A weak signal target recognition method fusing prior spectral information

Abstract

The application discloses a weak signal target recognition method fusing prior spectrum information and relates to the technical field of unmanned aerial vehicle spectral remote sensing inspection. The application establishes a recognition reference through a methane absorption prior spectrum, a background prior spectrum and an interference prior spectrum, forms a same frame local background baseline spectrum of the same background type for a to-be-recognized pixel in a same frame spectral image, and makes a weak absorption characteristic vector of the to-be-recognized pixel generate with the local background response as the reference. Therefore, the concave response of the low-concentration methane plume in the methane absorption characteristic wave band can be separated from the device surface reflectivity difference, the water surface reflection, the platform deck reflection and the shadow background change, the candidate methane response area is further filtered through the interference response marking, and the recognition result can reduce the non-methane response influence formed by the water vapor, the cloud shadow, the heat source reflection and the metal reflection.

Description

Technical Field

[0001] This invention relates to the field of UAV spectral remote sensing inspection technology, and in particular to a method for identifying weak signal targets by fusing prior spectral information. Background Technology

[0002] During inspections of oil and gas stations, pipeline valve chambers, wellhead equipment, and offshore platforms, methane leaks often manifest as low-concentration, short-duration, intermittent, or thin plumes. Influenced by wind direction, wind speed, thermal disturbance, equipment obstruction, and the orientation of the leak outlet, the spatial position, diffusion direction, and boundary morphology of the methane plume continuously change between adjacent moments, forming a typical scenario for weak signal target identification.

[0003] Current inspection methods typically employ point gas sensors, infrared thermal imaging, optical gas imaging, multispectral imaging, or hyperspectral imaging for methane leak identification. Among these, point gas sensors require drones to enter or traverse the plume cross section, and the sampling results are easily affected by plume deviation, local eddies, and flight paths, making it difficult to establish a stable correspondence between concentration anomalies and specific leak targets.

[0004] Multispectral or hyperspectral identification methods typically utilize methane absorption bands, reference bands, image segmentation, anomaly detection, or spectral line matching results to determine suspected leak areas. However, in oil and gas inspection scenarios, metal pipelines, valve housings, insulation layers, ground, water surfaces, platform decks, and shaded areas exhibit different spectral responses; water vapor, cloud shadows, heat source reflections, and metal reflections can also create non-methane response variations near the characteristic methane absorption bands. The weak absorption response corresponding to low-concentration methane plumes, when superimposed with the aforementioned background fluctuations, is easily obscured by local background changes or misjudged as a non-leakage disturbance.

[0005] Existing processing methods based on fixed methane prior spectra, global background subtraction, or differences between target and reference bands are ill-suited for inspection scenarios where multiple background types coexist within the same image frame. When the background reference and the background type of the pixel to be identified are inconsistent, the weak absorption response is affected by differences in surface reflectivity of equipment and local illumination changes. When the suspected response area remains only at the level of abnormal pixels or abnormal patches, the detection results are difficult to trace back to specific inspection objects such as valve groups, flanges, wellhead devices, tank interfaces, or pipeline interfaces.

[0006] Therefore, how to separate the weak absorption response of low-concentration methane plumes from UAV-borne spectral images in complex backgrounds and stably associate the identification results with specific inspection objects has become a technical problem that needs to be solved in oil and gas inspection. Summary of the Invention

[0007] This application provides a method for identifying weak signal targets by fusing prior spectral information, which solves the problem of difficulty in identifying and locating weak absorption in low-concentration methane plumes under complex backgrounds.

[0008] This invention provides a weak signal target recognition method that integrates prior spectral information, applied to the spectral identification of low-concentration methane plumes in oil and gas inspection scenarios by unmanned aerial vehicles (UAVs), comprising the following steps: The UAV-borne spectral sequence, UAV pose information, environmental flow vector, and spatial record of the inspected object are acquired in the inspection area. The UAV-borne spectral sequence covers the methane absorption characteristic band and the adjacent non-absorption reference band. The spatial record of the inspected object includes the inspected object identifier and the spatial range of the inspected object. A priori spectral information set is established, which includes methane absorption prior spectrum, background prior spectrum and interfering prior spectrum; Based on the background prior spectrum, the background type is marked in the same frame of spectral image, and candidate background pixels are determined from pixels that are in the same frame as the pixel to be identified, are different from the pixel to be identified, have the same background type, do not show methane absorption spectrum, and satisfy similar constraints in the adjacent non-absorption reference band. The candidate background pixels form the local background baseline spectrum of the same frame. Based on the methane absorption prior spectrum, the adjacent non-absorption reference band, and the local background baseline spectrum of the same frame, a weak absorption feature vector of the pixel to be identified is generated. Candidate methane response regions are generated based on the spectral correspondence between the weak absorption feature vector and the methane absorption prior spectrum. Interference response markers are generated based on the interference prior spectrum, and candidate methane response regions with interference response markers are filtered out. Response trajectories are formed for the filtered candidate methane response regions in consecutive frames. The response trajectories are projected onto the inspection area coordinate system based on the UAV pose information. A backtracking range is formed along the opposite direction of the environmental flow vector at the corresponding acquisition time. The intersection of the backtracking range and the spatial range of the inspection object is determined. When the intersection result in consecutive frames satisfies the continuous hit condition for the same inspection object identifier, the response trajectory is identified as the weak signal target of the methane plume corresponding to the inspection object identifier. Output the inspection object identifier and target identification record corresponding to the weak methane plume signal target.

[0009] In some embodiments, the methane absorption prior spectrum includes the absorption position, absorption width, absorption spectral shape of the methane absorption characteristic band, and the baseline spectral shape of the adjacent non-absorption reference band; When generating the weak absorption feature vector, the concave response of the pixel to be identified in the methane absorption feature band, the baseline response in the adjacent non-absorption reference band, and the residual response relative to the local background baseline spectrum of the same frame are extracted respectively.

[0010] In some embodiments, the background prior spectrum includes spectral templates for at least two background types, including metal pipeline surfaces, valve housings, insulation layers, ground, water surfaces, platform decks, and shaded areas. The background type is determined by the response of the neighboring non-absorbing reference band and the broadband reflection variation of the pixel to be identified.

[0011] In some embodiments, generating the candidate methane response region includes: Connectivity aggregation is performed on the weakly absorbing feature vectors of adjacent pixels; Preserve connected regions with methane absorption characteristic band concave response, baseline continuity of adjacent non-absorbing reference bands, and consistency of local background residuals within the same frame; The connected region is used as a candidate methane response region.

[0012] In some embodiments, the interfering prior spectrum includes at least one interfering spectral template selected from water vapor, cloud shadows, heat source reflection, metallic reflection, and shadows; When the weak absorption feature vector of the candidate methane response region forms a spectral correspondence with the interference prior spectrum, and the residual response of the candidate methane response region relative to the local background baseline spectrum of the same frame conforms to the residual change direction of the corresponding interference, an interference response label is generated for the candidate methane response region.

[0013] In some embodiments, determining the response trajectory includes: Spatial registration of consecutive frame spectral images is performed based on the UAV pose information. The candidate methane response regions searched and filtered in the registered consecutive frames have the same absorption spectrum, adjacent spatial positions and consecutive acquisition times. The searched response regions are connected according to the acquisition time to form a response trajectory.

[0014] In some embodiments, the response trajectory refers to the connection result of candidate methane response regions that meet the trajectory connection threshold, are spatially adjacent, have continuous acquisition times, and have weak absorption feature vectors that meet the consistency requirements in consecutive frames. When forming the backtracking range, the response trajectory is projected onto the inspection area coordinate system according to the UAV pose information, and the backtracking range is formed along the opposite direction of the environmental flow vector at the corresponding acquisition time, with the trajectory node position in the response trajectory as the starting point. The backtracking range is used to characterize the possible source area of ​​the weak signal target of the methane plume.

[0015] In some embodiments, the intersection of the backtracking range with the spatial range of the inspected object in the spatial record of the inspected object is determined, and whether the response trajectory points to the same inspected object identifier is determined based on the number of intersections, the intersection ratio and the continuity of the acquisition time in consecutive frames. When the backtracking range in consecutive frames meets the condition of continuous hits on the same inspection object identifier, a correspondence is established between the response trajectory and the inspection object identifier. When the backtracking range points to multiple inspection object identifiers at the same time, conflict resolution is performed based on the number of intersections, intersection ratio, lateral offset distance and collection time continuity of each inspection object. When it is impossible to distinguish a unique inspection object identifier, the multi-object conflict marker, candidate inspection object identifier and corresponding intersection result are output in the target identification record.

[0016] In some embodiments, the target identification record includes the acquisition time, UAV pose, methane absorption spectrum correspondence result, local background baseline spectrum in the same frame, weak absorption feature vector, interference response marker, response trajectory, environmental flow vector, inspection object identifier, inspection object spatial range, object corresponding status, multi-object conflict marker, candidate inspection object identifier, and corresponding intersection result.

[0017] In some embodiments, the prior spectral information set is established based on historical spectral data of the inspection area, on-site leak-free frame spectral data, and methane standard absorption spectrum; during the acquisition of the UAV-borne spectral sequence, the background prior spectrum is updated using on-site leak-free frame spectral data, and the updated background prior spectrum is used to redetermine the local background baseline spectrum of the same frame.

[0018] Through the above technical solution, the present invention can achieve at least the following beneficial effects: This invention establishes an identification reference by using methane absorption prior spectra, background prior spectra, and interference prior spectra. Within the same frame of the spectral image, a local background baseline spectrum of the same background type is generated for the pixels to be identified, allowing the weak absorption feature vector of the pixels to be identified to be generated with reference to the local background response. Thus, the concave response of a low-concentration methane plume in the methane absorption characteristic band can be separated from differences in equipment surface reflectivity, water surface reflection, platform deck reflection, and shadow background changes. After further filtering of candidate methane response regions using interference response markers, the identification results can reduce the influence of non-methane responses caused by water vapor, cloud shadows, heat source reflection, and metallic reflection.

[0019] By forming response trajectories of the filtered candidate methane response regions in consecutive frames, and then using these trajectories to form a backtracking range along the opposite direction of the environmental flow vector, and performing consecutive frame intersection analysis with the spatial range of the inspection target, a correspondence is established between the weak methane plume signal target and the inspection target identifier based on spectral response, temporal continuity, and spatial origin. Thus, the detection results can be further broken down from anomalous pixels or anomalous patches to specific inspection targets such as valve groups, flanges, wellhead devices, tank interfaces, and pipeline interfaces.

[0020] By extracting the concave response, baseline response, and residual response from the weak absorption feature vector, the methane absorption feature band, the adjacent non-absorption reference band, and the local background baseline spectrum of the same frame are all involved in the discrimination, thus weakening the influence of single-band brightness changes on the generation of candidate methane response regions.

[0021] By connecting response regions with the same absorption spectrum, adjacent spatial locations, and consecutive acquisition times in consecutive frames, short-term background disturbances are unlikely to form response trajectories that meet the temporal continuity conditions, thus preserving the temporal continuity of weak methane plume targets.

[0022] By writing the methane absorption spectrum correspondence results, local background baseline spectrum in the same frame, weak absorption feature vector, interference response marker, response trajectory, environmental flow vector and spatial range of the inspected object into the target identification record, the inspection and verification can be checked item by item according to the spectral basis, background reference basis, interference elimination basis and object correspondence basis. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0024] Figure 1 This is a flowchart of the weak signal target recognition method that integrates prior spectral information in the embodiments. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0026] All terms used in this application (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0027] To facilitate understanding, the terminology used in the embodiments of this application will be introduced first: Oil and gas inspection scenarios refer to inspection areas that include valve groups, flanges, wellhead equipment, compressors, tank interfaces, pipeline interfaces, or platform process equipment. UAV-borne spectral sequences refer to the set of spectral images continuously acquired by a UAV-borne spectral imaging device at each acquisition time. UAV pose information refers to the record of the UAV's position, attitude, and heading when acquiring each frame of spectral image. Environmental flow vectors refer to directional data characterizing the direction of gas diffusion within the inspection area, sourced from UAV-borne wind speed and direction sensors, station meteorological equipment, or local wind field records from the platform. Inspection object spatial records refer to the spatial data of the inspection object in the coordinate system of the inspection area, including at least the inspection object identifier, inspection object type, inspection object spatial range, and inspection object coordinates.

[0028] A low-concentration methane plume refers to a methane plume whose absorption response amplitude is close to the spectral fluctuation range of a leak-free frame in the same background type, and whose spatial morphology changes with the environmental flow direction in continuous frames. A weak signal target methane plume refers to a low-concentration methane plume that, after being jointly determined by the weak absorption feature vector, interference response marker, and response trajectory, corresponds to the inspection object identifier. The inspection object identifier is a data field used to uniquely characterize valve groups, flanges, wellhead devices, compressors, tank interfaces, pipeline interfaces, or platform process equipment. The spatial range of the inspection object refers to the two-dimensional outer contour range or three-dimensional projection range of the inspection object in the coordinate system of the inspection area; its fields include the inspection object identifier, inspection object type, outer contour boundary, coordinate position, and spatial range version identifier.

[0029] The UAV-borne spectral sequence, UAV pose information, environmental flow vector, and spatial record of the inspected object are stored under the same inspection task number. Each frame of the spectral image in the UAV-borne spectral sequence records the acquisition time, and a temporal correspondence is established between the acquisition time and the UAV pose information and environmental flow vector. The spatial record of the inspected object and the UAV-borne spectral sequence are spatially corresponded through the coordinate system of the inspection area.

[0030] The prior spectral information set refers to the set of spectral reference data established before identification and invoked during the identification process. Methane absorption prior spectrum refers to the spectral shape data used to characterize methane in the methane absorption characteristic band and adjacent non-absorption reference bands. Background prior spectrum refers to the spectral response data used to characterize the equipment surface, ground, water surface, platform deck, and shaded areas within the inspection area. Interference prior spectrum refers to the non-methane spectral response data used to characterize water vapor, cloud shadows, heat source reflection, metal reflection, and shadows. Background type refers to the classification label formed based on the background prior spectrum to determine the scene material or area attributes to which the pixel belongs. Broadband reflectance variation refers to the overall reflectance variation state of the pixel to be identified in multiple non-methane absorption bands covered by the UAV-borne spectral sequence. This overall reflectance variation state, together with the response of adjacent non-absorption reference bands, is used to distinguish metal pipeline surfaces, valve housings, insulation layers, ground, water surface, platform deck, and shaded areas. The background type labeling result is written into a data record composed of the pixel location, frame number, and background type identifier, and a corresponding relationship is established with the version identifier of the currently used prior spectral information set.

[0031] The methane absorption characteristic bands refer to the set of bands in the prior methane absorption spectrum that exhibit a methane absorption response. The adjacent non-absorption reference bands refer to the set of bands located in the adjacent region of the methane absorption characteristic bands that do not exhibit a methane absorption response and are used to form a baseline response. The spectral template refers to spectral reference data formed by band alignment, response normalization, and sensor response correction of methane standard absorption spectra, historical inspection spectral data, on-site leak-free frame spectral data, or interfering sample spectral data. The methane standard absorption spectrum refers to the methane absorption spectral shape data recorded in standard spectral libraries, experimental calibration data, or sensor manufacturer calibration data, and resampled according to the band order of the UAV-borne spectral sequence.

[0032] The data sources for the prior spectral information set include standard methane absorption spectra, historical inspection spectral data, on-site leak-free frame spectral data, and interference sample spectral data. When establishing the prior spectral information set, band alignment, response normalization, and sensor response correction are performed on spectral data from different sources to ensure that the methane absorption prior spectrum, background prior spectrum, interference prior spectrum, and UAV-borne spectral sequence have a consistent band order and response aperture.

[0033] Example 1: like Figure 1 As shown, this embodiment employs a weak signal target recognition method that integrates prior spectral information, applied to UAVs for spectral identification of low-concentration methane plumes in oil and gas inspection scenarios, including the following steps: Step S1: Obtain the UAV-borne spectral sequence, UAV pose information, environmental flow vector, and spatial record of the inspection object in the inspection area. The UAV-borne spectral sequence is a set of spectral images arranged according to the acquisition time and covering the methane absorption characteristic band and the adjacent non-absorption reference band. The spatial record of the inspection object includes the inspection object identifier and the spatial range of the inspection object. Step S2: Establish a priori spectral information set, which includes methane absorption prior spectrum, background prior spectrum and interfering prior spectrum; Step S3: In the same frame of spectral image, the background type is labeled according to the prior background spectrum. A spatial neighborhood is set with the pixel to be identified as the center. Pixels in the spatial neighborhood that are in the same frame as the pixel to be identified, are different from the pixel to be identified, have the same background type, do not show methane absorption spectrum, and meet similarity constraints in the nearby non-absorption reference band are selected as candidate background pixels. When the number of candidate background pixels is lower than the background baseline number threshold, candidate background pixels are supplemented along the extended neighborhood adjacent to the pixel to be identified and having the same background type. The spectral response of the candidate background pixels forms the local background baseline spectrum of the pixel to be identified in the same frame. Step S4: Generate a weak absorption feature vector for the pixel to be identified based on the prior methane absorption spectrum, the adjacent non-absorption reference band, and the local background baseline spectrum in the same frame. Step S5: Generate candidate methane response regions based on the spectral shape correspondence between the weak absorption feature vector and the prior methane absorption spectrum. Step S6: Generate interference response markers for the candidate methane response regions based on the interference prior spectrum, and filter out regions with interference response markers; Step S7: The filtered candidate methane response regions are used to form response trajectories in consecutive frames. The response trajectories are projected onto the inspection area coordinate system according to the UAV pose information. A backtracking range is formed along the opposite direction of the environmental flow vector at the corresponding acquisition time. The intersection of the backtracking range and the spatial range of the inspection object is judged. When the intersection result in consecutive frames meets the continuous hit condition for the same inspection object identifier, the response trajectory is determined as the weak signal target of the methane plume of the corresponding inspection object identifier. Step S8: Output the inspection object identifier and target identification record corresponding to the weak methane plume signal target; In one implementation, a candidate background pixel refers to a pixel that is in the same frame of the spectral image as the pixel to be identified, is different from the pixel to be identified, has the same background type, does not exhibit a methane absorption spectrum, and satisfies similarity constraints. The background baseline quantity threshold is the lower limit of the number of candidate background pixels required to form a local background baseline spectrum within the same frame. The spatial neighborhood refers to the pixel range within the same frame of the spectral image centered on the pixel to be identified, used for initial screening of candidate background pixels. The extended neighborhood refers to the pixel range within the same frame formed by extending along connected regions of the same background type when the number of candidate background pixels is lower than the background baseline quantity threshold. The similarity constraint means that the similarity difference between the candidate pixel and the pixel to be identified in the adjacent non-absorption reference bands is not greater than the reference band similarity difference threshold. The methane absorption spectrum refers to a spectral shape that exhibits an absorption dip corresponding to the prior methane absorption spectrum in the methane absorption characteristic band and maintains baseline continuity in the adjacent non-absorption reference bands.

[0034] When forming the local background baseline spectrum within the same frame, a spatial neighborhood is set centered on the pixel to be identified in the same frame of spectral image. Within this spatial neighborhood, pixels with the same background type as the pixel to be identified, which do not exhibit methane absorption spectral shape, and satisfy the similarity constraint of the adjacent non-absorbing reference band are selected as candidate background pixels. When the number of candidate background pixels is not less than the background baseline number threshold, the spectral responses of the candidate background pixels are weighted and synthesized or robustly aggregated to obtain the local background baseline spectrum within the same frame that is consistent with the background type of the pixel to be identified. When the number of candidate background pixels is less than the background baseline number threshold, the selection range is expanded along the connected regions adjacent to the pixel to be identified and having the same background type. The expansion process does not cross obvious background type boundaries, spatial range boundaries of the inspected object, or spectral saturation regions. The local background baseline spectrum within the same frame is formed by the candidate background pixels that still satisfy the non-methane absorption and reference band similarity constraints after expansion.

[0035] In one alternative embodiment, the methane absorption prior spectrum includes the absorption position, absorption width, absorption spectral shape of the methane absorption characteristic band, and the baseline spectral shape of the adjacent non-absorption reference band. When generating weak absorption feature vectors, the concave response of the pixel to be identified in the methane absorption feature band, the baseline response in the adjacent non-absorption reference band, and the residual response relative to the local background baseline spectrum of the same frame are extracted respectively. A weak absorption feature vector is a data record composed of the methane absorption feature band response, the response of the adjacent non-absorbing reference band, and the local background difference response within the same frame for the pixel to be identified. The concave response characterizes the decrease in methane absorption feature band relative to the adjacent non-absorbing reference band for the pixel to be identified. The baseline response characterizes the background reflectance level formed by the pixel to be identified in the adjacent non-absorbing reference band. The residual response characterizes the spectral difference between the pixel to be identified and the local background baseline spectrum within the same frame. All data in the weak absorption feature vector are generated using the same frame's spectral image, the same pixel to be identified, and the same local background baseline spectrum within the same frame. The methane absorption feature band response, the response of the adjacent non-absorbing reference band, and the local background difference response within the same frame are written into the concave response field, the baseline response field, and the residual response field, respectively. For the weak absorption feature vector to meet the consistency requirement, the concave direction of the methane absorption feature band must be consistent across consecutive frames or connected regions, the baseline variation of the adjacent non-absorbing reference band must be continuous, and the direction of variation of the local background residual response within the same frame must be consistent.

[0036] When generating weak absorption feature vectors, the concave response, baseline response, and residual response of the same pixel to be identified are written into the same data record according to the pixel location, frame number, and acquisition time. This data record is used for subsequent generation of candidate methane response regions, generation of interference response markers, and connection of response trajectories.

[0037] In one alternative implementation, the background prior spectrum includes spectral templates for at least two types of backgrounds, including metal pipeline surfaces, valve housings, insulation layers, ground, water surfaces, platform decks, and shaded areas. The background type is determined by the response of the neighboring non-absorbing reference band and the broadband reflection variation of the pixel to be identified. In one alternative implementation, generating the candidate methane response region includes: Connectivity aggregation is performed on the weakly absorbing feature vectors of adjacent pixels; Preserve connected regions with methane absorption characteristic band concave response, baseline continuity of adjacent non-absorbing reference bands, and consistency of local background residuals within the same frame; The connected regions are used as candidate methane response regions; Spectral shape correspondence refers to the correspondence between the spectral changes of the pixel or connected region to be identified and the prior methane absorption spectrum in terms of absorption position, absorption direction, and the trend of changes in adjacent reference bands. A connected region is a set of spatially adjacent pixels in the same frame of a spectral image whose weak absorption feature vectors satisfy the same discrimination condition. A candidate methane response region is a set of pixels formed by connected regions that satisfies the methane absorption spectral shape correspondence. When generating candidate methane response regions, spectral shape correspondence is first determined for the weak absorption feature vectors of individual pixels. Then, spatially adjacent pixels are connected and aggregated. A region that exhibits a concave response in the same direction in the methane absorption feature band, a continuous baseline response in the adjacent non-absorption reference band, and a consistent residual response is considered a candidate methane response region.

[0038] In one alternative implementation, the interfering prior spectrum includes at least one interfering spectral template from water vapor, cloud shadows, heat source reflections, metallic reflections, and shadows. When the weak absorption feature vector of the candidate methane response region forms a spectral correspondence with the interference prior spectrum, and the residual response of the candidate methane response region relative to the local background baseline spectrum of the same frame conforms to the residual change direction of the corresponding interference, an interference response label is generated for the candidate methane response region. Interference response markers are data fields that record the correspondence between candidate methane response regions and interfering prior spectra. They include at least the candidate methane response region number, interference type, interference spectral shape correspondence result, residual change direction, and marker time. The residual change direction refers to the direction of change of the candidate methane response region relative to the local background baseline spectrum in the same frame within the methane absorption characteristic band and the adjacent non-absorption reference band. When generating interference response markers, candidate methane response regions are first compared with the methane absorption prior spectrum for spectral shape correspondence determination, and then compared with the interfering prior spectrum for reverse determination. When a candidate methane response region simultaneously exhibits both the response change near the methane absorption position and the residual change direction corresponding to the interfering prior spectrum, the candidate methane response region is marked with an interference response marker.

[0039] In one alternative implementation, determining the response trajectory includes: Spatial registration of consecutive frame spectral images based on UAV pose information; The candidate methane response regions searched and filtered in the registered consecutive frames have the same absorption spectrum, adjacent spatial positions and consecutive acquisition times. Connect the searched response regions according to the acquisition time to form a response trajectory; In one optional implementation, the target identification record includes the acquisition time, UAV pose, methane absorption spectrum correspondence result, local background baseline spectrum in the same frame, weak absorption feature vector, interference response marker, response trajectory, environmental flow vector, inspection object identifier, inspection object spatial range, object corresponding status, multi-object conflict marker, candidate inspection object identifier, and corresponding intersection result. Target identification records refer to data records used to save the process and results of identifying weak methane plume targets. These records use the inspection task number, frame number, and inspection object identifier as index fields, and store the methane absorption spectrum correspondence results, the local background baseline spectrum of the same frame, the weak absorption feature vector, interference response markers, response trajectories, environmental flow vectors, and the spatial range of the inspection object. The object correspondence status refers to the correspondence determination status between the response trajectory and the inspection object identifier, with values ​​including "corresponded," "pending confirmation," and "multiple object conflict." The multi-object conflict marker is a status field written when the backtracking range simultaneously points to multiple inspection object identifiers and cannot distinguish a unique inspection object identifier. The candidate inspection object identifier refers to the set of inspection object identifiers that intersect with the backtracking range in the multi-object conflict state. The corresponding intersection results refer to the number of intersections, intersection ratio, lateral offset distance, and continuous record of acquisition time between the backtracking range and the spatial range of the inspection object.

[0040] In one optional implementation, the prior spectral information set is established based on the historical spectral data of the inspection area, the on-site leak-free frame spectral data, and the standard absorption spectrum of methane; during the acquisition of the UAV-borne spectral sequence, the background prior spectrum is updated using the on-site leak-free frame spectral data, and the updated background prior spectrum is used to redetermine the local background baseline spectrum of the same frame. Leak-free frame spectral data refers to spectral image data in which no weak methane plume signal targets were formed and no interference response markers were generated during the inspection process. Leak-free frame spectral data is collected after candidate methane response region generation, interference response marker generation, and weak methane plume signal target identification are completed in the current frame, and participates in background prior spectrum updates starting from subsequent frames. After the background prior spectrum is updated, the background type labeling, candidate background pixel selection, and local background baseline spectrum formation processes in subsequent frames within the same inspection task all use the updated background prior spectrum.

[0041] In a preferred embodiment of Example 1, when forming the local background baseline spectrum of the same frame from candidate background pixels, for the first... Pixels to be identified in a frame spectral image Within the same frame where the background type has been labeled, candidate background pixel constraints are performed. The prior background spectra used for background type labeling, candidate background pixel selection, reference band similarity difference threshold reading, and local background baseline spectrum generation within the same frame all use the same prior spectral information set version identifier. The prior version identifier is only used to record the version of the prior spectral information set used in the calculation of the same frame. When no leaked frame spectral data triggers the background prior spectrum update, the updated version of the prior spectral information set participates in the above calculations starting from the next frame, without replacing the local background baseline spectrum already generated in the current frame.

[0042] Candidate background pixels were selected in the order of consistent background type, spatial proximity, non-methane absorption, and similar reference band. Backgrounds of different materials, suspected methane absorption pixels, and pixels with abrupt changes in local reflection were not included in the candidate background pixel set.

[0043] The method for calculating the similarity difference of reference bands is as follows:

[0044] , in, This indicates the frame number in the UAV-borne spectral sequence. Indicates the first Candidate pixels in frame With the pixel to be identified The similarity difference in the adjacent non-absorbing reference band; Indicates the candidate pixels participating in the screening; Indicates the pixel to be identified; This represents the set of nearby non-absorbing reference bands; This indicates the number of bands in the set of neighboring non-absorbing reference bands; Indicates spectral band; Indicates the first Candidate pixels in frame In the band The spectral response; Indicates the first Pixels to be identified in the frame In the band The spectral response; Indicates the first Pixels to be identified in the frame When the average spectral response in the adjacent non-absorbing reference band is lower than the lower limit of the denominator calibrated by the sensor dark current level and the noise floor of the field-leaked frame spectral data, the lower limit of the denominator is used in the calculation of the similarity difference.

[0045] Based on background type consistency, spatial proximity, non-methane absorption constraints, and reference band similarity constraints, a candidate background pixel set is formed for background baseline calculation: , in, Indicates the first Pixels to be identified in the frame The corresponding set of candidate background pixels; Indicates the first Pixels to be identified in the frame Centered on and with radius Spatial neighborhood; Indicates the first The spatial neighborhood radius in the frame; Indicates the first Candidate pixels in frame Background type; Indicates the first Pixels to be identified in the frame Background type; Indicates the first Candidate pixels in frame The correspondence between the methane absorption spectrum and the prior methane absorption spectrum; the larger the correspondence, the closer the candidate pixel is to the methane absorption spectrum. This indicates the methane absorption spectrum rejection threshold; This represents the reference band similarity difference threshold; the remaining parameters follow the previous definitions.

[0046] The methane absorption spectrum rejection threshold is jointly calibrated using standard methane absorption spectra and leak-free frame spectral data from the field. The value is set such that only pixels with methane absorption spectrum values ​​not exceeding this threshold are allowed to enter the candidate background pixel set. The reference band similarity difference threshold is calibrated using leak-free frame spectral data from the field of the same background type. The value is set such that only pixels with reference band differences not exceeding this threshold are allowed to enter the candidate background pixel set.

[0047] The methane absorption spectrum removal threshold and the reference band similarity difference threshold are both bound to the background type of the pixel to be identified and the version of the prior spectral information set currently used; the same threshold is used for the same pixel in the process of candidate background pixel screening, supplementation and local background baseline spectrum synthesis in the same frame.

[0048] When the number of candidate background pixels is lower than the background baseline number threshold, neighboring regions are expanded within the same frame along connected regions of the same background type. The expansion process does not cross the spatial boundaries of the inspected object, obvious background type boundaries, or spectral saturation regions. A spectral saturation region refers to a pixel region where the spectral response reaches the upper limit of the effective response of the spectral imaging device. If the expansion radius is still insufficient after reaching a preset upper limit, the background prior spectrum of the leak-free frame spectral data of the same background type, after amplitude matching with the adjacent non-absorbing reference band of the current frame, is used as a conservative background baseline, and the insufficient background baseline is marked in the target identification record. A conservative background baseline is an alternative background baseline formed when the number of candidate background pixels is lower than the background baseline number threshold. Amplitude matching backoff refers to the process of using the unscaled background prior spectrum as a conservative background baseline and marking it as such when the reference amplitude is abnormal, the target amplitude is abnormal, or the similarity difference threshold of the reference band is still not met after scaling.

[0049] Amplitude matching refers to scaling the background prior spectrum as follows: the average spectral response of the pixel to be identified in the current frame in the neighboring non-absorption reference band is used as the target amplitude, and the average spectral response of the field leak-free frame spectral data of the same background type in the neighboring non-absorption reference band is used as the reference amplitude. The scaling factor is obtained by the ratio of the target amplitude to the reference amplitude and is limited to the lower and upper scaling limits calibrated by the reflection fluctuation range of the field leak-free frame spectral data of the same background type. When the reference amplitude is lower than the lower limit of the denominator, the pixel to be identified corresponding to the target amplitude is in the spectral saturation region, or the scaling factor reaches the lower or upper scaling limit but still cannot make the difference in the neighboring non-absorption reference band fall into the similarity difference threshold of the reference band, no scaling is performed, and the unscaled background prior spectrum is used as a conservative background baseline. At the same time, amplitude matching backtracking is marked in the target identification record.

[0050] When weighting candidate background pixels, pixels that are spatially closer and have smaller differences in reference bands receive a greater background contribution weight. The weighting method is as follows: , in, Indicates the first Candidate pixels in frame pixels to be identified Background contribution weight; Indicates the first In-frame summation index cell With the pixel to be identified The similarity difference in the adjacent non-absorbing reference band; Indicates the first Candidate pixels in frame With the pixel to be identified Spatial distance between them; This represents the spatial distance attenuation scale and is a positive value. This represents the attenuation scale of similarity differences in reference bands, and is taken as a positive value; This represents the summation index cell in the candidate background cell set; Indicates the first In-frame summation index cell With the pixel to be identified The spatial distance between them; the remaining parameters follow the definitions above. The spatial distance attenuation scale is calibrated by the ground sampling interval of the UAV-borne spectral image and the minimum resolvable width of the inspected object, and the reference band similarity difference attenuation scale is calibrated by the reference band fluctuation range of the on-site leak-free frame spectral data of the same background type.

[0051] The number of candidate background pixels participating in weighted synthesis is not less than the background baseline number threshold. The weight concentration threshold is a threshold used to determine whether the contribution of a single candidate background pixel to the local background baseline spectral data of the same frame is excessively concentrated. Its value is determined by the background contribution weight distribution in the spectral data of the same background type without leakage in the field. If the normalized weight of a single background contribution exceeds the weight concentration threshold, the candidate background pixel set is empty, or the weight normalization denominator is abnormal, conservative background baseline processing is applied. The weights are non-negative after normalization and their sum is 1.

[0052] The spectral responses of candidate background pixels are weighted and synthesized using normalized weights to obtain the local background baseline spectrum of the pixel to be identified within the same frame: , in, Indicates the first Pixels to be identified in the frame In the band The local background baseline spectrum of the same frame is used; the remaining parameters follow the previous definition. This local background baseline spectrum of the same frame is formed only by pixels in the current frame that have the same background type as the pixel to be identified, are spatially adjacent, and do not exhibit a methane absorption spectrum. It is used for the calculation of the residual response in the subsequent weak absorption feature vector.

[0053] Example 2: Based on Example 1, this example provides a specific method for spatially backtracking the response trajectory along the opposite direction of the environmental flow vector in a weak signal target recognition method that integrates prior spectral information. The response trajectory refers to the connection result of candidate methane response regions in consecutive frames that meet the trajectory connection threshold, are spatially adjacent, have continuous acquisition time, and have consistent weak absorption feature vectors. When performing spatial backtracking, the response trajectory is projected onto the coordinate system of the inspection area based on the UAV pose information. The backtracking range is formed by taking the position of the trajectory node in the response trajectory as the starting point and moving in the opposite direction of the environmental flow vector at the corresponding acquisition time. The backtracking range is used to characterize the possible source area of ​​the weak signal target of the methane plume. In one implementation, the intersection of the backtracking range with the spatial range of the inspected object in the spatial record of the inspected object is determined, and whether the response trajectory points to the same inspected object identifier is determined based on the number of intersections, the intersection ratio and the continuity of the collection time in consecutive frames. When the backtracking range in consecutive frames meets the condition of continuous hits on the same inspection object identifier, a correspondence is established between the response trajectory and the inspection object identifier; when the backtracking range points to multiple inspection object identifiers at the same time, conflict resolution is performed based on the number of intersections, intersection ratio, lateral offset distance and collection time continuity of each inspection object; when it is impossible to distinguish a unique inspection object identifier, multi-object conflict markers, candidate inspection object identifiers and corresponding intersection results are output in the target identification record. The trajectory connectivity threshold is a threshold used to determine whether candidate methane response regions in consecutive frames belong to the same response trajectory. Its inputs include the methane absorption spectral shape correspondence, spatial distance, acquisition time interval, and consistency result of the weak absorption feature vector. The low-concentration methane release calibration frame refers to a spectral image frame acquired and labeled with the methane plume position under known low-concentration methane release conditions. The trajectory connectivity threshold is jointly tuned by the spectral data from the low-concentration methane release calibration frame and the on-site leak-free frame. A trajectory node is a record of the candidate methane response region position corresponding to a single acquisition time within the response trajectory. Its fields include the acquisition time, candidate methane response region number, region center position, weak absorption feature vector, and methane absorption spectral shape correspondence result.

[0054] In a preferred embodiment of Example 2, when spatially backtracking the response trajectory in the opposite direction of the environmental flow vector, the response trajectory of the filtered candidate methane response region in consecutive frames is projected onto the inspection area coordinate system, and trajectory nodes are obtained according to the acquisition time.

[0055] Each trajectory node is calculated using the environmental flow vector at the corresponding acquisition time for reverse pointing; when synchronous data is missing, temporal interpolation is performed using the environmental flow vectors within adjacent consecutive frames. The effective flow velocity threshold is an implementation parameter used to determine whether the environmental flow vector participates in spatial backtracking, measured in meters per second. When the amplitude of the environmental flow vector is lower than the effective flow velocity threshold, the trajectory node does not form a separate inspection object correspondence and participates in target identification recording as a node to be confirmed. When the environmental flow vectors of adjacent consecutive frames are missing, the corresponding abnormal status is written into the target identification record.

[0056] When the number of trajectory nodes that meet the effective flow velocity threshold within a consecutive frame window is less than the hit count threshold, no correspondence between inspection objects is established. Only the intersection information of the response trajectory, low flow velocity marker, and spatial range of candidate inspection objects is retained in the target identification record.

[0057] In the When the magnitude of the environmental flow vector corresponding to a trajectory node is not lower than the effective flow velocity threshold, the reverse unit backtracking direction is determined according to the opposite direction of the environmental flow vector, and the calculation method is as follows: , in, Indicates the first The reverse unit backtracking direction of each trajectory node; This indicates the sequence number of the trajectory nodes in the response trajectory, sorted by the acquisition time. Indicates the first Each trajectory node corresponds to the environmental flow direction vector at the time of acquisition; Indicates the first Each trajectory node corresponds to the magnitude of the environmental flow direction vector at the time of acquisition.

[0058] The backtracking length is calibrated based on the ambient velocity amplitude, trajectory duration, and UAV spatial registration error. The backtracking length value does not exceed the effective observation range of the inspection area boundary and the inspection segment corresponding to the current UAV-borne spectral sequence.

[0059] The backtracking range width increases with the increase of spatial registration uncertainty and flow direction fluctuation. Its update is performed once on the trajectory node that meets the effective flow velocity threshold, and is limited by the lower limit and the upper limit of the backtracking range width. The lower limit of the backtracking range width is not greater than the upper limit of the backtracking range width. When the spatial registration uncertainty or flow direction fluctuation exceeds the corresponding abnormal threshold, the backtracking range width converges to the upper limit of the backtracking range width according to the calculation result, and the hit count threshold is adjusted to a value that is not lower than the hit count threshold under normal working conditions and does not exceed the number of trajectory nodes contained in the continuous frame window.

[0060] The calculation method for the backtracking range width is as follows: , in, Indicates the first The backtracking range of each trajectory node is wide; Indicates the upper limit of the backtracking range; This indicates the lower limit of the backtracking range; This indicates a wide range of basic backtracking; This represents the amplification factor of spatial registration uncertainty on the wide backtracking range, and takes a non-negative value; Indicates the first Spatial registration uncertainty corresponding to each trajectory node; This represents the amplification factor of the flow direction fluctuation on the wide range of backtracking, and takes a non-negative value; Indicates the first The flow direction fluctuation within the time neighborhood of each trajectory node.

[0061] The basic backtracking range is wide, determined by the ground sampling interval of the UAV-borne spectral image, the boundary error of the spatial range of the inspected object, the background disturbance range in the spectral data of the leak-free frame, and the lateral disturbance range of the plume in the low-concentration methane release calibration frame. Spatial registration uncertainty is given by the spatial registration residual of consecutive frames, and the flow direction fluctuation is given by the change in the direction of the environmental flow direction vector in adjacent consecutive frames. When the environmental flow direction vector is continuously missing within a consecutive frame window, no correspondence between inspected objects is established; only the response trajectory and abnormal flow direction markers are retained.

[0062] For the first one that meets the effective flow velocity threshold The trajectory nodes and their corresponding backtracking ranges are represented as follows: , in, Indicates the first The backtracking range corresponding to each trajectory node; Indicates the spatial location of the inspection area in the coordinate system; This indicates the effective observation area in the coordinate system of the inspection area; Indicates the first The spatial position of each trajectory node in the coordinate system of the inspection area; Represents the dot product of vectors; Indicates the first The backtracking length of each trajectory node; The aforementioned definition will be used; The aforementioned definition will be used; Represents the vector norm.

[0063] The intersection of the backtracking range of each trajectory node that meets the effective flow velocity threshold with the spatial range of the inspected object is determined. The intersection ratio is used to represent the degree of hit of the backtracking range with the spatial range of the inspected object. The calculation method is as follows: , in, Indicates the first The backtracking range of the first trajectory node and the second The proportion of intersection between the spatial ranges of each inspection object; Indicates the serial number of the object being inspected; Indicates the first The spatial range of each inspection target; Indicates the first The backtracking range of the first trajectory node and the second The area of ​​intersection of the spatial ranges of each inspection object; Indicates the first The area of ​​the backtracking range of each trajectory node; Indicates the first The area of ​​the space encompassing each inspection target; Represents the smaller of the two areas; area represents the area operator; when or When the value is zero, the intersection ratio between the trajectory node and the inspected object is not calculated, and the intersection judgment is recorded as an invalid hit. At the same time, the spatial range is marked as abnormal in the target recognition record.

[0064] The intersection ratio threshold is determined by the positioning error of the spatial range of the inspected object, the spatial registration error of the UAV, and the statistical results of false hits in the leak-free frame spectral data. When the intersection ratio is not less than the intersection ratio threshold, it is recorded as a hit of the trajectory node to the corresponding inspected object; when it is less than the intersection ratio threshold, it is not counted as an object hit.

[0065] Single-frame intersection determination does not establish a separate correspondence between inspected objects. Instead, continuous-frame cumulative hit determination is used to establish the correspondence between the response trajectory and the inspected object identifier. , in, Indicates the first The maximum cumulative number of hits for an inspection object within a consecutive frame window; This represents the starting trajectory node number of a continuous frame window, and its value is such that from the first... The trajectory node to the first All trajectory nodes belong to the same response trajectory; Indicates the number of trajectory nodes contained in a consecutive frame window; The indicator function representing the intersection ratio determination, when the first... The value is 1 when the intersection ratio of the trajectory nodes is not less than the intersection ratio threshold, and 0 otherwise. This represents the intersection ratio threshold; the remaining parameters follow the previous definitions. For trajectory nodes where the intersection ratio is not calculated or for intersection determination, the indicator function is set to zero.

[0066] A continuous frame window refers to the set of trajectory nodes in the response trajectory arranged continuously according to the acquisition time, used to count the number of hits on an inspected object. The hit count threshold is an implementation parameter used in the continuous frame cumulative hit determination to determine whether the same inspected object identifier is continuously pointed to. The number of trajectory nodes included in the continuous frame window is 5 when no scene tuning result is formed, with an adjustment range of 3~15; the hit count threshold is 3 when no scene tuning result is formed, with a value not less than 2 and not greater than the number of trajectory nodes included in the continuous frame window. The continuous hit condition means that within the continuous frame window, the effective hit count corresponding to the spatial range of the same inspected object is not less than the hit count threshold, and the methane absorption spectrum of the response trajectory remains consistent within the continuous frame window.

[0067] Trajectory nodes with flow rates below the effective flow rate threshold, trajectory nodes with a backtracking range area of ​​zero, and intersection determinations with an inspection object spatial range area of ​​zero are not counted as valid hits. The aforementioned trajectory nodes or intersection determinations can be retained in consecutive frame windows for time continuity recording, but this does not increase the maximum cumulative hit count.

[0068] When the maximum cumulative hit count of a certain inspection object within a consecutive frame window is not less than the hit count threshold When the methane absorption spectrum of the response trajectory remains consistent within the continuous frame window, the response trajectory is determined to point to the spatial range of the inspection target. The hit count threshold is jointly determined by the number of false backtracking hits in the on-site leak-free frame spectral data and the number of valid hits in the low-concentration methane release calibration frame; the value direction is higher than the upper limit of false hits in the on-site leak-free frame spectral data but not higher than the lower limit of valid continuous leak hits. When the upper limit of false backtracking hits in the on-site leak-free frame is not lower than the lower limit of valid hits in the low-concentration methane release calibration frame, the hit count threshold calibration interval is determined to be invalid, a single inspection target identifier is not automatically output, and a threshold calibration failure mark and a candidate inspection target identifier are output in the target identification record.

[0069] The backtracking range centerline refers to the central reference line extending along the backtracking range in the opposite direction to the environmental flow vector. If multiple inspection objects meet the continuous frame cumulative hit determination, the average intersection ratio of each inspection object within the hit trajectory node is compared, and conflict resolution is performed by combining the lateral offset distance between the inspection object and the backtracking range centerline. When the multi-object conflict condition is not met, the inspection object with a larger average intersection ratio and a smaller lateral offset distance is identified as the corresponding inspection object. The preset conflict interval threshold is used to determine whether the difference in the average intersection ratio is sufficient to distinguish the inspection objects. It is calibrated by converting the minimum spacing of the spatial range of adjacent inspection objects within the same inspection segment, the UAV spatial registration error, and the difference in the error intersection ratio in the on-site leak-free frame spectral data to the intersection ratio difference caliber. If the average intersection ratio sorting result is inconsistent with the horizontal offset distance sorting result from smallest to largest, and the difference between the highest average intersection ratio and the second highest average intersection ratio is lower than the preset conflict interval threshold, then it is treated as a multi-object conflict and no single inspection object correspondence is established; when the spatial ranges of two inspection objects intersect with the same backtracking range and cannot be distinguished by the stability of continuous frame pointing, it is also treated as a multi-object conflict, and the multi-object conflict mark, candidate inspection object identifier and corresponding intersection ratio are output in the target recognition record.

[0070] The lateral offset distance is the average vertical distance from the geometric center of the inspected object's spatial range to the center line of the corresponding backtracking range within the hit trajectory node. Continuous frame pointing stability is characterized by the continuity of hits and the fluctuation range of the intersection ratio for the same inspected object within the continuous frame window; fewer hit interruptions and smaller fluctuation range of the intersection ratio indicate higher continuous frame pointing stability.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0072] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments. For example, all the embodiments above can be used in any combination. The information disclosed in this background section is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

Claims

1. A weak signal target recognition method integrating prior spectral information, applied to the spectral recognition of low-concentration methane plumes in oil and gas inspection scenarios by unmanned aerial vehicles (UAVs), characterized in that... Includes the following steps: The UAV-borne spectral sequence, UAV pose information, environmental flow vector, and spatial record of the inspected object are acquired in the inspection area. The UAV-borne spectral sequence covers the methane absorption characteristic band and the adjacent non-absorption reference band. The spatial record of the inspected object includes the inspected object identifier and the spatial range of the inspected object. A priori spectral information set is established, which includes methane absorption prior spectrum, background prior spectrum and interfering prior spectrum; Based on the background prior spectrum, the background type is marked in the same frame of spectral image, and candidate background pixels are determined from pixels that are in the same frame as the pixel to be identified, are different from the pixel to be identified, have the same background type, do not show methane absorption spectrum, and satisfy similar constraints in the adjacent non-absorption reference band. The candidate background pixels form the local background baseline spectrum of the same frame. Based on the methane absorption prior spectrum, the adjacent non-absorption reference band, and the local background baseline spectrum of the same frame, a weak absorption feature vector of the pixel to be identified is generated. Candidate methane response regions are generated based on the spectral shape correspondence between the weak absorption feature vector and the methane absorption prior spectrum. Interference response markers are generated based on the interference prior spectrum, and candidate methane response regions with interference response markers are filtered out. The filtered candidate methane response regions are used to form response trajectories in consecutive frames. The response trajectories are projected onto the inspection area coordinate system based on the UAV pose information. A backtracking range is formed along the opposite direction of the environmental flow vector at the corresponding acquisition time. The intersection of the backtracking range and the spatial range of the inspection object is judged. When the intersection result in consecutive frames meets the continuous hit condition for the same inspection object identifier, the response trajectory is determined as the weak signal target of the methane plume corresponding to the inspection object identifier. Output the inspection object identifier and target identification record corresponding to the weak methane plume signal target.

2. The weak signal target recognition method fused with prior spectral information according to claim 1, characterized in that, The methane absorption prior spectrum includes the absorption position, absorption width, absorption spectral shape of the methane absorption characteristic band, and the baseline spectral shape of the adjacent non-absorption reference band; When generating the weak absorption feature vector, the concave response of the pixel to be identified in the methane absorption feature band, the baseline response in the adjacent non-absorption reference band, and the residual response relative to the local background baseline spectrum of the same frame are extracted respectively.

3. The weak signal target recognition method according to claim 1, characterized in that, The background prior spectrum includes spectral templates for at least two background types, including metal pipeline surfaces, valve housings, insulation layers, ground, water surfaces, platform decks, and shaded areas. The background type is determined by the response of the neighboring non-absorbing reference band and the broadband reflection variation of the pixel to be identified.

4. The weak signal target recognition method fused with prior spectral information according to claim 1, characterized in that, Generating the candidate methane response region includes: Connectivity aggregation is performed on the weakly absorbing feature vectors of adjacent pixels; Preserve connected regions with methane absorption characteristic band concave response, baseline continuity of adjacent non-absorbing reference bands, and consistency of local background residuals within the same frame; The connected region is used as a candidate methane response region.

5. The weak signal target recognition method according to claim 1, characterized in that, The interference prior spectrum includes at least one spectral template from water vapor, cloud shadows, heat source reflection, metallic reflection, and shadows. When the weak absorption feature vector of the candidate methane response region forms a spectral correspondence with the interference prior spectrum, and the residual response of the candidate methane response region relative to the local background baseline spectrum of the same frame conforms to the residual change direction of the corresponding interference, an interference response label is generated for the candidate methane response region.

6. The weak signal target recognition method according to claim 1, characterized in that, Determining the response trajectory includes: Spatial registration of consecutive frame spectral images is performed based on the UAV pose information. The candidate methane response regions searched and filtered in the registered consecutive frames have the same absorption spectrum, adjacent spatial positions and consecutive acquisition times. The searched response regions are connected according to the acquisition time to form a response trajectory.

7. The weak signal target recognition method according to claim 1, characterized in that, The response trajectory refers to the connection result of candidate methane response regions in consecutive frames that meet the trajectory connection threshold, are spatially adjacent, have continuous acquisition times, and have consistent weak absorption feature vectors. When forming the backtracking range, the response trajectory is projected onto the inspection area coordinate system according to the UAV pose information, and the backtracking range is formed along the opposite direction of the environmental flow vector at the corresponding acquisition time, with the trajectory node position in the response trajectory as the starting point. The backtracking range is used to characterize the possible source area of ​​the weak signal target of the methane plume.

8. The weak signal target recognition method according to claim 7, characterized in that, The intersection of the backtracking range with the spatial range of the inspected object in the spatial record of the inspected object is determined, and the response trajectory is determined to point to the same inspected object identifier based on the number of intersections, the intersection ratio and the continuity of the collection time in consecutive frames. When the backtracking range in consecutive frames meets the condition of continuous hits on the same inspection object identifier, a correspondence is established between the response trajectory and the inspection object identifier. When the backtracking range points to multiple inspection object identifiers at the same time, conflict resolution is performed based on the number of intersections, intersection ratio, lateral offset distance and collection time continuity of each inspection object. When it is impossible to distinguish a unique inspection object identifier, the multi-object conflict marker, candidate inspection object identifier and corresponding intersection result are output in the target identification record.

9. The weak signal target recognition method according to claim 1 or 8, characterized in that, The target identification record includes the acquisition time, UAV pose, methane absorption spectrum correspondence result, local background baseline spectrum in the same frame, weak absorption feature vector, interference response marker, response trajectory, environmental flow vector, inspection object identifier, inspection object spatial range, object corresponding status, multi-object conflict marker, candidate inspection object identifier, and corresponding intersection result.

10. The weak signal target recognition method according to claim 1, characterized in that, The prior spectral information set is established based on the historical spectral data of the inspection area, the on-site leak-free frame spectral data, and the standard absorption spectrum of methane. During the acquisition of the UAV-borne spectral sequence, the background prior spectrum is updated using the on-site leak-free frame spectral data, and the updated background prior spectrum is used to redetermine the local background baseline spectrum of the same frame.

Images