecutive frames, pulse backflush will be activated.

[0086] If the humidity is >95%RH for 5 minutes, turn on the electric heating decondenser (maintain at 50℃).

[0087] 4. Experimental Results and Comparative Analysis:

[0088] 5. Conclusion:

[0089] This embodiment verifies the high-precision identification capability of the present invention for FeO content in sintered ore under high temperature, high humidity and high dust environment. The identification error is significantly lower than that of traditional methods. The system has fast response and strong stability, meeting the requirements of continuous operation in industrial sites.

[0090] Example 2: Identification of the mixing ratio of metal slag and combustibles in a hazardous waste treatment line;

[0091] 1. Implementation scenario:

[0092] A hazardous waste incineration line requires component identification of mixed feed materials, distinguishing the mixing ratio of metallic slag (such as copper slag and steel slag) and combustibles (such as plastics and rubber), with an identification error of ≤3%. The ambient temperature is 70℃, humidity is 95%RH, and dust concentration is approximately 40mg / m³.

[0093] 2. System Configuration:

[0094] Sensor: Enable polarization hyperspectral module;

[0095] Model update: The mixed spectral characteristics of metal slag and combustibles are loaded into the SVM model;

[0096] Self-cleaning strategy: When humidity is >95%, the electrothermal film continues to operate to keep the optical window dry;

[0097] 3. Recognition process:

[0098] Real-time acquisition of multimodal data of the mixture;

[0099] Polarization hyperspectral processing: Separating water film interference and reconstructing the true spectrum;

[0100] ResNet50 identifies major material categories (metals / non-metals);

[0101] SVM model matching mixing ratio (e.g., 60% metal slag + 40% combustibles);

[0102] The output results are then uploaded to the DCS system for combustion control adjustment.

[0103] 4. Result Verification:

[0104] By comparing with the laboratory XRF test results, the mixing ratio error identified by the system is ±2.1%, which meets the industry requirement of error ≤3%.

[0105] Supplementary technical verification data:

[0106] In summary, through the detailed description and experimental verification of the two typical embodiments above, the present invention achieves high-precision and high-stability identification of material composition in complex industrial environments such as high temperature, high dust, and high condensation, and has good engineering feasibility and industry promotion value.

[0107] The parts of this invention not described in detail are prior art. It will be apparent to those skilled in the art that this invention is not limited to the details of the above exemplary embodiments, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects, and are intended to encompass all changes that fall within the meaning and scope of equivalents in the content of this invention.

Claims

1. A method for material identification in a high-temperature, dusty, and condensing environment, characterized in that: Includes the following steps: S1. Multimodal data synchronous acquisition: Data is acquired using a three-light fusion sensor, with the three sensors working together: a thermal infrared sensor, a millimeter-wave radar sensor, and a hyperspectral sensor. S2. Real-time monitoring of environmental parameters: Environmental data is collected through dust sensors and temperature and humidity meters. When T>80℃ and RH>90%, the anti-interference mode of the three-light fusion sensor is activated. S3. Dynamic Data Fusion: An adaptive weighted fusion algorithm is adopted, and the fusion function is expressed as follows: Fusion result = w1 × infrared feature + w2 × radar feature + w3 × spectral feature; Among them, weight = (1 / ), For each sensor, the signal-to-noise ratio is used. Dynamic weight calculation: , where σ is the noise variance and A is the signal amplitude; Uncertainty in fusion results: ; S4. Condensation interference elimination: Polarization hyperspectral processing, extracting 0°, 45°, and 90° polarization images, calculating Stokes vectors to separate specular reflection and diffuse reflection, and reconstructing spectral curves without water film interference; Stokes vector calculation: ； Reflectance reconstruction after defogging: ； S5. Material Matching and Recognition: Input the processed features into the pre-trained model: Primary classification: Material categories are determined based on ResNet50; Secondary identification: Specific components are determined by matching the material feature database using SVM; The recognition accuracy function is: ； in, = ; S6. Self-cleaning trigger: Optical window contamination monitoring. If the light transmittance drops by more than 10% for 3 consecutive frames, a 0.3s pulse backflushing will be initiated. If the humidity is >95%RH for 5 minutes, the electric heating decondensation will be activated.

2. The material identification method in a high-temperature, dusty, and condensing environment according to claim 1, characterized in that: In step S1, a thermal infrared sensor acquires the material temperature distribution with a resolution of 0.5℃; a millimeter-wave radar sensor generates a three-dimensional point cloud of the surface with an accuracy of ±2mm; and a hyperspectral sensor collects the reflectance in the 900-1700nm band with a 5nm interval.

3. The material identification method in a high-temperature, dusty, and condensing environment according to claim 1, characterized in that: In step S2, when the anti-interference mode of the three-light fusion sensor is activated, the sensor sampling rate enhancement function is expressed as follows: 。 4. The material identification method in a high-temperature, dusty, and condensing environment according to claim 1, characterized in that: In step S3, the sensor weights are adjusted in real time based on the prediction of dust concentration change trend using an LSTM network. When dust suddenly increases, the infrared weight w1 is increased to 0.

7.

5. The material identification method in a high-temperature, dusty, and condensing environment according to claim 1, characterized in that: In step S5, the material characteristic database includes spectral fingerprints of more than 200 industrial materials.

6. The material identification method in a high-temperature, dusty, and condensing environment according to claim 1, characterized in that: In step S6, the optical window is equipped with a 0.5MPa pulsed airflow nozzle and a 50W semiconductor heating film.

Images