Cascaded spectral imaging microsystem, spectrum acquisition method and fireproof coating identification method

Abstract

The invention provides a cascaded spectral imaging microsystem, a spectrum acquisition method and a fireproof coating identification method. The system comprises an imaging lens, a spectral imaging module and a processing circuit, the spectral imaging module comprises at least two cascaded spectral imaging sensors, the plurality of spectral imaging sensors respectively cover different free spectral ranges, and each spectral imaging sensor comprises unit group filtering structures which are periodically arranged on a photosensitive surface of the sensor; the unit group filtering structure consists of a plurality of pixel-level interference film filtering structures which are arranged in an array form; each spectral imaging sensor is provided with an imaging lens, and the plurality of spectral imaging sensors are arranged in parallel to synchronously acquire a target image; and the processing circuit receives the output data of the reading circuit for data processing. The method synchronously performs wide-range free spectral imaging on the target, reduces the measurement times, improves the recognition precision, reduces the instrument cost, meets the requirements of different application scenes, performs real-time information processing, and can be used for quickly identifying the target type.

Description

Technical Field

[0001] The present invention belongs to the technical field of spectrum recognition, and in particular relates to a cascaded spectrum imaging microsystem, a spectrum acquisition method, and a fire retardant coating identification method. Background Art

[0002] Existing spectral imaging systems have limited coverage of the free spectral range due to their own structural limitations. For example, spectral imaging sensors based on interference filtering are affected by the interference filter film material and structure, and the free spectral range of the filter structure is limited. The design of a single-chip spectral imaging sensor with a wide free spectral range formed by multiple interference filter structures or multiple materials cascaded is limited by the semiconductor process capabilities, and its implementation faces great process challenges. In addition, spectral instruments based on traditional grating spectrometry methods also have spectral range limitations, and this type of spectral instrument performs single-point spectral detection. In order to avoid single measurement errors, multiple measurements of the same sample are required, which increases the complexity of the measurement process. For the above reasons, the application of spectral imaging systems in the field of target recognition has many limitations, such as the rapid identification of fire retardant coatings for steel structures.

[0003] Steel structures have excellent architectural properties and are widely used in various large complexes. However, steel structures have poor fire resistance. When the temperature rises, the mechanical properties drop sharply, and the bearing capacity of such buildings also decreases. If they encounter high-temperature disasters such as fires, there will be a huge risk of collapse. Therefore, special fire retardant coatings are needed for protection. With the development of society, the demand for steel structures has increased, driving the demand for steel structure fire retardant coatings to increase day by day. However, the quality of steel structure fire retardant coatings in the fire retardant coatings market varies greatly, which brings hidden dangers to safety use, and requires fire supervision and inspection of their performance.

[0004] The appearance of ordinary coatings is no different from that of special coatings, but they do not contain flame retardant components or the proportion of flame retardant components cannot meet the standard requirements. Steel structure fire retardant coatings are special coatings. According to the flame retardant mechanism, fire retardant coatings can be divided into two categories: intumescent fire retardant coatings and non-intumescent fire retardant coatings. Non-intumescent fire retardant coatings are also called thick steel structure fire retardant coatings. The main component is inorganic thermal insulation material. The surface is granular and has a low density. The coating does not expand when heated. Because it has good thermal insulation, it is also called thermal insulation fire retardant coating. Its fire protection mechanism is to use the inherent good thermal insulation of the coating to prevent the heat of the fire from transferring to the steel, and form a dense glaze-like substance at high temperature, which effectively isolates oxygen and reflects heat, delaying the temperature rise of the steel structure and playing a role in fire protection. Intumescent fire retardant coatings are divided into thin and ultra-thin steel structure fire retardant coatings, which are mainly composed of base material, dehydrating agent, carbon forming agent and foaming agent. When the temperature rises to a certain level, the dehydrating agent promotes the dehydration and carbonization of polyhydroxy compounds. Under the action of a large amount of gas released by the decomposition of the foaming agent, the coating expands, and the expansion multiple can reach more than ten times or even dozens of times, forming a dense foam-like carbonized insulation layer, thereby preventing heat from being transferred to the steel structure and playing a fire protection role.

[0005] At present, the test method for the fire retardant performance of steel structure fire retardant coatings adopts the method specified in GB14907-2002 "Steel Structure Fire Retardant Coatings". According to the design load specified in GBJ17, the specified I-beam is loaded and the combustion experiment is carried out on a horizontal furnace. After the coating is applied, dried and cured, and tested on a specific experimental device or test furnace, the detection cycle is long, the process is complicated, and the test cost is high. Only a few professional institutions can implement it. In addition, the fire resistance limit given in the test report is the test result within a certain period of time after the coating is maintained, which has a certain gap with the actual use status of the fire retardant coating. The on-site inspection of project acceptance only tests physical indicators such as the thickness of the fire retardant coating. It is generally impossible to achieve on-site inspection of fire resistance, which has great limitations. Therefore, some fire retardant coating detection methods outside the standard regulations continue to develop, including cone calorimeter method, thermal analysis method, photoelectron spectroscopy method, scanning electron microscopy method, infrared spectroscopy method, X-ray diffraction analysis method, etc., which provide a theoretical basis for the study of the combustion mechanism of fire retardant coatings, the formulation and system of fire retardant coatings, but have not yet solved the problem of on-site inspection needs for project acceptance.

[0006] The release of the GA588-2012 "Rules for On-site Inspection and Judgment of Fire Products" shows that the country has increasingly higher requirements for on-site inspection of fire products. However, due to problems in equipment, technology, personnel and operation, there is currently no good technical means to quickly detect the on-site conditions of fire retardant coatings. Therefore, it is very necessary to carry out relevant research and experiments on the on-site rapid inspection of steel structure fire retardant coatings, to solve the bottleneck problem of difficult and inefficient quality inspection of fire retardant coatings at the project acceptance site during fire supervision and inspection, to explore the on-site rapid inspection method of fire resistance performance parameters of steel structure fire retardant coatings, and to use scientific data to assist fire supervision and inspection professionals in making comprehensive judgments on the eligibility of fire retardant coatings.

[0007] Spectral analysis is a non-contact photoelectric measurement method. Different substances have different "spectral fingerprints". Through non-contact detection of samples, the composition of samples can be quickly analyzed without destroying them. By changing the materials and relative proportions of each component of the sample, different spectra can be obtained. It is widely used in food safety, smart medical care, smart agriculture, fire rescue and other fields. Combined with intelligent learning algorithms, intelligent classification models are trained by learning the spectral characteristics of different targets, and intelligent classification models are used to quickly discriminate and analyze unknown samples. With the development of spectral imaging technology, traditional spectral imaging systems are difficult to meet the needs of increasingly complex detection scenarios. The development of spectral imaging microsystems has laid the foundation for the development of multi-field, multi-platform, lightweight spectral image intelligent detection technology. In particular, multi-spectral imaging sensors based on monolithic integration use interference filter thin film structures for spectroscopy. Relying on semiconductor processes, the spectral thin film structure is directly integrated on the CMOS image sensor to form an imaging chip with spectral selection function, which constitutes a spectral imaging sensor. Spectral imaging sensors are integrated into spectral imaging modules or spectral imaging microsystems of different forms according to the detection needs of different application fields, promoting the development of spectral imaging technology.

[0008] For the rapid detection of fire retardant coatings for steel structures, Sichuan Fire Research Institute has developed a "portable fire retardant coating test device" based on the thermal expansion characteristics of intumescent fire retardant coatings and the requirements of GA588 "Rules for on-site inspection and judgment of fire protection products". The device adopts the heating conditions in GB9978-88 "Test Methods for Fire Resistance of Building Components" to determine whether the fire retardant performance meets the standards by measuring the expansion state and expansion multiple of the intumescent steel structure fire retardant coating sample in the heating cup within a certain period of time. The device has an expansion resolution of about 1μm, an expansion value measurement range of 50mm, and can complete the detection in 15 minutes. First, the portable fire retardant coating test device is mainly used to detect intumescent fire retardant coatings, and it is not possible to perform rapid detection for non-intumescent fire retardant coatings. Secondly, the device mainly relies on the change in the state of the intumescent fire retardant coating after heating within a certain period of time to judge its performance. It is necessary to give the sample a certain amount of time for the development process of heat-expansion, which fails to meet the real-time detection requirements. Finally, this type of device requires taking uncoated coating samples for detection, and cannot detect samples that have been coated at the construction site.

[0009] Tianjin Fire Research Institute uses short-wave infrared spectra in the range of 1000-2500nm to carry out research on rapid identification technology of different brands of fire retardant coatings based on spectral analysis. The target near-infrared diffuse reflectance spectrum is collected by the short-wave infrared spectroscopy equipment of the optical fiber probe. After the standard normal variable transformation and the optimization processing of the second-order derivative of Norris, the Mahalanobis distance discriminant analysis model of different brands is established. By comparing the validation sets of different brands with the Mahalanobis distance discriminant analysis model, the average Mahalanobis distance and the Mahalanobis distance critical value of each brand are calculated. By collecting the short-wave infrared spectrum of unknown samples and comparing them with the established model of each brand, the calculated minimum Mahalanobis distance value of the unknown sample is compared with the brand's Mahalanobis distance critical value to determine the brand of the unknown sample. The technical effect is accurate results, simple operation, rapid analysis, low analysis cost and no pollution to the environment. It is suitable for on-site rapid non-destructive detection and identification of the quality and brand of steel structure fire retardant coatings. This method uses a short-wave infrared spectrometer based on the traditional grating spectrometry method for single-point spectral detection, covering a spectral range of 1000-2500nm. In order to avoid single measurement errors, the same sample needs to be measured multiple times, which increases the complexity of the measurement process. At the same time, the price of short-wave infrared instruments and equipment is relatively high due to the influence of detectors, and the commercial promotion of the system cannot be achieved for the time being. Summary of the invention

[0010] The present invention aims to solve one of the above-mentioned technical problems and provides a cascaded spectral imaging microsystem, a spectral acquisition method, and a fire retardant coating identification method. The cascaded spectral imaging microsystem adopts cascaded spectral imaging sensors to synchronously perform wide-range free spectral imaging of the target, thereby realizing visible-near-infrared wide-range spectral imaging, reducing the number of measurements, improving recognition accuracy, reducing instrument costs, meeting the needs of different application scenarios, and processing information in real time, which can be used to quickly identify target types.

[0011] The technical solution adopted by the present invention to solve the above technical problems is as follows:

[0012] The present invention provides a cascaded spectral imaging microsystem, comprising an imaging lens, a spectral imaging module, and a processing circuit; the spectral imaging module comprises at least two cascaded spectral imaging sensors and corresponding readout circuits thereof, a plurality of the spectral imaging sensors respectively cover different free spectrum ranges, each of the spectral imaging sensors comprises a unit group filter structure periodically arranged on a photosensitive surface of the sensor, the unit group filter structure comprises a plurality of pixel-level interference thin film filter structures arranged in an array form; an imaging lens is respectively arranged in front of each of the spectral imaging sensors, and a plurality of spectral imaging sensors are arranged in parallel to synchronously collect target images; the processing circuit receives output data of all the readout circuits and performs data processing.

[0013] Furthermore, the processing circuit comprises:

[0014] AI module, used to integrate the target identification algorithm, process and match the collected raw data, and automatically identify the target as one of the preset categories;

[0015] An interactive module, used to output the discrimination results;

[0016] A storage module, used for storing data;

[0017] Communication module, used to communicate with various terminals to implement command issuance and result display;

[0018] The cascaded spectral imaging microsystem also includes a power supply module, which is used to supply power to the cascaded spectral imaging microsystem.

[0019] Furthermore, the spectral imaging module includes two cascaded spectral imaging sensors, the unit group filter structure of the first spectral imaging sensor includes 3×3 interference thin film filter structures, covering the range of 400-600nm, and the unit group filter structure of the second spectral imaging sensor includes 4×4 interference thin film filter structures, covering the range of 600-1000nm. The cascaded spectral imaging microsystem is used to collect spectral image information of steel structure fire retardant coatings and ordinary coatings.

[0020] The present invention also provides a spectrum acquisition method of a cascaded spectrum imaging microsystem, using the cascaded spectrum imaging microsystem, the spectrum acquisition method comprises the following steps:

[0021] S1, adaptive parameter adjustment of multiple spectral imaging sensors;

[0022] S2, multiple spectral imaging sensors collect target raw data, obtain two-dimensional data through calibration, and obtain three-dimensional spectral data through downsampling;

[0023] S3, three-dimensional spectral data of multiple spectral imaging sensors are matched in real time through images to determine the transformation matrix;

[0024] S4. Extract a first target spectral range from the three-dimensional spectral data of a spectral imaging sensor through image processing, calculate the average spectrum of each spectral segment of the first target spectral range, calculate the coordinate area of ​​the first target spectral range in the three-dimensional spectral data of other spectral imaging sensors through a transformation matrix, and then calculate the average spectrum of each spectral segment in the coordinate area of ​​other spectral imaging sensors; combine the average spectra of all spectral imaging sensors to obtain a joint spectrum.

[0025] Furthermore, the method for adaptively adjusting parameters of the plurality of spectral imaging sensors includes:

[0026] S101, selecting a light source, and normalizing the light source spectrum to obtain a standardized light source spectrum;

[0027] S102, sorting the plurality of spectral imaging sensors from small to large according to the free spectral range values, taking the smallest spectral imaging sensor as a reference sensor, and determining an exposure time of the reference sensor;

[0028] S103, starting from the reference sensor, select two adjacent spectral imaging sensors in order, and adjust the exposure time of the next spectral imaging sensor according to the exposure time of the previous spectral imaging sensor, the method is as follows:

[0029]

[0030] Among them, F source (λ) is the light source response that varies with the central wavelength λ; is the maximum central wavelength of each spectral band of the previous spectral imaging sensor, is the minimum central wavelength of each spectral band of the latter spectral imaging sensor, is the previous spectral imaging sensor at exposure time τ The response of the corresponding spectrum band, The exposure time τ of the spectral imaging sensor is 2 Down The response of the corresponding spectrum band, The maximum central wavelength of the previous spectral imaging sensor The quantum efficiency of is the minimum central wavelength of the latter spectral imaging sensor Quantum efficiency of the spectrum, ∑ λ represents the sum over all λ.

[0031] Furthermore, the method for determining the exposure time of the reference sensor in step S102 is as follows:

[0032] Calculate the quantum efficiency of each spectral band of the reference sensor based on the standardized light source spectrum, and determine the spectral band with the highest and lowest quantum efficiency;

[0033] Set the gain of the reference sensor to 0, initialize the exposure time, collect the initialization image of the standard white colorimetric plate, and calculate the maximum grayscale value of the spectrum with the highest quantum efficiency and the minimum grayscale value of the spectrum with the lowest quantum efficiency in the initialization image;

[0034] Adjust the exposure time according to the maximum and minimum grayscale values ​​until the maximum grayscale value of the spectrum with the highest quantum efficiency is in the range of 180-255, and record the data of the standard white colorimetric plate and the exposure time of the reference sensor.

[0035] Furthermore, the method for adjusting the exposure time of the reference sensor according to the maximum and minimum grayscale values ​​is as follows:

[0036]

[0037] Among them, Max 0 is the maximum grayscale of the spectrum with the highest quantum efficiency, Min 0 is the minimum grayscale value of the spectrum with the lowest quantum efficiency, τ is the exposure time of the reference sensor, and τ 0 Initialize exposure time

[0038] Furthermore, the calibration method of the target raw data of each spectral imaging sensor is:

[0039] S201, fixing other parameters of the spectral imaging sensor, obtaining the spectral imaging sensor in an adjustable exposure time range τ min -τ max Internal dark noise variation curve;

[0040] S202, using a high-order polynomial function to fit the dark noise variation curve, obtain the functional relationship F(τ) between the average dark noise and the exposure time, and use F(τ) to calibrate the target raw data collected by the spectral imaging sensor:

[0041]

[0042] Among them, r wis the reflectance of the standard white colorimetric plate for each spectral band; I is the original data of the target scene; I τ is the standard white colorimetric plate data, I hw The identity matrix representing the size of the photosensitive surface of the spectral imaging sensor.

[0043] Furthermore, the first target spectral range is extracted by image processing from the three-dimensional spectral data of a spectral imaging sensor, and the average spectrum of each spectral segment of the first target spectral range is calculated, which specifically includes the following steps:

[0044] The first target spectral range in the three-dimensional spectral data is I 2d (x 1 ,y 1 ,i)~I 2d (x 2 ,y 2 ,i), in the two-dimensional coordinate (x 1 ,y 1 )(x 2 ,y 2 ) range, and calculate the average values ​​of the multiple pixels in each channel of the spectral imaging sensor to obtain the average spectrum of the target collected by the spectral imaging sensor.

[0045] The present invention further provides a method for identifying steel structure fire retardant coatings based on spectral imaging, which uses the cascaded spectral imaging microsystem to collect target raw data, uses the spectral acquisition method to obtain a joint spectrum, and also includes the following steps:

[0046] Collect the joint spectra of known common coatings and fire-retardant coatings, train and test the classifier, and optimize the classifier parameters according to the test accuracy;

[0047] Collect the joint spectrum of unknown paints and use the determined classifier to quickly identify the unknown paints.

[0048] The beneficial effects of the present invention compared with the prior art are as follows:

[0049] In view of the problems that existing short-wave infrared spectrometers need to perform multiple measurements for target spectral imaging and the high instrument cost, the present invention proposes a rapid detection method based on visible-near infrared range spectral imaging technology. A cascaded spectral imaging microsystem is designed, which synchronously performs wide-range spectral imaging of the target through multiple cascaded spectral imaging sensors. The spectral imaging sensor adopts semiconductor interference thin film spectroscopy technology, which overcomes the problem that the free spectral range of existing general-purpose spectral imaging sensors is difficult to expand due to the limitations of materials and semiconductor process capabilities. By cascading spectral imaging microsystems covering different free spectral ranges, the spectral detection range is expanded, and the fusion imaging technology reduces the number of measurements. Multi-spectral information can be obtained in one measurement, which improves recognition accuracy and reduces instrument costs. At the same time, it meets the needs of different application scenarios, and information is processed in real time, which can be used to quickly identify target types.

[0050] The present invention proposes a method for adaptive parameter adjustment of a cascaded spectral imaging microsystem and a method for calibrating raw data. According to the inherent characteristics of the spectral imaging sensor structure and the imaging law, the optimal initialization parameters of another cascaded sensor are automatically matched through adaptive parameter adjustment of a single sensor, so that the spectral data collected by different sensors are optimally matched and coupled. At the same time, a high-order polynomial function is introduced to simulate the variation law of dark noise, and the dark noise and its function are stored as prior data in the identification system storage module, which is automatically called according to each parameter value to realize automatic data calibration.

[0051] The present invention proposes a rapid identification system for fire retardant coatings for steel structures based on a cascade of integrated universal spectral imaging sensors. Under the condition of reducing the process requirements, the existing universal spectral imaging sensors covering different free spectral ranges within the range of 400-1000nm are cascaded, and each spectral imaging sensor is spectrally modulated by an interference film based on pixel-level narrow-band filtering to form a spectral imaging microsystem that can cover a wide spectral range of 400-1000nm, so as to quickly detect the relevant coatings of the target steel structure. This structure reduces the system's requirements for the design and process of the interference-level spectral film of the sensor core spectroscopy, and the design requirements for the free spectral range are reduced.

[0052] The present invention proposes a method for quickly identifying steel structure fire retardant coatings based on spectral imaging technology in the visible-near infrared spectral range of 400-1000nm. For typical expansive steel structure fire retardant coating samples, non-expandable steel structure fire retardant coating samples and some common coating samples with similar appearance, non-contact spectral image data collection of the samples can be performed, and a classification algorithm is designed to quickly identify the coatings based on the consistency characteristics of different fire retardant mechanism coatings in the visible-near infrared spectral range and the spectral differences between each other. The method has no requirements for the different physical states of the target, such as solid and liquid states, that is, the intelligent recognition method based on spectral imaging technology is equally effective for uncoated coatings that maintain their original physical state and coated and cured samples that have not been toned. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The included drawings are used to provide a further understanding of the embodiments of the present invention, which constitute a part of the specification, are used to illustrate the embodiments of the present invention, and together with the text description, explain the principles of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 A schematic diagram of the structure of a cascaded spectral imaging microsystem provided in a specific embodiment of the present invention;

[0055] Figure 2 A flow chart of a method for rapid identification of fire retardant coatings for steel structures based on spectral imaging provided in a specific embodiment of the present invention;

[0056] Figure 3 A schematic diagram of a dark noise fitting curve of a spectral imaging sensor provided in a specific embodiment of the present invention;

[0057] Figure 4 A schematic diagram of normalized spectrum fitting of natural light source provided by a specific embodiment of the present invention;

[0058] Figure 5 A schematic diagram of spectrum curves of different paint samples provided in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0059] Specific embodiments of the present invention are described in detail below. In the following description, for the purpose of explanation and not limitation, specific details are set forth to help fully understand the present invention. However, it will be apparent to those skilled in the art that the present invention may also be practiced in other embodiments that depart from these specific details.

[0060] It should be noted that in order to avoid obscuring the present invention due to unnecessary details, only the device structure and / or processing steps closely related to the scheme of the present invention are shown in the drawings, while other details that are not closely related to the present invention are omitted.

[0061] As one aspect of the present invention, a cascaded spectral imaging microsystem is provided, including an imaging lens, a spectral imaging module, and a processing circuit; the spectral imaging module includes at least two cascaded spectral imaging sensors and their corresponding readout circuits, multiple spectral imaging sensors respectively cover different free spectrum ranges, and multiple spectral imaging sensors cover the visible-near infrared range in combination; the imaging lens is correspondingly arranged in front of the spectral imaging sensor, and the processing circuit receives the output data of all the readout circuits and performs data processing, such as calculating the spectral fusion image and outputting the identification result. Multiple spectral imaging sensors are arranged in parallel so that they are at the same distance from the acquisition target, which is convenient for the subsequent processing of the acquired data. Each spectral imaging sensor includes a unit group filter structure periodically arranged on the photosensitive surface of the image sensor, and the unit group filter structure includes m 1 ×m 2 The present invention adopts multiple cascaded spectral imaging sensors to synchronously collect target spectral images, covering a wide free spectrum range, meeting the needs of different application scenarios, and processing information in real time, which can be used to quickly identify target types.

[0062] Preferably, the cascaded spectral imaging microsystem further includes a power supply module for supplying power to the cascaded spectral imaging microsystem.

[0063] Preferably, the cascaded spectral imaging microsystem processing circuit includes an AI module, an interaction module, a storage module, and a communication module: the AI ​​module is used to integrate a target identification algorithm, process and match the collected raw data for consistency, and automatically identify the target as one of the preset categories; the interaction module is used to output the identification result; the storage module is used to store data, including but not limited to prior data, classifier models and parameters, recognition results, original collected data, etc.; the communication module is used to communicate with a handheld terminal, a remote terminal, and a display terminal to realize command issuance and result display.

[0064] Furthermore, in order to realize the rapid identification of steel structure fire retardant coatings and ordinary coatings in the visible-near infrared range of 400nm-1000nm, a portable cascade spectral imaging microsystem for rapid identification of steel structure fire retardant coatings is designed. There are two cascade spectral imaging sensors. The unit group filter structure of the first spectral imaging sensor includes m 11 ×m 12The interference thin film filter structure covers the range of 400-600nm. The unit group filter structure of the second spectral imaging sensor includes m 21 ×m 22 An interference thin film filter structure covering the range of 600-1000nm. 11 、m 12 、m 21 、m 22 is a positive integer, the values ​​can be the same or different. To simplify the calculation, the four variables can take the same value.

[0065] The portable spectral imaging microsystem is used to quickly identify fire retardant coatings for steel structures, and to quickly identify and analyze the consistency of coated or uncoated fire retardant coatings for steel structures in complex engineering acceptance environments, giving consistency judgment results to assist engineering acceptance personnel in making comprehensive judgments on engineering acceptance. The spectral imaging microsystem has a simple structure, small size, portability, and fast detection speed. It does not require sampling, and can intelligently identify a wide range of coatings in the field of view by acquiring target spectral images at the same time, providing a scientific reference for engineering acceptance and construction personnel.

[0066] As another aspect of the present invention, a spectrum acquisition method of a cascaded spectrum imaging microsystem is provided, which uses at least two cascaded spectrum imaging sensors, adaptive parameter adjustment, raw data calibration, and an adaptive image matching method to perform multispectral image fusion in different spectral ranges. Specifically:

[0067] S1, adaptive parameter adjustment of multiple spectral imaging sensors;

[0068] S2, multiple spectral imaging sensors collect target raw data, obtain two-dimensional data through calibration, and obtain three-dimensional spectral data through downsampling;

[0069] S3, three-dimensional spectral data of multiple spectral imaging sensors are matched in real time through images to determine the transformation matrix;

[0070] S4. Extract a first target spectral range from the three-dimensional spectral data of a spectral imaging sensor through image processing, calculate the average spectrum of each spectral segment of the first target spectral range, calculate the coordinate area of ​​the first target spectral range in the three-dimensional spectral data of other spectral imaging sensors through a transformation matrix, and then calculate the average spectrum of each spectral segment in the coordinate area of ​​other spectral imaging sensors; combine the average spectra of all spectral imaging sensors to obtain a joint spectrum.

[0071] Furthermore, the method for adaptively adjusting parameters of multiple spectral imaging sensors in step S1 includes:

[0072] S101, selecting a light source, and normalizing the light source spectrum to obtain a standardized light source spectrum;

[0073] S102, sorting the plurality of spectral imaging sensors from small to large according to the free spectral range values, taking the smallest spectral imaging sensor as a reference sensor, and determining an exposure time of the reference sensor;

[0074] S103, starting from the reference sensor, select two adjacent spectral imaging sensors in order, and adjust the exposure time of the next spectral imaging sensor according to the exposure time of the previous spectral imaging sensor, the method is as follows:

[0075]

[0076] Among them, F source (λ) is the light source response that varies with the central wavelength λ; is the maximum central wavelength of each spectral band of the previous spectral imaging sensor, is the minimum central wavelength of each spectral band of the latter spectral imaging sensor, is the previous spectral imaging sensor at exposure time τ The response of the corresponding spectrum band, The exposure time τ of the spectral imaging sensor is 2 Down The response of the corresponding spectrum band, The maximum central wavelength of the previous spectral imaging sensor The quantum efficiency of is the minimum central wavelength of the latter spectral imaging sensor Quantum efficiency of the spectrum, ∑ λ Represents the sum of all λ. This configuration method is used to adaptively adjust the parameters of multiple spectral imaging sensors. When the exposure time of the reference sensor is determined, the exposure values ​​of subsequent sensors do not need to be manually adjusted. Moreover, after the adaptive parameter adjustment is adopted, the spectral matching conditions of multiple sensors are automatically achieved.

[0077] Furthermore, the method for determining the exposure time of the reference sensor in step S102 is as follows:

[0078] S1021. Calculate the quantum efficiency of each spectrum segment of the reference sensor based on the standardized light source spectrum, and determine the spectrum segment with the highest quantum efficiency and the spectrum segment with the lowest quantum efficiency;

[0079] S1022, setting the gain of the reference sensor to 0, initializing the exposure time, acquiring an initialization image of a standard white colorimetric plate, and calculating the maximum grayscale value of the spectrum with the highest quantum efficiency and the minimum grayscale value of the spectrum with the lowest quantum efficiency in the initialization image;

[0080] S1023, adjusting the exposure time according to the maximum and minimum grayscale values ​​until the maximum grayscale value of the spectrum with the highest quantum efficiency is in the range of 180-255, and recording the data of the standard white colorimetric plate and the exposure time τ of the reference sensor.

[0081] Furthermore, the method for adjusting the exposure time of the reference sensor according to the maximum and minimum grayscale values ​​is:

[0082]

[0083] Among them, Max 0 is the maximum grayscale of the spectrum with the highest quantum efficiency, Min 0 is the minimum grayscale value of the spectrum with the lowest quantum efficiency, τ is the exposure time of the reference sensor, and τ 0 Initialize the exposure time.

[0084] Furthermore, the calibration method of the target raw data of the spectral imaging sensor in step S2 is:

[0085] S201, fixing other parameters of the spectral imaging sensor, obtaining the spectral imaging sensor in an adjustable exposure time range τ min -τ max Internal dark noise variation curve;

[0086] S202, using a high-order polynomial function to fit the dark noise variation curve, obtain the functional relationship F(τ) between the average dark noise and the exposure time, and use F(τ) to calibrate the target raw data collected by the spectral imaging sensor:

[0087]

[0088] Among them, r w is the reflectance of the standard white colorimetric plate for each spectral band; I is the original data of the target scene; I τ is the standard white colorimetric plate data, I hw The identity matrix representing the size of the photosensitive surface of the spectral imaging sensor.

[0089] Further, in step S4, a first target spectral range is extracted from three-dimensional spectral data of a spectral imaging sensor through image processing, and an average spectrum of each spectral segment of the first target spectral range is calculated, which specifically includes the following steps:

[0090] The first target spectral range in the three-dimensional spectral data is I 2d (x 1 ,y 1 ,i)~I 2d (x 2 ,y 2 ,i), in the two-dimensional coordinate (x 1 ,y 1)(x 2 ,y 2 ) range, and calculate the average values ​​of the multiple pixels in each channel of the spectral imaging sensor to obtain the average spectrum of the target collected by the spectral imaging sensor.

[0091] Furthermore, the light source adopts a natural light source or a halogen lamp with a broad spectrum.

[0092] As another aspect of the present invention, a method for rapid identification of steel structure fire retardant coatings based on spectral imaging is provided, which adopts the above-mentioned spectral imaging microsystem, including at least two cascaded spectral imaging sensors, obtains a fused image through spectral acquisition of the cascaded spectral imaging microsystem, and then quickly identifies unknown coatings through a classifier, specifically:

[0093] S1, adaptive parameter adjustment of multiple spectral imaging sensors;

[0094] S2, multiple spectral imaging sensors collect target raw data, obtain two-dimensional data through calibration, and obtain three-dimensional spectral data through downsampling;

[0095] S3, three-dimensional spectral data of multiple spectral imaging sensors are matched in real time through images to determine the transformation matrix;

[0096] S4. Extracting a first target spectral range from three-dimensional spectral data of a spectral imaging sensor through image processing, calculating the average spectrum of each spectral segment of the first target spectral range, calculating the coordinate area of ​​the first target spectral range in the three-dimensional spectral data of other spectral imaging sensors through a transformation matrix, and then calculating the average spectrum of each spectral segment in the coordinate area of ​​other spectral imaging sensors; combining the average spectra of all spectral imaging sensors to obtain a joint spectrum;

[0097] S5. Collect the joint spectrum of known common coatings and fire-retardant coatings using the above method, train and test the classifier, and optimize the classifier parameters according to the test accuracy;

[0098] S6. Collect the joint spectrum of the unknown coating using the method of steps S1 to S4, and quickly identify the unknown coating using the determined classifier.

[0099] The calculation method of steps S1 to S4 in this method is consistent with the above, and the whole process is described for easy understanding.

[0100] It should be noted that the cascaded spectral imaging microsystem designed by the present invention can be used not only for the identification of steel structure fire retardant coatings, but also for snapshot spectral imaging detection applications with spectral feature distribution in the visible-near infrared range of not less than 200nm. For example, in the process of fire investigation on building steel structures, it is necessary to quickly identify the residual sample traces at the fire scene in a non-contact manner, and judge the occurrence and development of the fire by the morphology and color of the sample traces. Among them, in the process of recording, transmitting and analyzing colors using spectra, the detection spectrum range needs to span the full spectrum range of visible light 380nm-670nm. In addition to the customized design of spectral imaging sensors for steel structure fire investigation, the structure involved in the present invention can be directly used. By cascading existing general-purpose spectral imaging sensors covering different spectral ranges into a spectral imaging microsystem, full-spectrum information detection in the visible-near infrared spectral range is achieved, avoiding redesign and semiconductor process resource constraints.

[0101] As another aspect of the present invention, taking the rapid identification of fire retardant coatings for steel structures as an example, the technical solution of the present invention is described in detail.

[0102] The present invention adopts the spectral imaging technology of 400nm-1000nm in the visible-near infrared range to carry out the research work on the spectral characteristic consistency rapid identification technology of intumescent fire retardant coating, non-intumescent fire retardant coating and common coating for steel structure with different fire protection mechanisms, and carries out the spectral characteristic analysis of coating, and carries out the rapid analysis and judgment of the three types of coating by analyzing the consistency characteristics of intumescent fire retardant coating of different brands before and after color adjustment, the consistency characteristics of non-intumescent fire retardant coating of different brands and the consistency characteristics between common coatings. Based on the spectral analysis technology and the intelligent recognition algorithm, the coating that has been coated and cured, and the coating that has not been coated and maintained in the original physical state (such as intumescent fire retardant coating usually presents liquid state, and non-intumescent fire retardant coating usually presents solid state) can be rapidly identified and judged without contacting or destroying the sample.

[0103] The present invention proposes a method for quickly identifying steel structure fire retardant coatings with different fire retardant mechanisms and ordinary coatings based on spectral analysis technology. In the visible-near infrared 400nm-1000nm range, the method can quickly detect the coatings in the original physical state, which have been coated and cured or not coated, according to the spectral consistency of the un-tinted intumescent fire retardant coatings of different brands, the spectral consistency of the non-intumescent fire retardant coatings and the spectral consistency of the ordinary coatings.

[0104] The design covers the range of 400-1000nm. The portable spectral imaging microsystem is used for rapid detection of coatings: the system mainly includes an imaging lens, a spectral imaging module composed of a cascaded universal spectral imaging sensor and a readout circuit, a processing circuit and a power supply module. The system structure diagram is shown in the figure below. Figure 1As shown, the core spectral imaging module is composed of a cascade of universal snapshot spectral imaging sensors covering different spectral ranges, and the processing circuit includes AI modules, interaction modules, storage modules, communication modules and other functional modules.

[0105] Since the spectral imaging sensor based on interference filtering is affected by the interference filtering film material and structure, the free spectral range of the filtering structure is limited, and the design of a single-chip spectral imaging sensor with a wide free spectral range formed by cascading multiple interference filtering structures or multiple materials is also limited by the semiconductor process capabilities, and the realization faces great process challenges. Therefore, a cascaded universal spectral imaging sensor structure is adopted to cascade snapshot spectral image sensor modules covering different spectral ranges to form a snapshot spectral imaging module. For example, a universal spectral imaging sensor 1 covering 400-600nm is cascaded with a universal spectral imaging sensor 2 covering 600-1000nm. Through a single exposure imaging method, the spectral image data of the target scene in the range of 400-1000nm is simultaneously obtained for target processing and analysis.

[0106] Taking the universal spectral imaging sensor 1 of 400-600nm and the universal spectral imaging sensor 2 of 600-1000nm as examples, each adopts a pixel-level interference thin film filter structure (the interference thin film filter structure in the present invention is a well-known structure, using a common substrate Fabry-Perot cavity filter structure), each interference thin film stacking structure covers a pixel size, forming a spectral channel, such as the sensor 1 consists of a unit group filter structure:

[0107] Sensor 1 band 1, sensor 1 band 2, sensor 1 band 3;

[0108] Sensor 1 band 4, sensor 1 band 5, sensor 1 band 6;

[0109] Sensor 1 band 7, sensor 1 band 8, sensor 1 band 9.

[0110] The arrays are arranged on the photosensitive surface of the image sensor in a periodic arrangement.

[0111] Sensor 2 consists of a unit group filter structure:

[0112] Sensor 2 spectral band 1, sensor 2 spectral band 2, sensor 2 spectral band 3, sensor 2 spectral band 4;

[0113] Sensor 2 spectral band 5, sensor 2 spectral band 6, sensor 2 spectral band 7, sensor 2 spectral band 8;

[0114] Sensor 2 spectral band 9, sensor 2 spectral band 10, sensor 2 spectral band 11, sensor 2 spectral band 12;

[0115] Sensor 2 spectral band 13, sensor 2 spectral band 14, sensor 2 spectral band 15, sensor 2 spectral band 16.

[0116] The arrays are arranged on the photosensitive surface of the image sensor in a periodic arrangement.

[0117] In order to make the two sensors capture the same target scene, the two sensors are equipped with imaging optical systems to form a binocular system. Due to the spatial differences in the imaging scenes, an adaptive image matching method is introduced to fuse multispectral images in different spectral ranges. The data processing flow of the portable steel structure coating rapid inspection system is as follows: Figure 2 As shown, firstly, due to the difference in quantum efficiency between dual sensors and different spectral bands of sensors, it is necessary to automatically set the initial conditions of the instrument and match the parameters of the dual sensors. The present invention designs an adaptive parameter adjustment method for cascade sensors based on sensor response characteristics:

[0118] The light source uses natural light source or halogen lamp wide spectrum light source, and the spectra of the two types of light sources are normalized as standardized light source spectra. Taking sensor 1 as an example, sensor 1 covers 9 narrow-band spectrum bands in the range of 400-600nm. The quantum efficiency Q of each spectrum band of sensor 1 in the free spectrum range (400-600nm) under the standardized light source is calculated. n (n = 1, 2, ..., 9), the quantum efficiency of the spectrum is calculated based on the standardized light source spectrum, and the method is common knowledge. The spectrum with the highest quantum efficiency and the spectrum with the lowest quantum efficiency are selected and recorded as Q max , Q min , Q max , Q min ∈Q n , as the calibration spectrum. During the initialization phase of the device operation, the gain of sensor 1 is set to 0, and the exposure time τ is initialized. 0 (τ 0 ∈τ min ~τ max , τ min , τ max are the shortest and longest exposure times of sensor 1 respectively), and the reflectivity of each spectral band is r w The standard white colorimetric plate is placed in front of the camera to collect initial data. The collected initialization image is recorded as I w0 , calculate I w0 The maximum gray value Max in the spectrum with the highest quantum efficiency 0 And the minimum gray value Min of the spectrum with the lowest quantum efficiency 0 , make the following judgment and set the exposure time τ:

[0119]

[0120] And re-collect the standard white colorimetric plate data according to the latest exposure time setting I τ , and repeat the above judgment until the maximum grayscale value of the spectrum with the highest quantum efficiency in the standard white colorimetric plate data is in the range of 180-255. At this time, the standard white colorimetric plate data is recorded as I τ .

[0121] In order to couple the spectra of sensors in different spectral ranges, the exposure value of sensor 2 is adjusted based on the exposure value τ of sensor 1. The central wavelength f of the spectral range of sensor 1 1 n (n=1,2,…,9), sensor 2 spectral center wavelength f 2 n (n=1,2,…,16), satisfying f 1 1 <f 1 2 <…<f 1 9 <f 2 1 <f 2 2 <…<f 2 16 Taking the spectral coupling under natural light as an example, a high-precision spectrometer is used to measure the normalized natural light spectrum. Figure 4 As shown in Figure 1, firstly, a polynomial function is used to fit the normalized natural spectrum. Figure 4 The example fitting function can be expressed as F source (λ). Next, calculate the response of spectral band 9 under the exposure time τ of sensor 1, denoted as R 1 (f 1 9 , τ), sensor 2 exposure time τ 2 Should satisfy the spectrum segment 1 response R 2 (f 2 1 , τ 2 ) has the following relationship:

[0122]

[0123] Among them, F source (λ) is the natural light source response varying with the central wavelength λ; is the quantum efficiency of sensor 1 in band 9, is the quantum efficiency of sensor 2 in spectral band 1.

[0124] At this time, the exposure value of sensor 2 is τ 2 No manual adjustment is required, and the dual-sensor spectral matching conditions under natural light conditions are met.

[0125] Through the above method, the dual-sensor system can be adaptively adjusted in parameters and spectral matching, laying the foundation for obtaining reliable engineering spectral image data.

[0126] The present invention also provides a method for calibrating spectral imaging sensor data based on dark noise prior data: for the same sensor, the sensor dark noise changes with the initialization parameters, reflecting the characteristics of the instrument itself, and has nothing to do with the external light source. Therefore, the sensor dark noise response curve is established as prior data. Taking sensor 1 as an example, other parameters are fixed, and the sensor is in an adjustable exposure time range τ min -τ max The inner dark noise variation curve N(τ) is as follows: Figure 3 As shown, a high-order polynomial function F(τ) is used to simulate it, and the functional relationship between the average dark noise and the exposure time can be obtained:

[0127]

[0128] Where s is the degree of the polynomial, a m (m=1,2,…,s) is a set of coefficients. s is determined by the 2-norm of the difference between the measured dark noise variation N(τ) and the fitting function F(τ): Let ||N(τ)-F(τ)|| 2 The smallest s value less than 1 is the optimal s value.

[0129] by Figure 3 For example, when s=6, the above conditions are met, and the sixth-order polynomial function is used to fit the image average dark noise:

[0130]

[0131] F(τ) is used to calibrate the image. Sensor 1 is used to collect target scene data, and the target scene raw data I is corrected to calculate the target scene data I collected by sensor 1. 1 :

[0132]

[0133] Among them, r w is the reflectance of the standard white colorimetric plate for each spectral band; I is the original data of the target scene; I τ is the standard white colorimetric plate data, I hw The unit matrix representing the size of the sensor's photosensitive surface h×w.

[0134] Considering the differences in the distribution of dark noise in the two-dimensional focal plane of the sensor, the dark noise at each pixel position (x, y) is further refined and expressed as F(τ, x, y), and the following relationship exists:

[0135]

[0136] but

[0137] Among them, x and y represent the row and column positions of a single pixel in the two-dimensional focal plane, respectively, x∈{1,2,…,w}, y∈{1,2,…,h}.

[0138] The spectral imaging chip filters and screens the light information entering the system from the imaging lens, and converts it into an electronic signal through photoelectric conversion. The spectral image data is transmitted to the processing module through the post-readout circuit. After the processing module inputs the original spectral image data, it calibrates the dual sensor raw data according to the prior data and the standard white colorimetric plate information, and performs real-time automatic matching of the calibrated image based on the SIFT method to extract the spectrum of the target area. The target area spectrum extraction method is as follows:

[0139] First, the target images collected by sensor 1 and sensor 2 are calibrated according to the above method, and the calibrated data are respectively recorded as I and 1 (h,w),I 2 (h,w), where h and w are the number of rows and columns of pixels of the spectral image sensor, i.e., the height and width of the original spectral data. 2 Projected to I through the transformation matrix T 1 Since the general spectral imaging sensor adopts a pixel-level mosaic spectroscopic structure, the three-dimensional spectral data is obtained by downsampling the two-dimensional original image data. For example, for the 9-band sensor 1 with a unit group filter structure arranged in 3×3, the downsampled image is recorded as I 1d ([h / 3], [w / 3], 9), where “[]” means only the integer part is retained. Similarly, the spectrum data of sensor 2 after downsampling is recorded as I 2d ([h / 4],[w / 4],16).

[0140] Secondly, in the spectral image data I 2d In the spectral image I of the i-th (i∈16) spectral segment, the target spectrum is extracted. 2d ([h / 4], [w / 4], i), image processing methods such as automatic threshold extraction, edge detection, matching expansion and other conventional methods are used to extract the target's spatial position distribution information in the two-dimensional image. For example, if the target is located in the two-dimensional image, 2d (x 1 ,y 1 ,i)~I 2d (x 2 ,y 2 ,i)Range(x 1 , x 2 ∈[h / 4];y 1 ,y 2∈[w / 4]), in this two-dimensional coordinate (x 1 ,y 1 )(x 2 ,y 2 ) range, and calculate the average values ​​of the pixels in 16 channels, which are recorded as the jth spectrum Spec of the target O. 2A (j,16), j represents the number of spectra to be extracted for the same target, which is determined by the extraction requirements and the size of the pixel area occupied by the target in the single-channel image. Any target spectrum needs to be the average of multiple adjacent pixels (no less than 10) to avoid the error caused by random noise. If the size of the target O in the single-channel image is no less than 10×10 pixels, the value interval of a single average spectrum pixel point is 5×5 pixels (x 2 -x 1 =5,y 2 -y 1 =5), then in I 2d (x 1 ,y 1 ,i)~I 2d (x 2 ,y 2 ,i) No less than 36 sets of different spectral curves can be obtained within the range.

[0141] Thanks to the advantages of spectral image data, multiple sets of spectrum extraction of a single target can be performed multiple times in a single spectral image. One data collection can obtain the spectral data set of sensor 2 of the target. Through the conversion matrix T, 9-band spectrum retrieval is performed in the sensor 1 data. For example, (x 1 ,y 1 ) in sensor 1 corresponds to coordinate T([4x 1 / 3],[4y 1 / 3],9), the coordinate is marked as I 1d (x 1 ’ ,y 1 ’ , 9), calculate I 2d (x 2 ,y 2 ,i) corresponds to coordinate I in the image data of sensor 1 1d (x 2 ’ ,y 2 ’ , 9), calculate the average spectrum Spec 1A (j,9), the average spectrum combination of the two sensors can obtain the matching dual sensor joint spectrum Spec A (j, 25).

[0142] The intumescent fire retardant coating is marked as Class A, the non-intumescent fire retardant coating is marked as Class B, and the ordinary coating is marked as Class C. Parts of the target data are randomly selected as training sets, and feature learning and classifier training are performed on the spectral features of various targets in the training set. Feature learning and classification methods include but are not limited to machine learning methods such as linear discriminant analysis (LDA), principal component analysis (PCA), random forest (RF), support vector machine (SVM) or various conventional neural network models such as feedforward neural network (BP) and one-dimensional convolutional neural network (CNN). The part of the training set in various target data is used as the test set, and the trained classifier is used to quickly identify the test set data to obtain the predicted category, and the model accuracy is judged according to the true category of the test set data. If the accuracy meets the requirements, the model is used to identify samples of unknown categories; if the accuracy does not meet the requirements, the model is optimized, such as network adjustment or parameter adjustment, until the model prediction accuracy meets the requirements.

[0143] Finally, a spectral imaging device is used to collect spectral image data of unknown types of paint, and the above-mentioned algorithm model that meets the accuracy requirements is used to process the spectral image data and predict the target category.

[0144] For example, two different brands of non-expandable fire retardant coating samples for steel structures (sample1, sample2), two different brands of intumescent fire retardant coating samples for steel structures (sample3, sample4), and a common coating without fire retardant function (sample5) are selected, and the spectral image data of the 5 coatings in the range of 400-1000nm before coating, in the original physical state, and after coating and curing, with no less than 150 channels, are collected. After black and white correction preprocessing and target data extraction, the spectral curves of each sample can be obtained as follows: Figure 5 As shown. As can be seen from the figure, the spectra of various targets have high consistency before and after coating, and the spectral data of various samples are extracted from the hyperspectral image as a sample set. In order to avoid the influence of random noise, the single-target hyperspectral data takes the average value of the 5×5 pixel range spectrum of the target (calculated according to sensor 2) to make a total of 500 sample sets. 90% of the data of each category is randomly selected as the training set, and the 1-dimensional data features are obtained after the PCA method is used for dimensionality reduction. Combined with the LDA method, the A, B, and C category judgments are realized. The remaining 10% of the samples are taken for testing, and the recognition accuracy of the three categories of samples reaches 100%. That is, for the existing sample1~sample5 samples, including intumescent fire retardant coatings, non-intumescent fire retardant coatings and ordinary coatings without fire retardant properties, the above-mentioned PCA based on spectral images combined with the LDA model can realize the rapid and accurate distinction of unknown category samples.

[0145] According to user needs, data can be stored, analyzed, transmitted and other operations can be performed. The AI ​​module at the core of the processing circuit integrates an intelligent recognition algorithm, and forms an independent judgment mechanism based on the consistency analysis results of the target prior data. It can quickly process and match the collected raw data, and automatically identify the target as one of the preset categories. Finally, the interactive module outputs the judgment result and provides it to the user as a reference.

[0146] Features described and / or illustrated above for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or used in place of features in other embodiments.

[0147] It should be emphasized that the term "include / comprises" when used herein refers to the presence of features, integers, steps or components, but does not exclude the presence or addition of one or more other features, integers, steps, components or combinations thereof.

[0148] The many features and advantages of these embodiments are apparent from this detailed description, and thus the appended claims are intended to cover all such features and advantages of these embodiments that fall within their true spirit and scope. Furthermore, since numerous modifications and changes will readily occur to those skilled in the art, it is not intended that the embodiments of the invention be limited to the exact construction and operation illustrated and described, but rather all suitable modifications and equivalents falling within the scope thereof are intended to be covered.

[0149] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0150] Parts of the present invention that are not described in detail are well known to those skilled in the art.

Claims

1. A cascaded spectral imaging microsystem, characterized in that: It comprises an imaging lens, a spectral imaging module and a processing circuit; the spectral imaging module comprises at least two cascaded spectral imaging sensors and corresponding readout circuits, the multiple spectral imaging sensors respectively cover different free spectrum ranges, each spectral imaging sensor comprises a unit group filter structure periodically arranged on the photosensitive surface of the sensor, the unit group filter structure comprises a plurality of pixel-level interference thin film filter structures arranged in an array form; an imaging lens is respectively arranged in front of each spectral imaging sensor, and the multiple spectral imaging sensors are arranged in parallel to synchronously collect target images; the processing circuit receives the output data of all the readout circuits and performs data processing.

2. The cascaded spectral imaging microsystem according to claim 1, characterized in that: The processing circuit comprises: AI module, used to integrate the target identification algorithm, process and match the collected raw data, and automatically identify the target as one of the preset categories; An interactive module, used to output the discrimination results; A storage module, used for storing data; Communication module, used to communicate with various terminals to implement command issuance and result display; The cascaded spectral imaging microsystem also includes a power supply module, which is used to supply power to the cascaded spectral imaging microsystem.

3. The cascaded spectral imaging microsystem according to claim 1, characterized in that: The spectral imaging module includes two cascaded spectral imaging sensors. The unit group filter structure of the first spectral imaging sensor includes 3×3 interference thin film filter structures, covering the range of 400-600nm. The unit group filter structure of the second spectral imaging sensor includes 4×4 interference thin film filter structures, covering the range of 600-1000nm. The cascaded spectral imaging microsystem is used to collect spectral image information of steel structure fire retardant coatings and ordinary coatings.

4. A spectrum acquisition method for a cascaded spectrum imaging microsystem, characterized in that: Using the cascaded spectral imaging microsystem according to any one of claims 1 to 3, the spectral acquisition method comprises the following steps: S1, adaptive parameter adjustment of multiple spectral imaging sensors; S2, multiple spectral imaging sensors collect target raw data, obtain two-dimensional data through calibration, and obtain three-dimensional spectral data through downsampling; S3, three-dimensional spectral data of multiple spectral imaging sensors are matched in real time through images to determine the transformation matrix; S4. Extract a first target spectral range from the three-dimensional spectral data of a spectral imaging sensor through image processing, calculate the average spectrum of each spectral segment of the first target spectral range, calculate the coordinate area of ​​the first target spectral range in the three-dimensional spectral data of other spectral imaging sensors through a transformation matrix, and then calculate the average spectrum of each spectral segment in the coordinate area of ​​other spectral imaging sensors; combine the average spectra of all spectral imaging sensors to obtain a joint spectrum.

5. The spectrum acquisition method according to claim 4, characterized in that: The method for adaptively adjusting parameters of multiple spectral imaging sensors includes: S101, selecting a light source, and normalizing the light source spectrum to obtain a standardized light source spectrum; S102, sorting the plurality of spectral imaging sensors from small to large according to the free spectral range values, taking the smallest spectral imaging sensor as a reference sensor, and determining an exposure time of the reference sensor; S103, starting from the reference sensor, select two adjacent spectral imaging sensors in order, and adjust the exposure time of the next spectral imaging sensor according to the exposure time of the previous spectral imaging sensor, the method is as follows: Among them, F source (λ) is the light source response that varies with the central wavelength λ; is the maximum central wavelength of each spectral band of the previous spectral imaging sensor, is the minimum central wavelength of each spectral band of the latter spectral imaging sensor, is the previous spectral imaging sensor at exposure time τ The response of the corresponding spectrum band, The latter spectral imaging sensor is at exposure time τ2 The response of the corresponding spectrum band, The maximum central wavelength of the previous spectral imaging sensor The quantum efficiency of is the minimum central wavelength of the latter spectral imaging sensor Quantum efficiency of the spectrum, ∑ λ represents the sum over all λ.

6. The spectrum acquisition method according to claim 5, characterized in that: The method for determining the exposure time of the reference sensor in step S102 is as follows: Calculate the quantum efficiency of each spectral band of the reference sensor based on the standardized light source spectrum, and determine the spectral band with the highest and lowest quantum efficiency; Set the gain of the reference sensor to 0, initialize the exposure time, collect the initialization image of the standard white colorimetric plate, and calculate the maximum grayscale value of the spectrum with the highest quantum efficiency and the minimum grayscale value of the spectrum with the lowest quantum efficiency in the initialization image; Adjust the exposure time according to the maximum and minimum grayscale values ​​until the maximum grayscale value of the spectrum with the highest quantum efficiency is in the range of 180-255, and record the data of the standard white colorimetric plate and the exposure time of the reference sensor.

7. The spectrum acquisition method according to claim 6, characterized in that: The method for adjusting the exposure time of the reference sensor according to the maximum and minimum grayscale values ​​is as follows: Among them, Max0 is the maximum grayscale of the spectrum with the highest quantum efficiency, Min0 is the minimum grayscale value of the spectrum with the lowest quantum efficiency, τ is the reference sensor exposure time, and τ0 is the initialization exposure time.

8. The spectrum acquisition method according to claim 4, characterized in that: The calibration method of the target raw data of each spectral imaging sensor is as follows: S201, fixing other parameters of the spectral imaging sensor, obtaining the spectral imaging sensor in an adjustable exposure time range τ min -τ max Internal dark noise variation curve; S202, using a high-order polynomial function to fit the dark noise variation curve, obtain the functional relationship F(τ) between the average dark noise and the exposure time, and use F(τ) to calibrate the target raw data collected by the spectral imaging sensor: Among them, r w is the reflectance of the standard white colorimetric plate for each spectral band; I is the original data of the target scene; I τ is the standard white colorimetric plate data, I hw The identity matrix representing the size of the photosensitive surface of the spectral imaging sensor.

9. The spectrum acquisition method according to claim 4, characterized in that: Calculating the average spectrum of each spectral segment in the first target spectral range specifically includes the following steps: The first target spectral range in the three-dimensional spectral data is I 2d (x1, y1, i)~I 2d (x2, y2, i), select multiple pixels within the two-dimensional coordinate range of (x1, y1)(x2, y2), calculate the average values ​​of the multiple pixels in each channel of the spectral imaging sensor, and obtain the average spectrum of the target collected by the spectral imaging sensor.

10. A method for identifying fire retardant coatings for steel structures based on spectral imaging, characterized in that: The method of collecting target raw data by using the cascaded spectral imaging microsystem described in any one of claims 1 to 3 and obtaining a joint spectrum by using the spectral acquisition method described in any one of claims 4 to 9 further comprises the following steps: Collect the joint spectra of known common coatings and fire-retardant coatings, train and test the classifier, and optimize the classifier parameters according to the test accuracy; Collect the joint spectrum of unknown paints and use the determined classifier to quickly identify the unknown paints.

Images