A ground-based hyperspectral remote sensing system for rapid and high-resolution imaging of multi-component atmospheric trace components

Abstract

The present invention discloses a ground-based hyperspectral remote sensing system for rapid high-resolution imaging of trace components in the atmosphere of multiple components, a carrying platform, a shell carried by the carrying platform and capable of height adjustment and all-around rotation, and a scene sensing device, a front optical subsystem, a spectroscopic and detection subsystem, a collection control terminal, and an inversion calculation terminal are mounted outside the shell. The system has extremely high spectral resolution and advanced inversion processing algorithms, overcomes the drawbacks of applying two-dimensional imaging spectroscopy technology to VOCs component monitoring, obtains the concentration of each pollutant component and smoke plume diffusion information in real time, and can trace the source based on the concentration results and image information.

Description

Technical Field

[0001] The invention belongs to the technical field of optical remote sensing, and in particular relates to a ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components. Background Art

[0002] Various atmospheric trace components such as NO2, SO2, HCHO, VOCs and O3 in the world environment are polluting components that are endangering people's health to varying degrees. Among them, volatile organic compounds (VOCs) are an important source of air pollution. They participate in photochemical reactions, produce atmospheric pollutants such as O3, and cause problems such as photochemical smog and haze. High concentrations of VOCs not only harm the quality of the ambient air, but some VOC species are also toxic, carcinogenic, and have irritating odors, which will have a serious negative impact on the health and quality of life of people in the surrounding areas.

[0003] Most traditional methods of monitoring polluted gases are obtained through in-situ mass spectrometry. This method has extremely high accuracy, but its disadvantages are also very obvious: the equipment is expensive, the time resolution is low and the timeliness is poor, it is difficult to obtain the large-scale distribution of polluted gases, and it consumes a lot of manpower and material resources. One-dimensional imaging spectroscopy, as a remote sensing technology, has been applied to the monitoring of polluted gases, but there are still problems with insufficient temporal and spatial resolution, making it difficult to capture emission sources and difficult to reflect the dynamic characteristics of emission sources in real time. There is also the problem of difficulty in matching the observation field of view with the scene.

[0004] At the same spatial resolution, two-dimensional imaging spectroscopy has a temporal resolution that is 100 times greater than that of one-dimensional imaging spectroscopy. It can produce imaging results within minutes and can capture small changes in emission sources, which is beneficial to the establishment of a dynamic emission inventory. Two-dimensional imaging spectroscopy has been successfully applied to the inversion of common components such as NO2, SO2 and HCHO. For example, Wu Ziyang published a study on the two-dimensional distribution remote sensing of polluted gases based on imaging spectroscopy in the Journal of the University of Science and Technology of China. The literature (VOCs hyperspectral imaging: A new insight into evaluate emissions and the corresponding health risk from industries) discloses the observation of factory VOCs components through one-dimensional imaging spectroscopy. So far, no two-dimensional imaging spectroscopy has been applied to VOCs component monitoring, mainly because compared with one-dimensional imaging spectroscopy, the challenges faced by two-dimensional imaging spectroscopy in monitoring VOCs are more arduous. Not only must the accuracy of the absorption cross section and the cross-interference of various components be considered, but also the striping effect of the array CCD must be eliminated, sufficient signal intensity must be ensured, and appropriate image processing methods must be considered. Summary of the invention

[0005] In view of the above, the object of the present invention is to provide a ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components, so as to achieve large-scale high-resolution dynamic monitoring of trace gases including VOCs components.

[0006] To achieve the above-mentioned purpose of the invention, an embodiment provides a ground-based hyperspectral remote sensing system for rapid high-resolution imaging of trace components in a multi-component atmosphere, comprising: a carrying platform, a shell carried by the carrying platform and capable of height adjustment and omnidirectional rotation around, a scene sensing device, a front optical subsystem, a spectroscopic and detection subsystem, a collection control terminal, and an inversion calculation terminal are carried outside the shell;

[0007] The front optical subsystem includes a fused quartz glass window, a total reflection UV-coated quartz prism, a small motor, and a UV-coated achromatic lens. The solar scattered light transmitted through the fused quartz glass window is reflected by the total reflection UV-coated quartz prism driven by the small motor to adjust the observation field, and then received by the UV-coated achromatic lens and emitted to the light splitting and detection subsystem after adjusting the focal length.

[0008] The spectroscopic and detection subsystem includes an adjustable width slit, a reflective optical path, a controllable rotating grating, and a deep-cooled CCD. The outgoing light of the UV-coated achromatic lens passes through the adjustable width slit, is spectroscopically split through the reflective optical path and the controllable rotating grating, and is then received by the deep-cooled CCD and converted into an electrical signal.

[0009] The acquisition control terminal realizes the ultra-wide band acquisition mode based on the acquisition control function;

[0010] The inversion calculation terminal has data processing and visualization functions. Based on the data processing function, scene classification is performed on the scene image collected by the scene perception device, and inversion calculation is performed on the electrical signal to obtain the inversion result. Based on the visualization function, the inversion result is combined with the scene image for visualization presentation, wherein the scene classification result is used to guide the use of different acquisition schemes and / or data processing schemes during remote sensing imaging.

[0011] Preferably, in the spectrometry and detection subsystem, when the focal length and slit width are required to meet the signal-to-noise ratio requirements, the spectral resolution of each row after spectrometry is ≤0.35nm, and the slit functions of different rows are consistent. The controllable rotating grating is in the shape of a triangular pyramid with different blazing wavelengths, and the grating angle can be changed to change the band range according to the inversion component requirements.

[0012] Preferably, a calibration subsystem is also included, which includes a mercury lamp and a transmission scattering plate, which is used to calibrate the spectral wavelength. During calibration, the transmission scattering plate is placed in front of a total reflection UV-coated quartz prism, and the mercury lamp should be directly illuminated to the transmission scattering plate. The light of the mercury lamp is transmitted through the transmission scattering plate and then reflected by the total reflection UV-coated quartz prism and the UV-coated achromatic lens to be received by the spectroscopic and detection subsystem to ensure that the light is evenly projected on the CCD. After collecting the mercury lamp spectrum, a mapping relationship between the mercury lamp spectral line peak and the CCD spectral dimension channel is established for calibration.

[0013] Preferably, in the inversion calculation terminal, when performing inversion calculation on the electrical signal, the electrical signal is first preprocessed, including:

[0014] Performing a spatial dimension merging operation on the solar scattering spectrum, and / or performing a spectral dimension merging operation when inverting extremely weak signal components;

[0015] A sliding average filter will be added to the spectral dimension for filtering;

[0016] Before each inversion, the wavelength of the solar scattering spectrum is calibrated, and the dark current DC and bias spectrum OFS are measured at regular intervals during the measurement process. These spectra are interpolated from the time dimension to obtain the DC and OFS at each measurement, which are then deducted during inversion.

[0017] Preferably, in the inversion calculation terminal, the inversion result is visualized in combination with the scene image based on the visualization function, including:

[0018] The automatic overlay processing method is adopted. According to the original spectral data, the spectra collected in the visible light band channel are superimposed, and the grayscale image of the visible band is obtained synchronously. Based on this, the appropriate threshold is selected through histogram averaging or maximum inter-class variance method processing. According to the determined threshold, the pixels below the threshold are recorded, and the recorded coordinates are assigned NAN values ​​in the inversion concentration map corresponding to the inversion result, thereby realizing the subtraction of building obstruction, and then the inversion concentration map is superimposed with the grayscale image of the visible band.

[0019] Preferably, in the inversion calculation terminal, the inversion result is visualized in combination with the scene image based on the visualization function, including:

[0020] Based on the target detection of the scene classification result map, the emission sources are automatically identified and marked in the scene;

[0021] Overlay the scene map marked with emission sources and the inverted concentration map.

[0022] Preferably, in the inversion calculation terminal, when visually presenting the inversion result in combination with the scene image based on the visualization function, spline interpolation is also used to first fill in the data of the inversion result, and Kriging interpolation and wavelet transform processing are performed on the concentration image.

[0023] Preferably, the scene classification results are used to guide the use of different acquisition schemes and / or data processing schemes during remote sensing imaging, including:

[0024] When the scene classification results include smoke plume scenes, for scenes where smoke plumes can be clearly identified, the wind direction is detected by identifying the plume drift direction, and the upwind edge frame is used as the reference spectrum during the inversion calculation, without the need to subsequently collect the zenith spectrum as the reference spectrum;

[0025] When the scene classification results include road source scenes, that is, for road source scenes, the spectral dimension merging operation is performed during the inversion data preprocessing process, and the contrast of the inversion image is increased when the inversion results are visualized;

[0026] When the scene classification results include irregular emission scenes, for irregular emission scenes, the control system push-scan frame number increases and the scanning angular velocity decreases based on the acquisition control function;

[0027] When the scene classification result includes a scene that cannot be identified, a conventional push-broom mode is used for the scene that cannot be identified, and a spectrum is collected based on the push-broom mode.

[0028] Preferably, the acquisition control terminal implements the ultra-wide band acquisition mode based on the acquisition control function, including:

[0029] Based on the controllable rotating grating, the grating angle is controlled during the acquisition process, the band range of a single acquisition is expanded, and the inversion of more components is achieved. A stop-and-sweep acquisition mode is adopted, that is, the turntable included in the supporting platform rotates a fixed step and then stops, and the controllable rotating grating is started for wide-band acquisition, and the next rotation is entered after the acquisition is completed.

[0030] Preferably, the acquisition control terminal has an automatic focusing function of defining a core band, wherein the defined core band is a target gas inversion band.

[0031] Preferably, the acquisition control terminal is decoupled and is designed using a producer-consumer model architecture, thereby achieving complete decoupling of the acquisition process, the writing process, and the communication process;

[0032] The acquisition control terminal and the inversion calculation terminal are integrated, and the inversion algorithm part is written in Python. The entire part is integrated and packaged to generate a dynamic link library file, and the dynamic link library is called in the acquisition part to realize the integration of the software. The functions provided by the dynamic link library are highly encapsulated. Only the spectrum file and the photo path need to be provided. The algorithm will automatically identify the scene emission source according to the picture and select the algorithm for the corresponding scene. The reference spectrum and the observed spectrum are automatically judged according to the spectrum file naming and generation order, and multi-threaded inversion is performed. The inversion concentration map generated by the inversion will be returned to the acquisition part and displayed according to the selected result type.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The system of the present invention has extremely high spectral resolution and advanced inversion processing methods, which overcomes the shortcomings of applying two-dimensional imaging spectroscopy technology to VOCs component monitoring, obtains the concentration and plume diffusion information of each pollution component in real time, and can trace the source based on the concentration results and image information. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1 is a schematic structural diagram of a ground-based hyperspectral remote sensing system provided in an embodiment;

[0037] Figure 2 It is a block diagram of the implementation principle of the ground-based hyperspectral remote sensing system provided by the embodiment;

[0038] Figure 3 is a flow chart of automatically identifying scene sources provided by an embodiment;

[0039] Figure 4 It is the acquisition control terminal interface provided by the instance.

[0040] Figure 5 is an automated measurement inversion flow chart provided by an embodiment;

[0041] Figure 6 This is the final result diagram provided by the example. Taking the VOCs component formaldehyde as an example, the red box is the result of the identified source. There is no HCHO emission from other chimneys in the scene. DETAILED DESCRIPTION

[0042] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0043] In order to fill the technical gap of large-scale high-resolution dynamic monitoring of VOCs components as soon as possible, the present invention provides a ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components, such as Figure 1 Shown and Figure 2 As shown, it includes a carrying platform, a shell supported by the carrying platform and capable of height adjustment and all-round rotation, and the outside of the shell is equipped with a scene sensing device, a front optical subsystem, a spectroscopic and detection subsystem, a collection control terminal, and an inversion calculation terminal.

[0044] In the embodiment, the bearing platform includes a hydraulic lifting platform 1 and a high-precision one-dimensional turntable 2, which are used to adjust the height and azimuth angle of the shell and all subsystems therein. Specifically, the height of the shell and all subsystems therein can be adjusted between 1 and 3 meters by the hydraulic lifting platform 1 to meet the needs of different observation scenes and ensure that the field of view is not blocked by the surrounding environment. The one-dimensional turntable 2 is used to adjust the observation azimuth angle within the range of 0 to 360°, with an accuracy within 0.05°, to ensure the accuracy of the push-scan observation field of view.

[0045] In the embodiment, the scene perception device 4 is used to perceive the external scene and collect scene images, and may be a camera, which is mounted on the housing 3. The collected scene images are subjected to scene classification to obtain scene classification results.

[0046] In the embodiment, the front optical subsystem includes a fused silica glass window 9, a total reflection UV-coated quartz prism 6, a small high-precision motor 5, and a UV-coated achromatic lens 7, which are responsible for guiding the solar scattered light into the spectroscopic and detection system. Among them, the fused silica glass window 9 plays a role in protecting the optical system. The total reflection UV-coated quartz prism 6 is mounted on the small high-precision motor 5, and the pitch observation field of the prism can be adjusted with an accuracy of 0.05°. The use of the total reflection UV-coated quartz prism 6 instead of the overall system movement to turn the light path can avoid the error caused by the change of the position of the spectroscopic device. The solar scattered light transmitted through the fused silica glass window 9 is reflected by the total reflection UV-coated quartz prism 6 and then received by the UV-coated achromatic lens 7. The UV-coated achromatic lens 7 adjusts the focal length and emits the solar scattered light to the spectroscopic and detection subsystem, so that the spectroscopic and detection system can form a clear image with a field of view of 16°.

[0047] In the embodiment, the light splitting and detection subsystem includes an adjustable width slit 15, a reflective optical path 11, a controllable rotating grating 8, and a deep cooling CCD 10. These components cooperate to split the incoming solar scattered light from the front optical system and convert the optical signal into an electrical signal. Specifically, the outgoing light of the UV-coated achromatic lens 7 passes through the adjustable width slit 15, and then is split by the reflective optical path 11 and the controllable rotating grating 8, and then is received by the deep cooling CCD 10 and converted into an electrical signal.

[0048] Specifically, in the spectroscopic and detection subsystem, when the focal length and slit width are required to meet the signal-to-noise ratio requirements, the spectral resolution of each row after spectroscopic analysis is ≤0.35nm, and the slit functions of different rows are consistent; the reflection optical path 11 transmits the optical signal between the adjustable width slit 15, the controllable rotating grating 8 and the CCD 10. Among them, the controllable rotating grating 8 is in the shape of a triangular pyramid, with 150, 300 and 600 lines on the three sides, respectively, and has different blazing wavelengths. The grating angle can be changed to change the band range according to the inversion component requirements. The pixel specifications of CCD 10 are 2048 channels in the spectral dimension and 512 channels in the spatial dimension. It has a built-in deep refrigeration module that can maintain the temperature at -20°C to reduce wavelength drift and improve the accuracy of subsequent inversion work.

[0049] The ground-based hyperspectral remote sensing system provided in the embodiment also includes a calibration subsystem 17, which includes a mercury lamp and a transmission scattering plate, which is used to calibrate the wavelength of the spectrum. During calibration, the calibration subsystem 17 is fixed in front of the rotating total reflection UV-coated quartz prism 6 by screws, and the mercury lamp should be directly illuminated through the transmission scattering plate to the total reflection UV-coated quartz prism 6. The mercury lamp light is reflected by the total reflection UV-coated quartz prism 6 and then passed through the UV-coated achromatic lens 7 to be received by the spectroscopic and detection subsystem to ensure that the light is evenly projected on the CCD 10. After collecting the mercury lamp spectrum, the calibration subsystem 17 should be restored to its original position. After collecting the mercury lamp spectrum, a mapping relationship between the mercury lamp spectral line peak and the CCD spectral dimension channel is established, which is usually a polynomial of the third order or above, so as to perform calibration.

[0050] The ground-based hyperspectral remote sensing system provided in the embodiment also includes a temperature control subsystem, which is responsible for temperature control of the entire instrument to eliminate spectral drift caused by tiny displacements of optical devices caused by temperature changes. The temperature control subsystem includes a semiconductor refrigerator, a cooling fan, a high-precision temperature sensor 16 and an MCU circuit board, wherein the MCU circuit board and the high-precision temperature sensor 16 are fixed on the base plate 12, the semiconductor refrigerator is located at the top and bottom of the shell 3, and is attached to the instrument base plate 12, the cooling fan 14 is in close contact with the semiconductor refrigerator, and the temperature sensor is attached to the outside of the spectroscopic and detection subsystem. The MCU circuit board uses a PID algorithm to adjust the power of the semiconductor refrigerator to adjust the optical path temperature to 20±0.1°C.

[0051] In the embodiment, the control terminal and the inversion computing terminal are integrated together, such as integrated into PC 13, wherein the inversion computing terminal has data processing function and visualization function. Among them, based on the data processing function, the scene image collected by the scene perception device is classified, specifically, the scene image is feature extracted by the HOG algorithm, the vocabulary of the image feature is constructed by using the bag of words model, and the scene image is classified based on the image feature by using a classifier such as a support vector machine. The scene image and the scene classification results are convenient for guiding the use of different acquisition schemes and / or data processing schemes in subsequent remote sensing imaging, including:

[0052] When the scene classification results include a smoke plume scene, that is, for a smoke plume scene that can be clearly identified, the wind direction is detected by identifying the direction of the plume, and the upwind edge frame is used as the reference spectrum in the inversion calculation. There is no need to subsequently collect the zenith spectrum as a reference spectrum, thereby further improving the temporal resolution.

[0053] When the scene classification results include road source scenes, that is, for road source scenes, since the road source emission components are relatively simple, high resolution is not required, but complex scenes have higher requirements for the signal-to-noise ratio. Therefore, in the inversion data preprocessing process, the system signal-to-noise ratio is further improved by spectral dimension merging. At the same time, when the inversion results are visualized, the contrast of the inversion image is increased to highlight the diffusion phenomenon of road source emissions.

[0054] When the scene classification results include irregular emission scenes, for irregular emission scenes (containing potential multiple emission sources), based on the acquisition control function, the system push-scan frame number is increased, preferably doubled, and the scanning angular velocity is reduced, preferably halved, to obtain more imaging details to trace the emission sources.

[0055] When the scene classification results include scenes that cannot be identified, the conventional push-broom mode is used for the scenes that cannot be identified, and the spectrum is collected based on the push-broom mode. Different processing for different scenes can significantly improve the final presentation effect of the inversion.

[0056] Based on the data processing function, the electrical signal corresponding to the spectrum is inverted and calculated to obtain the inversion result, including the following steps:

[0057] Step 1: preprocess the electrical signal corresponding to the collected solar scattering spectrum.

[0058] Given that VOCs components have weak absorption and slow changes, inversion requires an extremely high signal-to-noise ratio. Therefore, the solar scattering spectrum will be merged in the spatial dimension to greatly improve the signal-to-noise ratio while ensuring spatial resolution. In order to filter out high-frequency white noise, a sliding average filter will be added to the spectral dimension.

[0059] For traditional inversion algorithms, dark current spectra (DC) and bias spectra (OFS) are collected before measurement and deducted from the measured spectrum to avoid the impact of instrument dark current and bias on the inversion results. For two-dimensional imaging instruments, the use of traditional methods to collect DC and OFS spectra will cause a huge waste of time resources and storage resources, and lack real-time performance. For this reason, in the present invention, in order to ensure the quality of the spectrum, the solar scattering spectrum will be calibrated once for wavelength before each inversion, and the dark current (DC) and bias spectrum (OFS) will be measured once every period of time during the measurement process, and interpolated from the time dimension according to these spectra to obtain DC and OFS at each measurement, which will be deducted during inversion. When inverting extremely weak components of the signal, the spectral dimension will also be considered to improve the signal-to-noise ratio.

[0060] Specifically, at the beginning of the acquisition, the shutter will be closed to acquire 20 frames of background spectra, and the same operation will be performed every half an hour. For each acquired raw spectrum, a correction process will be performed:

[0061]

[0062] Among them, i represents the spectral dimension, j represents the spatial dimension, k represents the scanned frame, S represents the spectrum after deducting DC and OFS, DN represents the original spectrum, and D represents the dark spectrum. and Respectively represent the dark spectra before and after the spectrum DN acquisition, count represents the number of frames, and ratio represents the proportion of the current spectrum acquisition time in the dark spectrum acquisition time period before and after. This method fully considers the real-time noise removal and the applicability to three-dimensional spectral data, and can effectively improve the inversion quality. In order to obtain smoother and more reasonable imaging results and remove stripe noise, the image will be processed by Kriging interpolation and wavelet transform.

[0063] In addition, since the data collected each time are in three-dimensional cubic form (vertical, horizontal, and spectral), the data will first be processed into a single std format for the next step of inversion.

[0064] Step 2: Inversion of atmospheric trace component column concentration (VCD).

[0065] Considering that different atmospheric components have different absorption characteristics, the slant column concentration (SCD) of various atmospheric pollutants (such as NO2, SO2, HONO, CS, C1, and VOCs such as aldehydes, ketones, esters and benzene series) can be inverted using the least squares fitting algorithm based on the collected solar scattering spectra at various horizontal azimuth and elevation angles. Considering the huge amount of data, ten threads will be opened for each inversion at the same time, and mutex locks will be added in this process to avoid read-write conflicts.

[0066] After obtaining the SCD, a simplified geometric calculation method is used to calculate the atmospheric photon path AMF based on the field of view angle and elevation angle, and the VCD is obtained as the inversion result according to VCD=SCD / AMF.

[0067] The inversion calculation terminal visualizes the inversion results combined with the scene image based on the visualization function. After filtering out the VCD data with too large errors, the remaining VCD data will be plotted as a concentration image. When visualizing the inverted concentration, an automatic overlay processing method can be used, that is, the mapping established based on the corresponding relationship between the spectral lens and the coaxial position of the visible camera is overlaid to obtain a scene map of the iterative inversion result map (inverted concentration map).

[0068] Specifically, according to the original spectral data, the spectra collected in the visible light band channel are superimposed to obtain the grayscale image of the visible band synchronously. Based on this, the appropriate threshold can be selected through processing methods such as histogram averaging and maximum inter-class variance method. According to the determined threshold, the pixels below the threshold are recorded, and the recorded coordinates are assigned NAN values ​​in the inversion concentration map corresponding to the inversion result, thereby realizing the deduction of building occlusion. The inversion concentration map and the grayscale image of the visible band are both from the instrument push scan, so they have the same number of pixels. Directly superimposing the inversion concentration map and the grayscale image of the visible band can achieve synchronous field of view matching between the inversion result and the real scene, and solve the field of view deviation caused by the field of view angle, distortion and imaging instrument of the visible camera.

[0069] When visualizing the inversion concentration corresponding to the inversion result, target detection based on the scene classification result map is also performed to automatically identify the emission source and mark the emission source in the scene map. This method is particularly suitable for complex multi-row outlet scenes, and can automatically mark the emission source for multiple components. In the embodiment, during target detection, YOLOv5 target recognition technology is used, and a targeted training set is constructed in combination with observation data, so that the YOLOv5 model can accurately identify the emission source. On the basis of the YOLOv5 framework, in addition to customizing targeted training data, targeted adjustments and optimizations are made to the prior frame, hyperparameters, loss functions, and data enhancement methods to improve recognition accuracy and robustness. The k-means++ clustering algorithm is used to cluster the width and height distribution of the emission source, and the size of the highest priority frame is determined to replace the default prior frame; the OneCycleLR learning rate scheduler is used to dynamically adjust the learning rate to improve the convergence speed and stability of the model; FocalLoss is used to adjust the classification loss so that the model pays more attention to emission sources that are difficult to detect; color jitter is mainly used in data enhancement to enhance the model's adaptability to environments with large changes in outdoor light intensity. Integrate automatic overlay processing and target detection, recognition and annotation processes into the visualization process, such as Figure 3 shown.

[0070] There is a very small probability that there will be missing values ​​in the inversion results, which will affect subsequent processing, so spline interpolation is generally used to fill in the data first. In order to obtain smoother and more reasonable imaging results and remove strip noise, the concentration image will be processed by Kriging interpolation and wavelet transform. After obtaining the final result, a display file will be generated. Based on the acquisition control function, the display file will be displayed and the next acquisition will be started.

[0071] The acquisition control terminal can realize an ultra-wide band acquisition mode based on the acquisition control function, including: based on the controllable rotating grating 8, the grating angle can be controlled during the acquisition process, the band range of a single acquisition can be expanded, a wide band scanning function can be realized, the inversion of more components can be realized, and the inversion components and quality can be improved by expanding the acquisition band. Compared with push-scan acquisition, the present invention can adopt a one-stop-one-scan observation acquisition mode, that is, the turntable 2 stops rotating after rotating a fixed step length, and the controllable rotating grating 8 starts to perform wide-band acquisition, and the next rotation is entered after the acquisition is completed. This method will sacrifice time resolution to a certain extent, but it can greatly expand the acquisition band, improve the inversion quality and realize the inversion of more components.

[0072] In the embodiment, based on the collection control function, the front optical subsystem is controlled to collect the solar scattered light according to the scene classification result. The specific collection scheme is as described above.

[0073] In the embodiment, the acquisition control terminal is decoupled and is designed using a producer-consumer model architecture, which achieves complete decoupling of the acquisition process, writing process, and communication process to avoid potential conflicts during the acquisition, writing, communication, and storage processes. This can further improve the imaging efficiency of the system, and a single imaging can be completed in one minute.

[0074] In the embodiment, the acquisition control terminal also has a function of automatically focusing by defining a core band. Since chromatic aberration cannot be completely eliminated, simultaneous focusing cannot be achieved in a wide band. In the acquisition control terminal, a core band can be defined, usually a target gas inversion band, to achieve optimal focusing for the core band and ensure its spectral quality.

[0075] In the embodiment, the acquisition control terminal and the inversion calculation terminal are integrated. For traditional remote sensing instruments, due to the large amount of work in the acquisition and inversion parts, the acquisition and inversion are usually performed in two different modules, which greatly weakens the integrity of the software, and the cumbersome inversion steps will also bring negative user experience. In this system, the inversion algorithm part is written in Python, and the entire part is integrated and packaged to generate a dynamic link library file, and the dynamic link library is called in the acquisition part to realize the integration of the software. Only the spectrum file and the photo path need to be provided. The algorithm will automatically identify the scene emission source according to the picture and select the algorithm for the corresponding scene, and automatically judge the reference spectrum and the observation spectrum according to the spectrum file naming and generation order, and perform multi-threaded inversion. The inversion concentration map generated by the inversion will be returned to the acquisition part and displayed according to the selected result type. Figure 4 The visual interface of the integrated acquisition control terminal is demonstrated.

[0076] The embodiment also provides a process for automatically measuring and inverting using the above system, such as Figure 5 As shown in the figure, taking HCHO as an example, based on this process, high-resolution imaging is successfully achieved in complex scenes, and automatic identification and marking of emission sources are performed, as shown in the figure. Figure 6 shown.

[0077] The ground-based hyperspectral remote sensing system provided in the above embodiment can realize rapid high-resolution imaging of multi-component atmospheric trace components, especially imaging of weakly absorbing VOCs. Due to the weak absorption of a large number of VOCs components and poor absorption cross-section quality, there is a huge technical gap in large-scale high-resolution dynamic monitoring of VOCs emission sources. In the system of the present invention, all optical elements are coated with ultraviolet anti-reflection films to increase the received scattered light flux, and spectral processing methods such as two-dimensional filtering and merging of spatial and spectral dimensions, dark spectrum subtraction of time dimension interpolation, and optimal focusing to reduce spectral focusing errors, at least one of these means can obtain high-resolution data with a sufficiently high comprehensive signal-to-noise ratio, and realize the inversion of common pollutants (NO2, O3, SO2, etc.) and a variety of weakly absorbing VOCs components (benzaldehyde, formaldehyde, glyoxal, benzoic acid, and 2-hydroxynicotinaldehyde, etc.).

[0078] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components, characterized in that: include: A carrying platform, a shell carried by the carrying platform and capable of height adjustment and all-around rotation, the shell having a scene sensing device, a front optical subsystem, a spectroscopic and detection subsystem, a collection control terminal, and an inversion calculation terminal; The front optical subsystem includes a fused quartz glass window, a total reflection UV-coated quartz prism, a small motor, and a UV-coated achromatic lens. The solar scattered light transmitted through the fused quartz glass window is reflected by the total reflection UV-coated quartz prism driven by the small motor to adjust the observation field, and then received by the UV-coated achromatic lens and emitted to the light splitting and detection subsystem after adjusting the focal length. The spectroscopic and detection subsystem includes an adjustable width slit, a reflective optical path, a controllable rotating grating, and a deep-cooled CCD. The outgoing light of the UV-coated achromatic lens passes through the adjustable width slit, is spectroscopically split through the reflective optical path and the controllable rotating grating, and is then received by the deep-cooled CCD and converted into an electrical signal. The acquisition control terminal realizes the ultra-wide band acquisition mode based on the acquisition control function; The inversion calculation terminal has a data processing function and a visualization function. Based on the data processing function, the scene image collected by the scene perception device is classified, the electrical signal is inverted and calculated to obtain the inversion result, and the inversion result is combined with the scene image for visualization based on the visualization function. The scene classification result is used to guide the use of different acquisition schemes and / or data processing schemes during remote sensing imaging; In the inversion calculation terminal, when performing inversion calculation on the electrical signal, the electrical signal is first preprocessed, including: performing a spatial dimension merging operation on the solar scattering spectrum, and / or performing a spectral dimension merging operation when the inversion signal is extremely weak; adding a sliding average filter to the spectral dimension for filtering; before each inversion, the solar scattering spectrum is calibrated for wavelength, and the dark current DC and the bias spectrum OFS are measured at regular intervals during the measurement process, and interpolation is performed from the time dimension based on these spectra to obtain the DC and OFS at each measurement, which are deducted during inversion; In the inversion calculation terminal, the inversion result is visualized in combination with the scene image based on the visualization function, including: using an automatic overlay processing method, superimposing the spectrum collected in the visible light band channel according to the original spectral data, and synchronously obtaining a grayscale image of the visible band, and selecting an appropriate threshold through histogram averaging or maximum inter-class variance method processing means, and recording pixels below the threshold according to the determined threshold, assigning NAN values ​​to the recorded coordinates in the inversion concentration map corresponding to the inversion result, thereby achieving the subtraction of building occlusion, and then superimposing the inversion concentration map with the grayscale image of the visible band; or, based on the target detection of the scene classification result map, automatically identifying the emission source, and marking the emission source in the scene; superimposing the scene map marked with the emission source and the inversion concentration map.

2. The ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components according to claim 1, characterized in that: In the spectrometry and detection subsystem, when the focal length and slit width are required to meet the signal-to-noise ratio requirements, the spectral resolution of each row after spectrometry is ≤0.35nm, and the slit functions of different rows are consistent. The controllable rotating grating is in the shape of a triangular pyramid with different blazing wavelengths. The grating angle can be changed to change the band range according to the inversion component requirements.

3. The ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components according to claim 1, characterized in that: It also includes a calibration subsystem, which includes a mercury lamp and a transmission scattering plate, which is used to calibrate the spectral wavelength. During calibration, the transmission scattering plate is placed in front of a total reflection UV-coated quartz prism, and the mercury lamp should be directly illuminated to the transmission scattering plate. The light of the mercury lamp is transmitted through the transmission scattering plate and then reflected by the total reflection UV-coated quartz prism and the UV-coated achromatic lens to be received by the spectroscopic and detection subsystem to ensure that the light is evenly projected on the CCD. After collecting the mercury lamp spectrum, a mapping relationship between the mercury lamp spectral line peak and the CCD spectral dimension channel is established for calibration.

4. The ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components according to claim 1, characterized in that: In the inversion calculation terminal, when visualizing the inversion result in combination with the scene image based on the visualization function, spline interpolation is used to first fill the data of the inversion result, and Kriging interpolation and wavelet transform are performed on the concentration image.

5. The ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components according to claim 1, characterized in that: The scene classification results are used to guide different acquisition schemes and / or data processing schemes for remote sensing imaging, including: When the scene classification results include smoke plume scenes, for scenes where smoke plumes can be clearly identified, the wind direction is detected by identifying the plume drift direction, and the upwind edge frame is used as the reference spectrum during the inversion calculation, without the need to subsequently collect the zenith spectrum as the reference spectrum; When the scene classification results include road source scenes, that is, for road source scenes, the spectral dimension merging operation is performed during the inversion data preprocessing process, and the contrast of the inversion image is increased when the inversion results are visualized; When the scene classification results include irregular emission scenes, for irregular emission scenes, the control system push-scan frame number increases and the scanning angular velocity decreases based on the acquisition control function; When the scene classification result includes a scene that cannot be identified, a conventional push-broom mode is used for the scene that cannot be identified, and a spectrum is collected based on the push-broom mode.

6. The ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components according to claim 1, characterized in that: The acquisition control terminal implements the ultra-wide band acquisition mode based on the acquisition control function, including: Based on the controllable rotating grating, the grating angle is controlled during the acquisition process, the band range of a single acquisition is expanded, and the inversion of more components is achieved. A stop-and-sweep acquisition mode is adopted, that is, the turntable included in the supporting platform rotates a fixed step and then stops, and the controllable rotating grating is started for wide-band acquisition, and the next rotation is entered after the acquisition is completed.

7. The ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components according to claim 1, characterized in that: The acquisition control terminal has an automatic focusing function of defining a core band, wherein the defined core band is a target gas inversion band.

8. The ground-based hyperspectral remote sensing system for rapid high-resolution imaging of multi-component atmospheric trace components according to claim 1, characterized in that: The acquisition control terminal is decoupled and is designed using a producer-consumer model architecture, achieving complete decoupling of the acquisition process, writing process, and communication process; The acquisition control terminal and the inversion calculation terminal are integrated. The inversion algorithm part is written in Python. The whole part is integrated and packaged to generate a dynamic link library file. The dynamic link library is called in the acquisition part to realize the integration of the software. The functions provided by the dynamic link library are highly encapsulated. Only the spectrum file and the photo path need to be provided. The algorithm will automatically identify the emission source of the scene according to the picture and select the algorithm for the corresponding scene. The reference spectrum and the observation spectrum are automatically determined according to the spectrum file naming and the generation order, and multi-threaded inversion is performed. The inversion concentration map generated by the inversion will be returned to the acquisition part and displayed according to the selected result type.

Images