A vegetation reflectance spectrum measurement method and system for forest monitoring

Abstract

The application discloses a vegetation reflectance spectrum measurement method and system for forest monitoring, acquires original spectrum collection signals of a target area of a forest and carries out canopy penetration attenuation correction to establish a benchmark reflectance spectrum set, identifies a forest canopy projection area and atmospheric scattering interference of the benchmark reflectance spectrum set to form a ground true reflectance spectrum, carries out red edge feature positioning and band ratio analysis based on the ground true reflectance spectrum to form a vegetation reflectance spectrum index, carries out multi-time spectrum difference detection on the vegetation reflectance spectrum index and removes phenology interference to determine a change confidence interval, screens abnormal spectra based on the change confidence interval and carries out forest health grade labeling after correction by a spectrum evolution coefficient to output forest coverage change measurement results. The application can effectively eliminate the interference of forest canopy shadow and atmospheric scattering, distinguish normal phenology fluctuation from real health abnormality, and provide reliable spectrum data support for forest vegetation state monitoring.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing monitoring and forestry resource management, and in particular to a method and system for measuring vegetation reflectance spectra for forest farm monitoring. Background Technology

[0002] Remote sensing technology provides an effective means for large-scale forest vegetation status surveys. By analyzing the electromagnetic wave reflection characteristics of ground features, physiological parameters of vegetation can be obtained. The acquisition of spectral data for forest areas is affected by various factors. When incident light penetrates the canopy, energy attenuation occurs due to shading by branches and leaves. In stands with high canopy closure, the upper canopy blocks the lower layer, resulting in lower reflected signal intensity. Atmospheric aerosols and water vapor scatter different wavelengths differently, with shorter wavelengths attenuating more strongly than longer wavelengths, affecting the accuracy of calculations for spectral ratio-based vegetation indices.

[0003] The spectral response of vegetation exhibits regular changes throughout its growth cycle, with chlorophyll content and canopy structure varying at different stages in deciduous tree species. Coniferous and broadleaf forests show systematic differences in red-edge characteristics due to variations in leaf structure; using a uniform band combination to calculate indices can lead to inter-species assessment bias. This phenological rhythm is similar to spectral changes caused by abnormal factors such as pests and diseases; failing to distinguish between them can lead to misinterpreting normal fluctuations as anomalies, and relying solely on difference detection is insufficient to verify the temporal consistency of results. Summary of the Invention

[0004] This invention discloses a vegetation reflectance spectrum measurement method and system for forest farm monitoring, aiming to solve the problem of accurate acquisition of spectral data under the interference of canopy shadow and atmospheric scattering. It constructs differentiated spectral index calculation strategies for different tree species, performs phenological background separation and abnormal change identification on multi-temporal spectral data, and then corrects the reliability of change detection results through time series correlation analysis. Finally, it outputs forest stand health level assessment and cover change measurement results, providing technical support for forest resource surveys, pest and disease early warning and ecological status monitoring.

[0005] The first aspect of this invention proposes a method for measuring vegetation reflectance spectra for forest farm monitoring, comprising the following steps:

[0006] Acquire the original spectral acquisition signal of the target area of ​​the forest farm, and establish a reference reflectance spectral set by performing canopy penetration attenuation correction on the original spectral acquisition signal;

[0007] Canopy projection region identification is performed on the reference reflectance spectrum set to generate a canopy shadow correction factor. Atmospheric scattering interference is identified in the visible light and near-infrared bands of the reference reflectance spectrum set to determine the scattering attenuation distribution. Reflectance compensation is performed by combining the scattering attenuation distribution with the canopy shadow correction factor to form the true surface reflectance spectrum.

[0008] Based on the true surface reflectance spectrum, the red edge feature is used to locate the spectral response range of the tree species. The band ratio analysis is performed using the spectral response range of the tree species to obtain the spectral index component. The vegetation reflectance spectral index is formed by combining the spectral index component with the reference reflectance spectrum set.

[0009] The spectral variation amplitude is determined by multi-temporal spectral difference detection based on the vegetation reflectance spectral index. The change confidence interval after phenological interference is determined based on the spectral variation amplitude. Anomaly spectral feature set is formed by screening based on the change confidence interval.

[0010] The spectral evolution coefficient is obtained by performing a time-series correlation analysis between the true surface reflectance spectrum and the vegetation reflectance spectral index. Based on the spectral evolution coefficient, the spectral change feature set is corrected to form a corrected spectral feature. The corrected spectral feature is then labeled with the forest stand health level to output the forest cover change measurement results.

[0011] A second aspect of this invention provides a vegetation reflectance spectrum measurement system for forest farm monitoring, comprising:

[0012] The spectral acquisition module is used to acquire the original spectral acquisition signal of the target area of ​​the forest farm, and to perform canopy penetration attenuation correction on the original spectral acquisition signal to establish a reference reflectance spectral set;

[0013] The shadow correction module is used to identify the canopy projection area of ​​the reference reflectance spectrum set to generate a canopy shadow correction factor, identify atmospheric scattering interference in the visible light band and near-infrared band of the reference reflectance spectrum set to determine the scattering attenuation distribution, and perform reflectance compensation by combining the scattering attenuation distribution with the canopy shadow correction factor to form the true surface reflectance spectrum.

[0014] The index extraction module is used to determine the spectral response range of tree species by red edge feature localization based on the true reflectance spectrum of the land surface, to perform band ratio analysis using the spectral response range of tree species to obtain spectral index components, and to form a vegetation reflectance spectral index based on the spectral index components and the reference reflectance spectrum set.

[0015] The change detection module is used to determine the spectral variation amplitude by performing multi-temporal spectral difference detection based on the vegetation reflectance spectral index, determine the change confidence interval after phenological interference is removed based on the spectral variation amplitude, and form a spectral change feature set by screening abnormal spectra based on the change confidence interval.

[0016] The results output module is used to perform time-series correlation analysis between the true surface reflectance spectrum and the vegetation reflectance spectral index to obtain the spectral evolution coefficient, correct the spectral change feature set based on the spectral evolution coefficient to form corrected spectral features, and label the corrected spectral features with forest stand health level to output the forest cover change measurement results.

[0017] The beneficial effects of this invention are reflected in the following aspects: First, addressing the energy attenuation problem when incident light penetrates the canopy, a reliable benchmark reflectance spectrum set is established based on canopy penetration attenuation correction using the differences in canopy closure and leaf area index. On this basis, a shade intensity grading strategy correlated with canopy closure is adopted to replace the traditional fixed threshold method. The correspondence between shade and canopy is established based on the spatial matching relationship between the canopy edge contour and candidate shade areas. Simultaneously, differentiated compensation, rather than uniform coefficient correction, is performed to address the band selectivity characteristics of atmospheric scattering, improving the accuracy of restoring the true reflectance spectrum of the ground surface in complex forest stand environments. Second, overcoming the limitations of traditional vegetation indices that use fixed band combinations, this invention constructs index calculation parameters adapted to the characteristics of coniferous forests (with shorter wavelengths and gentler slopes at the red edge) and broadleaf forests (with longer wavelengths and steeper slopes at the red edge). This makes the vegetation reflectance spectral indices of different tree species within the same forest area comparable, avoiding health assessment biases caused by differences in the spectral characteristics of tree species in mixed forest areas. Finally, a phenological background separation mechanism based on multi-year spectral statistics was established. Unlike simple time-series difference detection methods, this mechanism decomposes spectral variation into phenological and non-phenological components by constructing a forest-type phenological baseline. It only makes anomaly judgments for changes that exceed the allowable range of phenology, and introduces spectral evolution coefficients to verify the temporal consistency of the detection results, effectively suppressing false alarms caused by seasonal phenological fluctuations. Attached Figure Description

[0018] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0019] Figure 1 This is a flowchart illustrating a vegetation reflectance spectrum measurement method for forest farm monitoring according to the present invention.

[0020] Figure 2 This is a schematic diagram of the multiple scattering optical path between the canopy layers in this invention;

[0021] Figure 3 This is a schematic diagram of the typical red-edge spectral characteristics of vegetation according to the present invention;

[0022] Figure 4 This is a schematic diagram of the multi-temporal phenological change curves of the present invention;

[0023] Figure 5This is a structural block diagram of a vegetation reflectance spectrum measurement system for forest farm monitoring according to the present invention.

[0024] Wherein: 1-Upper canopy; 2-Middle canopy; 3-Lower canopy; 4-Incident light; 5-First scattering point; 6-First scattering path; 7-Second scattering point; 8-Second scattering path; 9-Outgoing light; 10-Vertical thickness of canopy; 11-Length of multiple reflection paths; 12-Spectral curve of coniferous forest; 13-Spectral curve of broadleaf forest; 14-Red light absorption valley; 15-Red edge position of coniferous forest; 16-Red edge position of broadleaf forest; 17-Red edge area range; 18-Near-infrared plateau area; 19-Green light reflection peak; 20-Phenological curve of deciduous forest; 21-Phenological curve of evergreen forest; 22-Phenological node; 23-Abnormal change point; 24-Phenological baseline. Detailed Implementation

[0025] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0026] It should be understood that, when used in this application specification, the term "comprising" indicates the presence of the described feature, integral, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0027] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0028] The technical solutions of the embodiments of this application will be described below.

[0029] like Figure 1 As shown, this embodiment of the invention provides a method for measuring vegetation reflectance spectra for forest farm monitoring, including the following steps S110-S150:

[0030] Step S110: Obtain the original spectral acquisition signal of the target area of ​​the forest farm, and perform canopy penetration attenuation correction on the original spectral acquisition signal to establish a reference reflectance spectral set.

[0031] Specifically, the raw spectral acquisition signal of the target area of ​​the forest farm is obtained. The target area of ​​the forest farm is scanned and acquired by a hyperspectral imager mounted on a UAV. The imager records the reflected radiation energy of ground objects in the visible to short-wave infrared range according to the set spectral resolution. The radiation values ​​of each band are arranged according to spatial location to form the raw spectral acquisition signal. During the acquisition, geometric parameters such as solar altitude angle, observation zenith angle and relative azimuth angle, as well as atmospheric visibility information at the time of acquisition, need to be recorded. During a clear summer day, a hyperspectral imager conducted aerial remote sensing operations over a state-owned forest farm. Flying along a pre-defined route at a fixed altitude, it traversed different terrain types, including larch plantations, natural broadleaf mixed forests, and forest clearings. The imager collected reflectance values ​​from each terrain type in the visible to short-wave infrared spectral range, band by band. The larch plantation area exhibited low radiance in the red band due to the strong absorption of red light by its needles, but high radiance in the near-infrared band due to scattering by leaf cell structures. The natural broadleaf mixed forest area had slightly lower overall radiance than the coniferous forest due to the denser broadleaf canopy. The forest clearing area, with its mix of bare soil and herbaceous vegetation, displayed radiance characteristics distinctly different from the forest floor. These differentiated radiance records collectively constituted a raw spectral acquisition signal data cube covering the entire target area. The spatial dimension of the data cube corresponds to the ground pixel location, and the spectral dimension corresponds to each acquisition band. The raw spectral acquisition signal contained mixed information from various components, including surface reflection, atmospheric scattering, and sensor noise, requiring subsequent correction processing to extract the true surface reflectance characteristics.

[0032] A baseline reflectance spectral set was established by performing canopy penetration attenuation correction on the original spectral acquisition signal. The spectral curves of each pixel in the original spectral acquisition signal reflect the combined effect of incident irradiance and surface reflectance. The incident irradiance was calculated based on the solar-sensor geometry at the acquisition time. The apparent reflectance was obtained by dividing the reflected radiation value of each band in the original spectral acquisition signal by the corresponding incident irradiance. Then, the canopy penetration attenuation coefficient was calculated based on the shading and absorption characteristics of the canopy on the incident light. The corrected reflectance R_base=R_apparent / τ_canopy was calculated for each band of each pixel, where R_apparent is the apparent reflectance and τ_canopy is the canopy penetration attenuation coefficient, which ranges from 0 to 1. The canopy penetration attenuation coefficient is obtained by looking up tables or calculating based on parameters such as stand type, canopy closure, and leaf area index. The corrected reflectance values ​​of each band of each pixel were arranged according to spatial location and band order to form the baseline reflectance spectral set. The original spectral data collected by a hyperspectral imager showed that the near-infrared reflected radiation value of a spruce forest area was significantly lower than that of the adjacent forest clearing. This is because the high canopy closure of the spruce forest blocked and absorbed incident light through multiple scattering. Based on the canopy closure and leaf area index of the area, the canopy penetration attenuation coefficient was calculated to be approximately 0.6. After dividing the apparent reflectance by this attenuation coefficient, the near-infrared reflectance of the spruce forest area recovered from its previously low value to a level consistent with healthy coniferous forests. The corrected reflectance value was included in the baseline reflectance spectral set. The forest clearing area, due to the absence of canopy shading, had an attenuation coefficient close to 1, and the reflectance change before and after correction was minimal; it was also included in the baseline reflectance spectral set. The baseline reflectance spectral set eliminated the influence of canopy penetration attenuation, but still contained interference components such as canopy shadows and atmospheric scattering, requiring further correction to obtain the true surface reflectance spectrum.

[0033] Step S120: Canopy projection region identification is performed on the reference reflectance spectrum set to generate canopy shadow correction factor; atmospheric scattering interference identification is performed on the visible light band and near-infrared band of the reference reflectance spectrum set to determine the scattering attenuation distribution; reflectance compensation is performed by combining the scattering attenuation distribution with the canopy shadow correction factor to form the true surface reflectance spectrum.

[0034] In some embodiments, the step of identifying canopy projection regions and generating canopy shadow correction factors from the reference reflectance spectrum set includes: extracting near-infrared low-reflectance regions from the reference reflectance spectrum set to establish candidate shadow regions; performing canopy edge contour matching on the candidate shadow regions to generate canopy association identifiers; establishing shadow intensity distribution by performing canopy closure association shadow intensity grading on the canopy association identifiers; and constructing canopy shadow correction factors based on the shadow intensity distribution.

[0035] Candidate shadow areas are established by extracting low-reflectance regions in the near-infrared band from a baseline reflectance spectrum set. The reflectance values ​​of each pixel in the near-infrared band in the baseline reflectance spectrum set can effectively distinguish between well-lit areas and shaded areas. Pixels in the baseline reflectance spectrum set with reflectance below a dynamic threshold are marked as potential shadow pixels. The dynamic threshold is determined by multiplying the average near-infrared reflectance of fully lit vegetation pixels in the baseline reflectance spectrum set by an attenuation coefficient, typically ranging from 40% to 60% of the average. Spatially adjacent potential shadow pixels are aggregated to form connected regions as candidate shadow areas. When a hyperspectral imager captured data from a Pinus sylvestris forest, the near-infrared reflectance of the fully illuminated area exhibited the typical high reflectance characteristics of a healthy coniferous forest. Using this as a benchmark and setting a threshold, pixel-by-pixel scanning revealed that pixels with significantly low reflectance were mainly distributed on the north and east sides of the canopy projection. These low-reflectance pixels showed spatial stripe and patchy distribution characteristics, extending outwards along the canopy edge to form several connected areas, which were marked as candidate shadow areas. Clearings and sparse forest areas, due to ample sunlight, generally maintained high reflectance levels and did not meet the shadow characteristics, thus not being included in the candidate shadow areas. Candidate shadow areas may include genuine canopy shadows or inherent low-reflectance features such as water bodies and bare rocks, requiring further differentiation through subsequent canopy edge matching.

[0036] Canopy edge contour matching is performed on candidate shadow areas to generate canopy association identifiers. The spatial location and morphological characteristics of candidate shadow areas contain key information for distinguishing between true and false shadows. The theoretical shadow projection direction and distance corresponding to each canopy edge are calculated based on the solar altitude angle and azimuth angle. The spatial location of the candidate shadow area is matched with the theoretical shadow projection range. Canopy association identifiers are formed by assigning corresponding canopy identifiers to candidate shadow areas that fall within the theoretical projection range. The canopy association identifiers record which tree or grove of trees projected the shadow area. When the hyperspectral imager collected data on a larch forest in the afternoon, the sun was positioned in the southwest at a low altitude. Based on solar geometry, calculations showed that the canopy shadow should project northeastward, with a fixed proportional relationship between the projection length and tree height. The centroid positions of each candidate shadow area were compared with the relative azimuth and distance to the nearest canopy edge. One candidate shadow area, located on the northeast side of a larch canopy, perfectly matched the theoretical projection range and was identified as the shadow cast by that canopy, thus being assigned a canopy association identifier. Another candidate shadow area, located in a depression at the edge of the collection area, was found to be a small pond formed by seasonal water accumulation. It lacked surrounding tall canopies and its location did not match any theoretical projection range; therefore, it was determined to be a low-reflectivity water body rather than a canopy shadow and was not assigned a canopy association identifier. Matching the canopy edge contour effectively distinguishes the true canopy shadow from other low-reflectivity features.

[0037] A shadow intensity distribution was established by classifying shadow intensity based on canopy closure correlation of canopy association markers. The canopy number recorded in the canopy association markers can be used to find the canopy closure information of the corresponding canopy. The higher the canopy closure, the deeper the shadow cast and the stronger the shading effect. Based on the canopy closure value associated with the canopy association markers, the shadow intensity was divided into three levels: light shadow, moderate shadow, and heavy shadow. A canopy closure of less than 40% corresponds to light shadow because the sparse canopy still allows a lot of light to pass through. A canopy closure between 40% and 70% corresponds to moderate shadow with a moderate shading effect. A canopy closure of more than 70% corresponds to heavy shadow, which almost completely blocks direct sunlight. The intensity levels of each shadow pixel were arranged according to spatial location to form the shadow intensity distribution. The hyperspectral imager captured forest stands at different developmental stages and with varying canopy closures. A young larch forest, in its rapid growth phase with an incompletely closed canopy, showed low canopy closure and good light transmittance according to canopy association identifiers. Its projected shadow pixels were classified as lightly shaded in the shadow intensity distribution, indicating limited shading. A mature spruce forest, after years of growth, had a dense, interwoven canopy. Canopy association identifiers showed high canopy closure and almost no light transmittance. Its projected shadow pixels were classified as heavily shaded in the shadow intensity distribution, indicating a need for significant correction. The forest edge transition zone showed alternating sparse and dense canopy closure at a moderate level, corresponding to moderate shading. All shadow pixels with canopy association identifiers were classified according to the above rules to create a shadow intensity distribution map covering the entire captured area. Different intensity levels were represented by different values ​​in the map to facilitate subsequent calculation of correction factors.

[0038] A canopy shadow correction factor is constructed based on the shadow intensity distribution. The shadow level of each pixel in the shadow intensity distribution directly determines the magnitude of the correction compensation. The corresponding correction coefficient is obtained by looking up a table or calculating according to the shadow level of each pixel in the shadow intensity distribution. The canopy shadow correction factor K_shadow=1 / (1-α×I_shadow), where α is the shadow attenuation coefficient, which reflects the difference in shading characteristics of the canopy of different tree species according to the forest stand type labeling, and I_shadow is the shadow intensity value of the pixel in the shadow intensity distribution. Light shadow, moderate shadow, and heavy shadow correspond to different intensity values. The canopy shadow correction factor of non-shadow pixels is 1, indicating that no correction is required. The shadow intensity distribution of a forest area acquired by a hyperspectral imager shows that the northeastern region is heavily shaded by mature coniferous forests, exhibiting the highest shadow intensity value. Based on the shadow intensity level and the shading characteristics of the coniferous forests, a larger canopy shadow correction factor is needed to compensate for the reflectance attenuation caused by the shadows in this area. The southwestern region, consisting of open forest grassland under full illumination, shows no shadow markers in the shadow intensity distribution, and its canopy shadow correction factor is 1, requiring no compensation. The central transition zone is moderately affected by the shadows cast by sparse forests, and its canopy shadow correction factor is at a moderate level. The canopy shadow correction factors of each pixel are arranged according to their spatial location to form a correction factor matrix with the same spatial size as the reference reflectance spectral set.

[0039] Atmospheric scattering interference was identified and the scattering attenuation distribution was determined for the visible and near-infrared bands of the reference reflectance spectrum. The reflectance values ​​of the blue and near-infrared bands in the reference reflectance spectrum showed significant differences in their response to atmospheric scattering. Utilizing the characteristic that short wavelengths are sensitive to atmospheric molecular scattering while long wavelengths are relatively insensitive, the atmospheric scattering contribution was calculated for each pixel and band in the reference reflectance spectrum. The scattering contribution of each pixel and band was arranged according to spatial location and band order to form the scattering attenuation distribution, S_dist=f(λ,AOD,θ), where λ is the wavelength, AOD is the atmospheric optical thickness, and θ is the observed zenith angle. Scattering attenuation is greater at shorter wavelengths and smaller at longer wavelengths. When the hyperspectral imager collected data from the forest area, the atmospheric visibility was at a moderate level, corresponding to a certain level of aerosol concentration. The reflectance of the reference reflectance spectrum was generally high in the blue light band and the spatial distribution was relatively uniform. This is the result of the superposition of atmospheric Rayleigh scattering and aerosol scattering. The atmospheric scattering path radiation was estimated from low-reflectance features such as forest shadows and water bodies using the dark target method to determine the scattering attenuation distribution covering the entire collection area. When collecting data in high-altitude forest areas, the scattering effect was significantly weakened due to the thin atmosphere, and the overall value of the identified scattering attenuation distribution was smaller, reflecting the low scattering characteristics of the thin atmosphere.

[0040] In some embodiments, the step of forming a true surface reflectance spectrum by combining the scattering attenuation distribution with the canopy shadow correction factor for reflectance compensation includes: determining an attenuation reference value based on the band attenuation characteristics of the scattering attenuation distribution; estimating the canopy multiple scattering additional amount on the attenuation reference value to generate a scattering compensation amount; performing band-by-band reflectance correction on the scattering compensation amount in combination with the canopy shadow correction factor to establish a corrected reflectance spectrum; and forming a true surface reflectance spectrum based on the corrected reflectance spectrum.

[0041] The attenuation reference value is determined based on the band attenuation characteristics of the scattering attenuation distribution. The variation of attenuation with wavelength in each band of the scattering attenuation distribution reflects the physical characteristics of atmospheric scattering. By analyzing the attenuation difference between short and long wavelengths in the scattering attenuation distribution, the reference band with the least influence from atmospheric scattering is selected as the reference. The attenuation reference value S_base = S_dist(λ_ref), where S_dist(λ_ref) is the attenuation of the scattering attenuation distribution at the reference wavelength λ_ref. The reference wavelength is usually selected from the near-infrared or short-wave infrared bands because these bands are least affected by atmospheric molecular scattering and can reflect the background level of aerosol scattering. The scattering attenuation distribution of a forest area collected by a hyperspectral imager exhibits typical wavelength-dependent characteristics. The blue band shows the largest attenuation because short-wavelength photons have the highest probability of scattering due to collisions with atmospheric molecules. The attenuation decreases sequentially in the green and red bands because the scattering probability decreases with increasing wavelength. The near-infrared band shows the smallest attenuation, almost entirely affected by aerosol scattering. This aligns with the physical law that Rayleigh scattering by atmospheric molecules decreases with the fourth power of wavelength. The near-infrared band was selected as the reference band, and the attenuation baseline value was determined by reading the values ​​of this band from the scattering attenuation distribution. This attenuation baseline value mainly reflects the contribution of aerosol scattering, while the contribution of atmospheric molecule scattering has decreased to a very low level. When collecting data in high-altitude forest areas, the overall scattering effect is weakened due to the thin atmosphere, and the values ​​of each band in the scattering attenuation distribution are generally lower. The corresponding attenuation baseline values ​​are also significantly lower than those in low-altitude areas, reflecting the weak scattering characteristics of the thin atmosphere.

[0042] For example, the step of estimating the additional amount of canopy multiple scattering based on the attenuation reference value to generate a scattering compensation amount includes: performing inter-canopy optical path analysis based on the attenuation reference value to generate multiple reflection path lengths; using the multiple reflection path lengths to separate the scattering contributions of each canopy layer to form canopy scattering components; weighting and superimposing the canopy scattering components by canopy closure to generate a cumulative scattering amount; and correcting the cumulative scattering amount by band differences to form a scattering compensation amount.

[0043] Based on the attenuation benchmark value, interlayer optical path analysis is performed to generate multiple reflection path lengths. The atmospheric scattering intensity level reflected by the attenuation benchmark value directly affects the optical path extension effect inside the canopy. Combining the vertical structure information of the canopy, the propagation path of incident light inside the canopy is analyzed. Based on the attenuation benchmark value, the total path length of light from the canopy top after multiple reflections and scatterings to the sensor is calculated. The multiple reflection path length L_path = h_canopy × (1 + n × r_scatter), where h_canopy is the vertical thickness of the canopy, n is the average number of reflections which is related to the canopy closure (the higher the canopy closure, the more reflections), and r_scatter is the single scattering path extension coefficient, and r_scatter = k × S_base, where k is the canopy scattering response coefficient and S_base is the attenuation benchmark value. The larger the attenuation benchmark value, the stronger the atmospheric scattering, and the larger the corresponding r_scatter value. A mature larch forest, as captured by a hyperspectral imager, has a large vertical canopy thickness and high canopy closure. Incident light undergoes multiple reflections and scatterings among the branches and leaves before reaching the sensor. The calculated r_scatter value, based on the attenuation benchmark, is large, indicating a long path length for multiple reflections in this area. The tortuous propagation path of light through dense foliage is equivalent to several times the vertical thickness of the canopy, demonstrating a significant optical path lengthening effect. Conversely, a young plantation has a thinner canopy and lower canopy closure, resulting in a shorter light penetration path and fewer reflections. The relatively shorter path length for multiple reflections, calculated based on the attenuation benchmark, reflects the limited influence of the sparse canopy on the optical path. A longer path length for multiple reflections indicates more scattering and absorption of light within the canopy, and a greater additional scattering effect. Figure 2 As shown, after the incident light 4 enters the canopy, it is scattered at the first scattering point 5 in the upper canopy 1, and propagates along the first scattering path 6 to the second scattering point 7 in the middle canopy 2. It then propagates through the second scattering path 8 to the lower canopy 3, and finally forms the outgoing light 9. The vertical thickness of the canopy 10 and the number of reflections together determine the length of the multiple reflection paths 11.

[0044] The scattering contribution of each canopy layer is separated using the multiple reflection path length to form the canopy scattering component. The distribution ratio of the multiple reflection path length in the vertical direction of the canopy determines the contribution of each layer to the total scattering. The canopy is divided into upper, middle, and lower layers according to its vertical height. The contribution of each layer to the total scattering is calculated based on the distribution ratio of the multiple reflection path length in each layer. The canopy scattering component S_layer_i = S_base × (L_i / L_path) × k_i, where S_layer_i is the scattering component of the i-th layer, S_base is the attenuation reference value, L_i is the optical path length in the i-th layer, L_path is the multiple reflection path length, and k_i is the scattering efficiency coefficient of the i-th layer, which is related to the leaf density and optical characteristics of the leaves in that layer. A multi-layered mixed forest, as captured by a hyperspectral imager, exhibits a typical multi-layered vertical structure, comprising three distinct vertical layers: an upper layer of dominant trees, a middle layer of companion trees, and a lower layer of shrubs. The distribution of multiple reflection path lengths in each layer is proportional to the thickness of each layer. The upper layer of dominant trees, with its dense canopy and high leaf density, exerts the strongest initial interception and scattering effect on incident light, and the calculated scattering component of the upper layer accounts for the majority of the total scattering. The middle layer of companion trees, under pressure and with moderate leaf density, receives incident light that has already been filtered by the upper layer, resulting in a moderate scattering component. The lower layer of shrubs, due to insufficient light and sparse leaves, primarily receives transmitted and scattered light, and the calculated scattering component of the lower layer is relatively small.

[0045] The cumulative scattering amount is generated by weighting and summing the scattering components of each forest layer based on canopy closure. The canopy closure of each forest layer determines its ability to intercept light, thus affecting the weight of its scattering contribution. The scattering components of each forest layer are weighted and summed according to their canopy closure. Forest layers with higher canopy closure have a higher weight because they have a stronger effect on light interception and scattering. The cumulative scattering amount S_cum = Σ(S_layer_i × C_i), where S_layer_i is the scattering component of the i-th forest layer, and C_i is the canopy closure of the i-th forest layer. The summation is performed across all forest layers. In a multi-layered mixed forest acquired by a hyperspectral imager, the upper layer has a high canopy closure, resulting in a strong ability to intercept incident light. The scattering component of the upper forest layer receives a large weight in the weighting process, and the multiple scattering caused by the dense foliage in this layer contributes the most to the total cumulative scattering amount. The middle layer has a moderate canopy closure and a moderate weight, so the weighted contribution of the scattering component of the middle forest layer is secondary. The lower layer has a low canopy closure and a small weight, so the scattering component of the lower forest layer has a limited contribution to the total amount. The cumulative scattering of a forest stand is obtained by weighting and summing the scattering components of each forest layer with the corresponding canopy closure, reflecting the overall scattering effect. A mature pure spruce forest with a very high canopy closure has a strong effect on light interception and multiple scattering due to its extremely dense canopy, and the calculated cumulative scattering is significantly greater than that of a sparse forest stand, reflecting the strong scattering characteristics of a high canopy closure forest stand.

[0046] The cumulative scattering is corrected for band differences to form the scattering compensation. The cumulative scattering reflects the overall intensity of multiple scattering from the canopy, but different bands have different susceptibility to scattering, requiring band-specific adjustments. The specific compensation value for each band is calculated based on the cumulative scattering and wavelength dependence. The scattering compensation S_comp(λ) = S_cum × (λ_ref / λ)^β, where S_comp(λ) is the scattering compensation at wavelength λ, S_cum is the cumulative scattering, λ_ref is the reference wavelength, and β is the wavelength exponent, reflecting the rate of scattering change with wavelength and related to atmospheric and canopy scattering characteristics. The scattering compensation is large at short wavelengths and small at long wavelengths, which conforms to the wavelength dependence characteristics of scattering. Based on the cumulative scattering and wavelength index, the data collected by the hyperspectral imager for a certain forest area were differentiated for each band. The blue band, with the shortest wavelength, was most affected by scattering and received the largest scattering compensation. The scattering compensation for the green and red bands decreased sequentially with increasing wavelength. The near-infrared band, with the longest wavelength, had a scattering compensation equal to the cumulative scattering itself and did not require wavelength amplification. Under the condition of high aerosol concentration, the wavelength index value was larger, indicating a stronger wavelength dependence of scattering and more significant differences in scattering compensation among different bands.

[0047] The scattering compensation amount is combined with the canopy shading correction factor to perform band-by-band reflectance correction to establish the corrected reflectance spectrum. The scattering compensation amount and the canopy shading correction factor correct for atmospheric canopy scattering and canopy shading interference, respectively. The two are combined and applied to the reflectance value of each pixel in the reference reflectance spectrum set. The scattering compensation amount of the corresponding band is subtracted to eliminate the influence of atmospheric and canopy scattering, and then multiplied by the canopy shading correction factor of that pixel to compensate for the reflectance attenuation caused by shading. The corrected reflectance values ​​are arranged in band order to form the corrected reflectance spectrum of that pixel. A pixel in a moderately shaded area of ​​a forest region, acquired by a hyperspectral imager, was located in the shadow cast by the mature broadleaf canopy and was also affected by atmospheric scattering. After subtracting the scattering contribution band by band based on the scattering compensation amount corresponding to the pixel's location, the blue light reflectance decreased significantly and tended to normal, eliminating the artifact of high reflectance caused by scattering. Then, after amplifying and compensating the reflectance of each band according to the canopy shadow correction factor of this pixel, the near-infrared reflectance recovered from its low state to the typical range of a healthy broadleaf forest. The dual correction of scattering compensation subtraction and canopy shadow correction factor amplification effectively eliminated the superposition interference of scattering and shadow, forming the corrected reflectance spectrum of this pixel. For pixels in fully illuminated, shadowless areas, only scattering compensation subtraction was performed without shadow amplification because the canopy shadow correction factor is 1; the resulting corrected reflectance spectrum only eliminated scattering interference. The above operations were repeated for all pixels to build a corrected reflectance spectrum dataset covering the entire acquisition area.

[0048] The true surface reflectance spectrum is generated based on the corrected reflectance spectrum. After scattering subtraction and shading compensation, the corrected reflectance spectrum may contain values ​​that exceed the limits, requiring range checks and outlier handling. The reflectance values ​​of each pixel and band in the corrected reflectance spectrum are limited to a physically reasonable range of 0 to 1. For pixel bands with negative values ​​in the corrected reflectance spectrum, 0 is set to indicate that the band has almost no reflection. For pixel bands with values ​​exceeding 1 in the corrected reflectance spectrum, 1 is set to indicate that the reflectance value exceeds 1, or the correction parameters are checked back. The reflectance spectrum after range constraints is the true surface reflectance spectrum. In the corrected reflectance spectrum of a forest area acquired by a hyperspectral imager, the reflectance of most pixels in all bands fell within the physically reasonable range of 0 to 1, indicating that the correction parameters were set appropriately. After inspection, these pixels were directly included in the true reflectance spectrum of the ground surface. However, in some deep shadow areas, the blue light bands had slightly higher reflectance values ​​than the upper limit due to an excessively large canopy shadow correction factor. These outliers were truncated to the upper limit and marked as suspicious pixels requiring manual review before being included in the true reflectance spectrum of the ground surface. Pixels at the edge of a water body had near-zero reflectance in the corrected reflectance spectrum due to strong near-infrared absorption by the water itself, which is a normal spectral characteristic of water bodies and did not require special processing; they were directly included in the true reflectance spectrum of the ground surface. The true reflectance spectrum of the ground surface eliminates multiple interferences such as atmospheric scattering, canopy multiple scattering, and canopy shadow, and can accurately reflect the intrinsic spectral characteristics of the forest floor and vegetation.

[0049] Step S130: Based on the real surface reflectance spectrum, red edge feature localization is performed to determine the spectral response range of tree species. The band ratio analysis is performed using the spectral response range of tree species to obtain the spectral index component. Based on the spectral index component and the reference reflectance spectrum set, the vegetation reflectance spectral index is formed.

[0050] Specifically, the spectral response range of tree species is determined by red-edge feature localization based on the true reflectance spectrum of the ground surface. The reflectance curve in the 680-780 nm band of the true reflectance spectrum contains the core absorption and reflection characteristics of vegetation chlorophyll. The position of the band with the most dramatic reflectance change in the true reflectance spectrum within this range is calculated as the red-edge position. The characteristic response band range of different tree species is determined based on the combination of the red-edge position and the red-edge slope, forming the spectral response range of the tree species. A red-edge position biased towards the short wavelength direction usually indicates that the vegetation has low chlorophyll content or is under stress, while a red-edge position biased towards the long wavelength direction indicates that the vegetation is growing vigorously and has high chlorophyll content. The surface reflectance spectrum of a mixed coniferous and broadleaf forest area collected by a hyperspectral imager shows that the red edge of the larch stand is concentrated around 715 to 720 nm with a relatively gentle slope. This is because the slender needles and relatively sparse chlorophyll distribution result in a gentler transition from red light absorption to near-infrared reflection. In contrast, the red edge of the adjacent birch stand is concentrated around 725 to 735 nm with a steeper slope. This is because the broad, flat, and densely distributed chlorophyll in the broad leaves creates stronger red light absorption and a faster jump in near-infrared reflectance. Based on this, the spectral response ranges of the tree species in the coniferous and broadleaf forests were determined respectively. Figure 3 As shown, the X-axis represents wavelength and the Y-axis represents reflectance. The spectral curves of coniferous forest (12) and broadleaf forest (13) show local peaks at the green light reflection peak (19) and the lowest reflectance at the red light absorption valley (14). The reflectance increases rapidly within the red edge region (17). The red edge position (15) of coniferous forest is biased towards shorter wavelengths and has a gentler slope than the red edge position (16) of broadleaf forest. In the near-infrared plateau region (18), both types of curves tend to be stable.

[0051] In some embodiments, the step of using the spectral response range of the tree species to perform band ratio analysis to obtain spectral index components includes: mapping the spectral response range of the tree species to the red light band and near-infrared band to construct a response intensity comparison table; performing band difference analysis on the response intensity comparison table to obtain band response differences; extracting red edge slope features and near-infrared plateau features based on the band response differences; and performing coniferous and broad-leaved forest type difference adaptation on the red edge slope features and near-infrared plateau features to form spectral index components.

[0052] A response intensity comparison table is constructed by mapping the spectral response intervals of tree species to the red and near-infrared bands. The characteristic band ranges of each tree species defined by the spectral response intervals indicate the specific locations in the red and near-infrared bands. Based on the spectral response intervals, the red-edge starting band of each tree species is determined as the representative of the red band, and the red-edge ending band is determined as the representative of the near-infrared band. The reflectance values ​​at the corresponding locations of the spectral response intervals of each tree species are summarized by tree species type. The mean and standard deviation of the reflectance of each tree species in the two bands are arranged by tree species type to form a response intensity comparison table. The rows of the response intensity comparison table correspond to different tree species types, and the columns correspond to red reflectance, near-infrared reflectance, and their statistical characteristics. The hyperspectral imager collected data on a forest area containing four main tree species: larch, spruce, birch, and Mongolian oak. Based on the spectral response range of each tree species, the red and near-infrared bands were located, and the corresponding reflectance values ​​were read. Larch showed low reflectance in the red band because, although its needles are thin, chlorophyll absorption of red light is still significant. Its moderate reflectance in the near-infrared band is due to the simple scattering ability of its needle cells. Spruce showed low reflectance in both bands because its needles are dark green with high chlorophyll concentration and a thick waxy layer, resulting in strong overall absorption of incident light. Birch showed high reflectance in the near-infrared band but low reflectance in the red band because its broadleaf spongy tissue has developed strong multiple scattering ability, exhibiting typical broadleaf high reflectance characteristics. Mongolian oak's spectral characteristics are between those of needles and broadleaf trees because, although its leaves are broadleaf, their texture is thick and highly leathery. These data were compiled into a response intensity comparison table containing the response characteristics of the four tree species.

[0053] Band response difference analysis was performed on the response intensity comparison table to obtain the band response difference values. The difference in reflectance of each tree species in the red and near-infrared bands in the response intensity comparison table directly reflects the strength of vegetation photosynthetic activity. The difference between the near-infrared reflectance and the red reflectance of each tree species in the response intensity comparison table was calculated. The band response difference value ΔR = R_NIR - R_Red, where ΔR is the band response difference value, R_NIR is the near-infrared band reflectance, and R_Red is the red band reflectance. The larger the band response difference value, the stronger the photosynthetic activity and the higher the chlorophyll content of the vegetation. The band response difference value of healthy vegetation is usually between 0.3 and 0.5. Based on the reflectance data of each tree species in a forest area collected by a hyperspectral imager, the band response difference was calculated using data from two bands recorded in a response intensity comparison table. The birch forest exhibited the largest band response difference due to its large broad-leaved photosynthetic area and loosely arranged mesophyll cells, which facilitate gas exchange and photosynthesis, resulting in high and evenly distributed chlorophyll content. The larch forest showed a moderate band response difference because although the individual needles are small, the densely packed needle population still possesses considerable photosynthetic capacity. The spruce forest had a lower band response difference because the small, tightly packed needles and mutual shading between leaves limited the effective photosynthetic area, and near-infrared reflectance was also limited due to multiple shading. A birch forest under drought stress showed wilting leaves, closed stomata, hindered photosynthesis, and chlorophyll degradation, resulting in a significantly lower band response difference than a healthy birch forest in the same area. The band response difference directly reflects the contrast intensity between red light absorption and near-infrared reflectance in vegetation and is an important indicator for assessing vegetation vitality.

[0054] The red-edge slope feature and near-infrared plateau feature were extracted based on the band response difference. The distribution gradient of the band response difference at different band positions in the red-edge region contains key information about vegetation chlorophyll activity. The increasing pattern of the band response difference from the red-edge initiation band to the red-edge termination band was analyzed. The rate of change of the band response difference between the red-edge initiation band and the red-edge termination band was used as the red-edge slope feature. The larger the rate of change, the steeper the red-edge transition and the stronger the vegetation chlorophyll activity. The mean and dispersion of the band response difference of each band in the near-infrared plateau region were analyzed. A high mean and small dispersion indicate that the near-infrared plateau feature is stable, reflecting the integrity of the leaf cell structure and uniform scattering ability. The band response difference values ​​of a healthy birch forest in a certain forest area, acquired by a hyperspectral imager, showed a rapid increasing trend in the red edge region from 680 to 750 nm. The value quickly rose from a low value at the beginning of the red edge to a high value at the end of the red edge. This is because the high chlorophyll concentration of healthy broad leaves makes them more capable of absorbing red light, while the well-developed mesophyll sponge tissue also makes them more capable of scattering near-infrared light. The extracted red edge slope characteristics are steep. At the same time, the band response difference values ​​of each band in the near-infrared region from 750 to 900 nm remained at a stable high level with very little fluctuation. The extracted near-infrared plateau characteristics were flat and the values ​​were high, indicating that the leaf structure was sound. In a birch forest infected with leaf spot disease, the leaf tissue showed localized necrosis. The increasing trend of the band response difference in the red edge region slowed down because the chlorophyll in the lesions was destroyed, reducing the red light absorption capacity. The extracted red edge slope feature became significantly flatter. The irregular fluctuations in the band response difference of each band in the near-infrared region were due to the uneven scattering capacity caused by the alternating distribution of lesions and healthy tissue. The extracted near-infrared plateau feature was no longer flat and its overall level decreased. This feature change can serve as a basis for early identification of the disease.

[0055] Spectral index components were generated by adapting red edge slope characteristics and near-infrared plateau characteristics to coniferous and broadleaf forest types. Systematic differences exist between red edge slope and near-infrared plateau characteristics in coniferous and broadleaf forests. Based on these differences, a type-specific weighted combination strategy was adopted to generate spectral index components optimized for different forest types. For coniferous forests, the weight of near-infrared plateau characteristics was emphasized because the red edge slope of coniferous forests is relatively gentle, and the near-infrared plateau characteristic better reflects their health status. For broadleaf forests, the weight of red edge slope characteristics was emphasized because the steep red edge of broadleaf forests is sensitive to chlorophyll changes. In a mixed coniferous and broadleaf forest area, data collected by a hyperspectral imager were analyzed. For coniferous stands such as larch and spruce, a weighting scheme primarily based on near-infrared plateau features was used. Since the red edges of coniferous forests are relatively gentle, even in healthy conditions, the slope of the red edges is not significant. Therefore, emphasizing near-infrared plateau features can more sensitively capture water stress and chlorosis in conifers, and the generated coniferous forest type spectral index component accurately reflects the actual health level of the coniferous forest. For broadleaf stands such as birch and Mongolian oak, a weighting scheme primarily based on red edge slope features was used. Since the red edges of broadleaf forests are steep and highly sensitive to chlorophyll loss, emphasizing the red edge slope features can more sensitively capture broadleaf senescence and disease infection. The generated broadleaf forest type spectral index component accurately reflects chlorophyll changes in broadleaf forests. A balanced weighting scheme of both types of features was used in the transitional areas at the edge of the mixed forest to adapt to the mixed characteristics of the forest types. This forest type adaptation strategy allows the spectral index components to more accurately reflect the true health status of vegetation in different forest types.

[0056] A vegetation reflectance spectral index is formed by combining spectral index components with a reference reflectance spectral set. While spectral index components can reflect the relative health status of vegetation, they lack the ability to characterize absolute vegetation cover. This deficiency can be overcome by fusing spectral index components with the original reflectance characteristics from the reference reflectance spectral set. The vegetation reflectance spectral index VRSI = I_comp × (1 + k × R_base_NIR), where I_comp is the spectral index component, k is the fusion coefficient used to adjust the contribution of the original reflectance information, and R_base_NIR is the average reflectance in the near-infrared band of the reference reflectance spectral set. This fusion strategy retains the anti-interference ability of the normalized index while introducing the absolute reflectance information of the reference reflectance spectral set to enhance the sensitivity to vegetation cover. In areas of a forest region where the spectral index component is high, if the near-infrared reflectance of the reference reflectance spectral set is also high, the vegetation reflectance spectral index will further increase. In a mature birch forest with high canopy closure, the health spectral index component of a single tree is high, and the dense canopy also results in high near-infrared reflectance in the reference reflectance spectral set. After fusion, the vegetation reflectance spectral index reaches a very high level, accurately reflecting the actual state of the forest stand, which is both healthy and dense. In some sparse forest areas, although the spectral index components are acceptable, the near-infrared reflectance of the reference reflectance spectral set is not high due to the mixing of forest land and bare soil. After fusion, the vegetation reflectance spectral index is at a medium level, which can accurately reflect the actual state of the sparse forest. In bare land and water areas, the spectral index components are inherently low. After fusion, the vegetation reflectance spectral index remains low, and misjudgment will not occur.

[0057] Step S140: Spectral variation amplitude is determined by multi-temporal spectral difference detection based on vegetation reflectance spectral index; confidence interval of change after phenological interference removal is determined based on spectral variation amplitude; and abnormal spectrum is screened based on the confidence interval of change to form a spectral variation feature set.

[0058] In some embodiments, determining the spectral variation amplitude by multi-temporal spectral difference detection based on the vegetation reflectance spectral index includes: performing temporal registration based on the vegetation reflectance spectral index to obtain a homologous spectral sequence; extracting adjacent temporal differences from the homologous spectral sequence to generate a difference sequence; performing extreme value screening of continuous canopy coverage areas in the difference sequence to extract the maximum spectral shift; and determining the spectral variation amplitude based on the maximum spectral shift.

[0059] Homologous spectral sequences are obtained through temporal registration based on vegetation reflectance spectral index. Due to differences in flight attitude and GPS positioning during data collection, pixel positions in multi-period vegetation reflectance spectral index data may shift. Spatial registration of the vegetation reflectance spectral index data from different periods ensures precise correspondence between pixel positions. The registration accuracy must be better than the size of a single pixel to ensure that the time series reflects the spectral changes of the same ground object rather than spurious changes caused by spatial displacement. The index values ​​of each temporal phase are read from the registered vegetation reflectance spectral index data according to pixel position and arranged chronologically to form a homologous spectral sequence. A hyperspectral imager collected data from a forest farm four times in April, July, September, and November. Due to slight differences in flight attitude during each collection, the pixel positions of the vegetation reflectance spectral index images were shifted at the sub-pixel level. The data from each period were precisely registered using ground control points and image cross-correlation matching technology. After registration, the vegetation reflectance spectral index values ​​of the central pixel of a larch plot at the four time phases were arranged in chronological order to form a homologous spectral sequence containing four elements. Each element in the sequence corresponds to the index state of the pixel in different seasons. If a pixel is missing data at a certain time phase due to cloud cover, the corresponding element in the homologous spectral sequence at that location is marked as an invalid value and is not included in subsequent calculations.

[0060] Difference sequences are generated by extracting the differences between adjacent time phases of homologous spectral sequences. The difference between the index values ​​of two adjacent time phases in a homologous spectral sequence reflects the changes in vegetation vitality during that period. The differences between the index values ​​of two adjacent time phases in a homologous spectral sequence are calculated, and the differences are arranged in chronological order to form a difference sequence, D_seq=[d_12,d_23,d_34,...], where d_ij is the index difference between the i-th time phase and the j-th time phase. A positive difference indicates that the index increases and vegetation vitality increases, while a negative difference indicates that the index decreases and vegetation vitality decreases. The absolute value of the difference reflects the degree of drastic change. The homologous spectral sequence of a larch forest collected by a hyperspectral imager contains index values ​​for four phases: spring, summer, autumn, and winter. After calculating the difference between adjacent phases of the homologous spectral sequence, a difference sequence containing three elements was obtained. The positive difference from spring to summer reflects the increased vitality of larch during the greening and leaf unfolding period. The slightly negative difference in the summer to autumn difference sequence reflects a slight decline in the later growth stage. The significantly negative difference in the autumn to winter difference sequence reflects a sharp drop in vitality during the dormant period. This difference pattern is consistent with the normal phenological rhythm of larch. In a larch forest affected by pine caterpillars, the abnormally negative difference from summer to autumn far exceeds the normal decline level, and the difference sequence shows an abnormal pattern that does not conform to normal phenology.

[0061] The maximum spectral shift is extracted by extreme value screening of continuous canopy coverage areas in the difference sequence. The absolute value of each element in the difference sequence reflects the drastic degree of spectral change in the corresponding time period. The difference with the largest absolute value in the difference sequence is selected as the maximum spectral shift. During the screening, only pixels in continuous canopy coverage areas are considered to exclude interference from edge effect sensitive areas such as forest edges and gaps. Continuous canopy coverage areas are determined by vegetation reflectance spectral index threshold segmentation and connectivity analysis. In the difference sequence of a forest area acquired by a hyperspectral imager, the absolute value of the difference values ​​of each phase in the continuous larch pure forest area is the largest for the element corresponding to the summer to autumn difference sequence, but it is still within the normal phenological fluctuation range. The maximum spectral shift obtained from the difference sequence is about 0.15, indicating that the spectral change in this area is mild throughout the year. In a forest fire-affected area, the index value drops sharply from spring to summer. The absolute value of the difference in this group in the difference sequence is much higher than that of other groups. The maximum spectral shift obtained from the selection reaches more than 0.6, which is significantly beyond the normal range. The difference fluctuation of the forest edge transition zone is large due to the influence of mixed pixels from farmland or grassland, but it is an edge effect and is excluded from the selection range, so it will not interfere with the extraction of the maximum spectral shift.

[0062] The spectral variation amplitude is determined based on the maximum spectral shift. The extraction process of the maximum spectral shift simultaneously records the statistical baseline information of the corresponding pixel's homologous spectral sequences. The spectral variation amplitude is obtained by normalization using the maximum spectral shift and its associated sequence mean, defined as A_var = |d_max| / μ_base, where A_var is the spectral variation amplitude and a dimensionless ratio, d_max is the maximum spectral shift, and μ_base is the mean of the homologous spectral sequences of the pixel calculated simultaneously during the extraction of the maximum spectral shift, serving as the normalization baseline. This normalization process ensures the comparability of spectral variation amplitudes across different vegetation cover types. In a forest area, data collected by a hyperspectral imager was analyzed. While extracting the maximum spectral shift, the mean of the homologous spectral sequences for each pixel was calculated. In a healthy larch stand, although the maximum spectral shift reached 0.15 due to phenological variations, the mean of the associated sequences was relatively high, resulting in a calculated spectral variation amplitude of approximately 0.25, which is within the normal fluctuation range. In a stand affected by windfall damage, the maximum spectral shift reached 0.5, and the mean of the associated sequences also decreased significantly. The calculated spectral variation amplitude exceeded 1.0, indicating an extreme anomaly. In an evergreen spruce forest, the spectral variation was very small throughout the year, with a maximum spectral shift of only about 0.05. The calculated spectral variation amplitude was less than 0.1, reflecting the stable characteristics of evergreen coniferous forests. The spectral variation amplitude, as a relative quantity, characterizes the degree of drastic change in the spectral composition of each pixel relative to its baseline level.

[0063] In some embodiments, determining the confidence interval of change after phenological interference removal based on the spectral variation amplitude includes: segmenting the spectral variation amplitude into seasonal periods to obtain a phenological change interval; extracting the normal phenological fluctuation range within the phenological change interval to generate a phenological baseline; removing the amount of change within the phenological baseline from the spectral variation amplitude to obtain a non-phenological variation; and determining the confidence interval of change after phenological interference removal based on the non-phenological variation.

[0064] Phenological variation intervals were obtained by seasonally segmenting the spectral variation amplitude. The spectral variation amplitude exhibits significant periodic fluctuations in different seasons. The spectral variation amplitude was segmented into four phenological stages: spring greening period, summer growing season, autumn decline period, and winter dormancy period. The distribution range of the spectral variation amplitude within each phenological stage was statistically analyzed to form phenological variation intervals. The phenological variation intervals of different phenological stages show significant differences, reflecting the different physiological states of vegetation in each season. A hyperspectral imager accumulated rich multi-year, multi-temporal spectral variation data for a forest farm through quarterly monitoring over three consecutive years. Statistical analysis revealed that the spectral variation of larch stands during the spring greening period was mainly concentrated between 0.15 and 0.30, reflecting changes in the vigor of greening and leaf unfolding; during the summer growing season, it was concentrated between 0.05 and 0.15, reflecting relative stability during the vigorous growth period; during the autumn decline period, it was concentrated between 0.20 and 0.40, reflecting drastic changes in yellowing and leaf fall; and during the winter dormancy period, it was concentrated between 0.02 and 0.08, reflecting extreme stability during the dormancy period. The variation ranges of each season were defined as four phenological variation intervals for spring, summer, autumn, and winter. Figure 4 As shown, the X-axis represents time and the Y-axis represents the vegetation reflectance spectral index. The deciduous forest phenological curve 20 shows obvious seasonal fluctuations, passing through four phenological nodes 22: greening, vigorous growth, decline, and dormancy. The evergreen forest phenological curve 21 shows smaller fluctuations throughout the year. The phenological baseline 24 defines the normal fluctuation range, and abnormal change points 23 that exceed this range are potential non-phenological anomalies.

[0065] For example, the step of extracting the normal phenological fluctuation range within the phenological change interval to generate a phenological baseline includes: establishing a multi-year spectral comparison relationship based on the phenological change interval; setting multiple phenological nodes in the multi-year spectral comparison relationship; extracting the spectral change amplitude from the phenological nodes to construct the interannual fluctuation range; and performing segmented statistical analysis of the interannual fluctuation range for evergreen and deciduous forest types to form a phenological baseline.

[0066] A multi-year spectral comparison relationship was established based on phenological change intervals. The seasonal time ranges defined by the phenological change intervals provide a time window for extracting historical data for the same period. Spectral variation data of the same seasonal stage in multiple years were located from historical monitoring data based on the phenological change intervals. The spectral variation amplitudes of each year within the same phenological change interval were cross-grouped and arranged by year number and phenological stage, forming a multi-year spectral comparison relationship matrix with year as the row index and each time period within the phenological change interval as the column index. Each element in the matrix records the statistical characteristics of the variation amplitude of the corresponding year and corresponding time period. A hyperspectral imager has accumulated rich quarterly data through five years of continuous monitoring of a forest farm. For the phenological change interval of spring greening, the spectral variation data for each spring from year 1 to year 5 were extracted to form a multi-year spectral comparison relationship for spring. This multi-year comparison relationship shows that while there are some interannual differences in the variation amplitude of spring greening over the five years, the overall fluctuation range is limited. In one year, due to a warm spring, the variation amplitude of early greening was slightly higher than in other years, which is a normal climate response. Similarly, for the phenological change interval of autumn decline, a multi-year spectral comparison relationship for autumn was established. This multi-year comparison relationship shows that the interannual differences in the variation amplitude of autumn are greater than those of spring, reflecting the complex impact of autumn climate change on the leaf fall process.

[0067] Multiple phenological nodes are set in the multi-year spectral comparison relationship. There are several key turning points on the time axis of the multi-year spectral comparison relationship that need to be marked. Phenological nodes are set on the time axis of the multi-year spectral comparison relationship according to the key turning points of vegetation phenology. Typical phenological nodes include the greening start point, the vigorous growth point, the decline turning point, and the dormancy start point. Each phenological node corresponds to an inflection point or extreme point on the phenological curve. The node position is determined according to the average pattern of the multi-year data in the multi-year spectral comparison relationship. The spectral comparison of a larch forest over many years, collected by a hyperspectral imager, showed that the phenological node for the onset of greening typically occurs in mid-April when the vegetation reflectance spectral index begins to rise rapidly; the phenological node for vigorous growth occurs in mid-July when the index reaches its annual peak; the phenological node for the transition from decline occurs in early September when the index begins to fall; and the phenological node for the onset of dormancy occurs in late October when the index drops to its annual lowest point and tends to stabilize. These four time points were set as phenological nodes, and the typical spectral variation amplitude values ​​at each phenological node were recorded. Due to slightly different growth rhythms, the phenological node positions of birch forests differ from those of larch by about two weeks. Phenological nodes suitable for birch were set accordingly.

[0068] The interannual fluctuation range is constructed by extracting spectral variation amplitudes from phenological nodes. Key time positions of each phenological node are marked in the multi-year spectral comparison relationship, and the corresponding spectral variation data for that period are indexed. Based on the time index of the phenological node, the spectral variation amplitude values ​​at that node for each year are read from the multi-year spectral comparison relationship. The distribution characteristics of the multi-year spectral variation amplitude at each phenological node are statistically analyzed, and the mean and standard deviation of the variation amplitude at each phenological node are calculated. The mean plus or minus a certain number of standard deviations are used as the upper and lower limits of the interannual fluctuation range for that node. In five years of data collected by a hyperspectral imager from a forest farm, the spectral variation of the greening-initiation phenological node index varied from year to year. The variation was lower in the third year due to higher-than-average spring precipitation and earlier greening in the fourth year, while the variation was higher in the other years due to a warmer spring. The mean and standard deviation of the phenological node were calculated, and the interannual fluctuation range was determined by adding or subtracting twice the standard deviation from the mean. Variation within this range is considered normal interannual variation, while variation outside this range is considered abnormal and requires attention. The variation of the vigorous growth phenological node index was more concentrated and had a narrower interannual fluctuation range, while the decline and transition phenological node was more affected by climate and had a wider interannual fluctuation range.

[0069] The interannual fluctuation range is segmented and statistically analyzed for evergreen and deciduous forest types to form phenological baselines. Since there are fundamental differences in the interannual fluctuation range between evergreen and deciduous forests, it is necessary to conduct statistical analysis by forest type. Based on the essential differences in phenological characteristics between evergreen and deciduous forests, the interannual fluctuation range of each phenological node in both forest types is statistically analyzed to form phenological baselines for each forest type. The phenological baseline B_pheno includes the upper and lower limits of the fluctuation for each phenological node in each forest type, serving as a reference standard for judging whether spectral changes belong to normal phenological fluctuations. The interannual fluctuation ranges of deciduous tree species such as larch and birch, and evergreen tree species such as spruce and fir, were statistically analyzed in a mixed coniferous and broad-leaved forest farm using hyperspectral imaging. Due to seasonal leaf fall, the interannual fluctuation ranges of each phenological node of deciduous tree species are generally wide, forming a loose deciduous forest phenological baseline. The interannual fluctuation ranges of each node of evergreen tree species are very narrow due to the year-round coniferous state, forming a strict evergreen forest phenological baseline. In practical applications, the corresponding phenological baseline is selected for comparison and judgment based on the forest type attribute of the pixel. Deciduous forest pixels refer to the deciduous forest phenological baseline, and evergreen forest pixels refer to the evergreen forest phenological baseline to avoid misjudgment caused by using incorrect baseline standards.

[0070] Non-phenological variability is obtained by removing changes within the phenological baseline from the spectral variation amplitude. The spectral variation amplitude includes both normal phenological fluctuations and abnormal changes, which need to be separated. The spectral variation amplitude of each pixel is compared with the phenological baseline B_pheno of the corresponding forest type. If the spectral variation amplitude falls within the phenological baseline range, it is determined to be a normal phenological fluctuation and its variation is set to zero. If the spectral variation amplitude exceeds the phenological baseline range, the excess part is calculated as non-phenological variability, and the non-phenological variability is V_non=max(0,A_var-B_upper), where A_var is the spectral variation amplitude and B_upper is the upper limit of the phenological baseline. A hyperspectral imager captured the spectral variation of a larch forest stand in a certain forest area. The variation was 0.30, indicating that it was in the autumn decline phase. The upper limit of the autumn phenological baseline for deciduous forests is 0.40. This spectral variation falls within the phenological baseline range and is considered a normal phenological fluctuation with zero non-phenological variation. A birch forest stand had a spectral variation of 0.55, also in autumn, but this value exceeded the upper limit of the autumn phenological baseline for deciduous forests (0.40). The calculated non-phenological variation was 0.15, indicating an abnormal change exceeding the normal phenological fluctuation.

[0071] The confidence interval for changes after phenological interference removal was determined based on non-phenological variability. The overall distribution characteristics of non-phenological variability in each pixel of the forest area reflect the spatial pattern of abnormal changes. Statistical analysis was performed based on the distribution of non-phenological variability to distinguish normal phenological pixels with zero non-phenological variability from potential abnormal pixels with non-phenological variability greater than zero. Pixels with non-phenological variability greater than zero were sorted according to their numerical values. A specific quantile of the statistical distribution was used as the high-confidence judgment boundary. The change confidence interval CI = [V_threshold, V_max], where CI is the change confidence interval and only includes changes exceeding the phenological tolerance range, V_threshold is the specific quantile of the non-phenological variability distribution corresponding to the confidence judgment boundary, and V_max is the maximum value of non-phenological variability during the monitoring period. Statistical results of pixels across a state-owned forest farm collected by a hyperspectral imager showed that the non-phenological variability of the vast majority of pixels was zero, indicating that the spectral changes in these areas were normal phenological fluctuations and would not be included in the change confidence interval. In a larch stand in the northeast of the forest farm, the non-phenological variability of pixels was generally high and concentrated in the high quantile of the statistical distribution. The non-phenological variability of these pixels all fell into the change confidence interval and were judged as real changes with high confidence. On-site verification confirmed that pine caterpillar infestation had indeed occurred in this area, causing abnormal needle shedding. Although the non-phenological variability of a few pixels in a mixed forest on the western edge of the forest farm was greater than zero, the values ​​were low and near the confidence threshold. They were not included in the change confidence interval but were marked as areas to be observed and required continuous monitoring.

[0072] Anomaly spectrum screening based on change confidence intervals forms a spectral change feature set. The numerical range defined by the change confidence interval provides a criterion for screening high-confidence anomalous pixels. The spectral variation amplitude and non-phenological variation of all pixels are compared with the change confidence interval. Pixels and their spectral features falling within the change confidence interval are judged as high-confidence changes and included in the spectral change feature set. The spectral change feature set includes the spatial location, variation amplitude, non-phenological variation, and timing of the change for each anomalous pixel. In a forest area acquired by a hyperspectral imager, several high-confidence change pixels were screened based on change confidence intervals and included in the spectral change feature set. Some of these pixels were concentrated in a larch stand in the northeast of the forest farm, and the changes occurred in summer. On-site verification confirmed that pine caterpillar infestation caused a large amount of needle loss in this area. These pixels and their features were included in the spectral change feature set and labeled as pest type. Another part was distributed on both sides of a valley in the western part of the forest farm, and the changes occurred in spring. Verification confirmed that early spring frost damage caused damage to new buds in this area. These pixels were also included in the spectral change feature set and labeled as frost damage type. The spectral variation feature set gathers all high-confidence anomalous change information detected during the monitoring period, providing basic data for stand health assessment and disaster early warning.

[0073] Step S150: Perform time-series correlation analysis between the true surface reflectance spectrum and the vegetation reflectance spectral index to obtain the spectral evolution coefficient. Based on the spectral evolution coefficient, correct the spectral change feature set to form the corrected spectral features. Label the forest stand health level on the corrected spectral features and output the forest cover change measurement results.

[0074] Specifically, a temporal correlation analysis is performed between the true surface reflectance spectrum and the vegetation reflectance spectral index to obtain the spectral evolution coefficient. There is an inherent temporal correlation between the key band reflectance sequence of the true surface reflectance spectrum and the vegetation reflectance spectral index sequence. The temporal changes in the near-infrared band reflectance of the true surface reflectance spectrum and the temporal changes in the vegetation reflectance spectral index typically exhibit synchronous fluctuation characteristics. The temporal correlation and consistency of the changing trends between the true surface reflectance spectrum sequence and the vegetation reflectance spectral index sequence are calculated. The spectral evolution coefficient E_coef = Cov(R_seq, I_seq) / (σ_R×σ_I), where E_coef is the spectral evolution coefficient and its value ranges from -1 to 1, Cov is the covariance of the true surface reflectance spectrum sequence and the vegetation reflectance spectral index sequence, and σ_R and σ_I are the standard deviations of the two sequences, respectively. A coefficient close to 1 indicates a high positive correlation and consistent changing trends, while a coefficient close to 0 indicates a weak correlation. In multi-temporal data of a healthy larch forest stand collected by a hyperspectral imager, the near-infrared reflectance of the true surface reflectance spectrum showed regular seasonal changes, and the vegetation reflectance spectral index also showed similar seasonal fluctuations with highly synchronized changes. The calculated spectral evolution coefficient was close to 1, indicating a strong correlation between spectral response and index changes. In a forest stand affected by pests and diseases, the true surface reflectance spectrum showed an abnormal increase in the visible light band and an abnormal decrease in the near-infrared band, resulting in a non-synchronous trend between the vegetation reflectance spectral index and the single-band reflectance changes. The calculated spectral evolution coefficient was low, reflecting the disturbance of the spectral structure by the disease.

[0075] The spectral change feature set is corrected based on the spectral evolution coefficient to form the corrected spectral features. The spectral evolution coefficient reflects the inherent consistency of spectral changes and can be used as a confidence weight. The change amplitude of each pixel in the spectral change feature set is corrected with confidence weight based on the spectral evolution coefficient. The corrected spectral feature is F_corr = F_raw × E_coef, where F_corr is the change amplitude in the corrected spectral feature, F_raw is the original change amplitude in the spectral change feature set, and E_coef is the spectral evolution coefficient. Changes with high correlation are amplified to emphasize their reliability, while changes with low correlation are reduced and their weight is decreased. The high spectral evolution coefficient of a certain pine caterpillar infestation area in the spectral change feature set of a certain forest area collected by a hyperspectral imager indicates that the spectral changes have good internal consistency. After correction, the weight of the amplitude of the corrected spectral feature changes in this area remains at a high level, confirming it as a credible change event. The spectral evolution coefficient of a certain forest edge transition zone pixel is low due to the mixed pixel effect. After correction, the weight of the amplitude of the corrected spectral feature changes in this area is reduced, indicating that it should be treated with caution as there may be false detections. The spectral evolution coefficient of a certain cloud shadow edge pixel is close to zero, indicating that its spectral change lacks internal logic and is likely a pseudo-change caused by cloud shadow residue. After correction, the amplitude of the corrected spectral feature changes approaches zero and is effectively eliminated.

[0076] In some embodiments, the step of labeling the corrected spectral features with forest stand health levels and outputting forest cover change measurement results includes: extracting chlorophyll-sensitive band responses from the corrected spectral features to generate a chlorophyll distribution map; dividing the chlorophyll distribution map into chlorophyll health intervals to determine health level boundaries; labeling the corrected spectral features with level assignments based on the health level boundaries to generate forest stand health level labels; and spatially overlaying the forest stand health level labels with the corrected spectral features to output forest cover change measurement results.

[0077] The chlorophyll distribution map is generated by extracting the chlorophyll-sensitive band response from the corrected spectral features. The change information recorded by each pixel in the corrected spectral features is associated with the original spectral data of the corresponding pixel. The chlorophyll-sensitive band is located from the spectral data associated with the corrected spectral features for response value extraction. The 550 nm green light reflectance peak and the 680 nm red light absorption valley are the core locations of the chlorophyll-sensitive band. The relative index of chlorophyll content is obtained by weighting the reflectance of the two bands of the associated spectral data according to the confidence weight of each pixel in the corrected spectral features. The chlorophyll index values ​​of each pixel are then rendered according to their spatial location to form the chlorophyll distribution map. In a forest area acquired by a hyperspectral imager, the spectral data associated with high-confidence pixels in the corrected spectral features showed significant differences in reflectance at the 550 nm and 680 nm bands. Healthy larch stands exhibited a distinct green reflectance peak at 550 nm and deep red light absorption at 680 nm due to abundant chlorophyll content and strong red light absorption capacity, resulting in a high calculated chlorophyll index value, which appeared as a dark green color on the chlorophyll distribution map. In a larch stand infected with pine needle rust, the needles turned yellow and chlorophyll was lost, the 550 nm reflectance peak weakened, and the 680 nm absorption became lighter because chlorophyll decomposition led to a decrease in red light absorption capacity, resulting in a low chlorophyll index value, which appeared as a light yellow color on the chlorophyll distribution map. Forest clearings, lacking vegetation cover, had a chlorophyll index close to zero, appearing as white or gray on the chlorophyll distribution map.

[0078] The chlorophyll distribution map was used to divide the chlorophyll health intervals and determine the health level boundaries. The statistical distribution characteristics of chlorophyll index values ​​in the chlorophyll distribution map reflect the overall chlorophyll content pattern of the forest area. Based on the statistical distribution characteristics of chlorophyll index values ​​in the chlorophyll distribution map, the range of index values ​​was divided into several intervals corresponding to different health levels. The health level boundary B_health=[b_1,b_2,b_3], where b_1 is the boundary value between excellent and good, b_2 is the boundary value between good and slightly damaged, and b_3 is the boundary value between slightly damaged and severely damaged. The boundary values ​​were calibrated according to the index value range of healthy plots in the chlorophyll distribution map. The chlorophyll index values ​​in the chlorophyll distribution map of a larch forest collected by a hyperspectral imager were mainly distributed between 0.2 and 0.8. After statistical analysis of the index value range of healthy plots in the chlorophyll distribution map, values ​​above 0.6 were classified as excellent, corresponding to sufficient chlorophyll content and vigorous vegetation; values ​​between 0.4 and 0.6 were classified as good, corresponding to normal chlorophyll content and healthy vegetation growth; values ​​between 0.25 and 0.4 were classified as slightly damaged, corresponding to the beginning of chlorophyll loss and possible slight stress; and values ​​below 0.25 were classified as severely damaged, corresponding to a large loss of chlorophyll and a worrying state of vegetation health. Based on this, the health level boundary of the larch forest in this forest farm was determined to be [0.6, 0.4, 0.25]. Due to the generally higher chlorophyll content, the threshold values ​​of the health level boundary of the broadleaf forest were adjusted upwards accordingly.

[0079] Based on health grade boundaries, the corrected spectral features are labeled with their corresponding grade to generate stand health grade labels. The threshold values ​​of each interval defined by the health grade boundaries provide a quantitative standard for pixel grade determination. The chlorophyll index values ​​associated with each pixel in the corrected spectral features are compared with the health grade boundaries. The health grade of the pixel is determined according to which interval the index value falls into. A corresponding grade label is assigned to each pixel in the corrected spectral features to form a stand health grade label. The stand health grade label is recorded in the form of numerical coding or text description. A hyperspectral imager collected data on a forest area and determined the health level of each pixel in the corrected spectral features based on the health level boundary. In a healthy larch stand, the chlorophyll index value was between 0.65 and 0.75, all exceeding the excellent threshold of 0.6 in the health level boundary. The corresponding pixels were assigned the health level label "Excellent," indicating that the stand was in optimal health. In a slightly water-deficient stand, the chlorophyll index value was between 0.35 and 0.45. Some pixels fell into the good range of the health level boundary and were labeled "Good," while others fell into the slightly damaged range and were labeled "Slightly Damaged," reflecting the spatial heterogeneity of the stand's health status. In a severely insect-infested area, the chlorophyll index value was generally below 0.2. The corresponding pixels were assigned the health level label "Severely Damaged," triggering an early warning.

[0080] The forest stand health grade labels and corrected spectral features are spatially overlaid to output the forest cover change measurement results. The forest stand health grade label layer and the corrected spectral feature layer respectively carry two key information categories: health grade and change range. The forest stand health grade label layer and the corrected spectral feature layer are spatially overlaid and registered in a geographic coordinate system to form a comprehensive result layer containing multiple attributes such as forest stand health grade labels, change range, change type, and confidence score. Statistical summaries are then compiled according to the forest compartments and sub-compartments managed by the forest farm to form a forest cover change measurement result report. In a state-owned forest farm, data collected by a hyperspectral imager, after overlaying the forest stand health grade labels and corrected spectral features, showed that within a certain forest compartment, 85% of the pixels with excellent health grade labels were good, 10% were slightly damaged, 4% were severely damaged, and 1% were seriously damaged. The overall assessment was that the health status was good, but there were localized minor problems. These statistical information, along with the spatial distribution... Figure 1 The system outputs the stand cover change measurement results for each forest compartment. In one forest compartment that suffered fire, 60% of the pixels were severely damaged. The stand cover change measurement results indicate this as a severely damaged forest compartment, with the change type labeled as fire and the time of occurrence specified in the year and month. Timely post-disaster recovery assessment and afforestation planning are recommended. The stand cover change measurement results comprehensively present the health status and dynamic changes of the forest vegetation in a combination of charts and graphs, providing scientific data support for forest resource monitoring, disaster early warning, and management.

[0081] To implement the above-described method embodiments, a vegetation reflectance spectrum measurement method for forest farm monitoring is proposed to achieve the corresponding functions and technical effects. See also Figure 5 , Figure 5 This diagram illustrates a structural block diagram of a vegetation reflectance spectral measurement system 500 for forest farm monitoring, as provided in an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The vegetation reflectance spectral measurement system 500 for forest farm monitoring provided in this embodiment includes:

[0082] The spectral acquisition module 501 is used to acquire the original spectral acquisition signal of the target area of ​​the forest farm, and to perform canopy penetration attenuation correction on the original spectral acquisition signal to establish a reference reflectance spectral set.

[0083] The shadow correction module 502 is used to identify the canopy projection area of ​​the reference reflectance spectrum set to generate a canopy shadow correction factor, identify atmospheric scattering interference in the visible light band and near-infrared band of the reference reflectance spectrum set to determine the scattering attenuation distribution, and perform reflectance compensation by combining the scattering attenuation distribution with the canopy shadow correction factor to form the true surface reflectance spectrum.

[0084] The index extraction module 503 is used to determine the spectral response range of tree species by red edge feature localization based on the true reflectance spectrum of the land surface, to perform band ratio analysis using the spectral response range of tree species to obtain spectral index components, and to form a vegetation reflectance spectral index based on the spectral index components and the reference reflectance spectrum set.

[0085] The change detection module 504 is used to determine the spectral variation amplitude by performing multi-temporal spectral difference detection based on the vegetation reflectance spectral index, determine the change confidence interval after phenological interference removal based on the spectral variation amplitude, and form a spectral change feature set by screening abnormal spectra based on the change confidence interval.

[0086] The result output module 505 is used to perform time-series correlation analysis between the true surface reflectance spectrum and the vegetation reflectance spectral index to obtain the spectral evolution coefficient, correct the spectral change feature set based on the spectral evolution coefficient to form corrected spectral features, and label the corrected spectral features with forest stand health level to output the forest cover change measurement results.

[0087] The vegetation reflectance spectral measurement system 500 for forest farm monitoring described above can implement the vegetation reflectance spectral measurement method for forest farm monitoring described in the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining contents of this application embodiment can be referred to the contents of the above method embodiments, and will not be repeated in this embodiment.

[0088] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for measuring vegetation reflectance spectra for forest farm monitoring, characterized in that, include: Acquire the original spectral acquisition signal of the target area of ​​the forest farm, and establish a reference reflectance spectral set by performing canopy penetration attenuation correction on the original spectral acquisition signal; The process of identifying canopy projection regions and generating canopy shadow correction factors for the reference reflectance spectrum set includes: extracting low-reflectance regions in the near-infrared band from the reference reflectance spectrum set to establish candidate shadow areas; performing canopy edge contour matching on the candidate shadow areas to generate canopy association identifiers; establishing shadow intensity distribution by classifying shadow intensity based on canopy closure on the canopy association identifiers; constructing canopy shadow correction factors based on the shadow intensity distribution; identifying atmospheric scattering interference in the visible and near-infrared bands of the reference reflectance spectrum set to determine scattering attenuation distribution; and performing reflectance compensation by combining the scattering attenuation distribution with the canopy shadow correction factors to form the true surface reflectance spectrum. Based on the actual surface reflectance spectrum, red-edge feature localization is used to determine the spectral response interval of tree species. Band ratio analysis is then performed using these intervals to obtain spectral index components. This includes: mapping the tree species spectral response intervals to red and near-infrared bands to construct a response intensity comparison table; performing band difference analysis on the response intensity comparison table to obtain band response differences; extracting red-edge slope features and near-infrared plateau features based on the band response differences; performing coniferous and broad-leaved forest type difference adaptation on the red-edge slope features and near-infrared plateau features to form spectral index components; and combining these spectral index components with the baseline reflectance spectrum set to form a vegetation reflectance spectral index. The spectral variation amplitude is determined by multi-temporal spectral difference detection based on the vegetation reflectance spectral index. Based on the spectral variation amplitude, a confidence interval for change after phenological interference removal is determined. This includes: segmenting the spectral variation amplitude into seasonal periods to obtain phenological change intervals; extracting normal phenological fluctuation ranges within the phenological change intervals to generate a phenological baseline; removing changes within the phenological baseline from the spectral variation amplitude to obtain non-phenological variation; determining a confidence interval for change after phenological interference removal based on the non-phenological variation; and screening for abnormal spectra based on the change confidence intervals to form a spectral change feature set. The spectral evolution coefficient is obtained by performing a time-series correlation analysis between the true surface reflectance spectrum and the vegetation reflectance spectral index. Based on the spectral evolution coefficient, the spectral change feature set is corrected to form a corrected spectral feature. The corrected spectral feature is then labeled with the forest stand health level to output the forest cover change measurement results.

2. The method according to claim 1, characterized in that, The process of forming the true surface reflectance spectrum by combining the scattering attenuation distribution with the canopy shading correction factor includes: The attenuation reference value is determined based on the band attenuation characteristics of the scattering attenuation distribution; The attenuation reference value is used to estimate the additional amount of canopy multiple scattering to generate a scattering compensation amount; The scattering compensation amount is combined with the canopy shading correction factor to perform band-by-band reflectance correction to establish the corrected reflectance spectrum; The true surface reflectance spectrum is formed based on the corrected reflectance spectrum.

3. The method according to claim 1, characterized in that, The determination of spectral variation amplitude based on multi-temporal spectral difference detection using the vegetation reflectance spectral index includes: Based on the vegetation reflectance spectral index, temporal registration is performed to obtain homologous spectral sequences; The phase difference between adjacent spectral sequences is extracted to generate a difference sequence; The maximum spectral shift is extracted by screening extreme values ​​of continuous canopy coverage areas in the difference sequence. The spectral variation magnitude is determined based on the maximum spectral shift.

4. The method according to claim 1, characterized in that, The step of labeling the corrected spectral features with forest stand health levels and outputting forest cover change measurement results includes: The chlorophyll distribution map is generated by extracting the chlorophyll-sensitive band response of the corrected spectral features. The chlorophyll distribution map is divided into healthy chlorophyll intervals to determine the boundaries of the health level; Based on the health level boundary, the corrected spectral features are labeled with the level classification to generate forest stand health level labels; The forest stand health level label and the corrected spectral features are spatially superimposed to output the forest stand cover change measurement results.

5. The method according to claim 2, characterized in that, The step of estimating the additional amount of canopy multiple scattering on the attenuation reference value to generate a scattering compensation amount includes: Based on the attenuation reference value, perform inter-canopy optical path analysis to generate multiple reflection path lengths; The scattering contribution of each forest layer is separated by using the multiple reflection path lengths to form the forest layer scattering component; The cumulative scattering amount is generated by weighting and superimposing the forest scattering components by canopy closure. The cumulative scattering amount is corrected for band differences to form a scattering compensation amount.

6. The method according to claim 1, characterized in that, The step of extracting the normal phenological fluctuation range within the phenological change range to generate a phenological baseline includes: Establish a multi-year spectral comparison relationship based on the aforementioned phenological change range; Multiple phenological nodes are set in the multi-year synchronous spectral comparison relationship; The interannual fluctuation range is constructed by extracting the spectral variation amplitude from the phenological nodes. The interannual fluctuation range is segmented and statistically analyzed for evergreen and deciduous forest types to form a phenological baseline.

7. A vegetation reflectance spectrum measurement system for forest farm monitoring, characterized in that, include: The spectral acquisition module is used to acquire the original spectral acquisition signal of the target area of ​​the forest farm, and to perform canopy penetration attenuation correction on the original spectral acquisition signal to establish a reference reflectance spectral set; The shadow correction module is used to identify canopy projection regions and generate canopy shadow correction factors from the reference reflectance spectrum set. This includes: extracting low-reflectance areas in the near-infrared band from the reference reflectance spectrum set to establish candidate shadow areas; performing canopy edge contour matching on the candidate shadow areas to generate canopy association identifiers; establishing a shadow intensity distribution by classifying the shadow intensity based on canopy closure of the canopy association identifiers; constructing canopy shadow correction factors based on the shadow intensity distribution; identifying atmospheric scattering interference in the visible and near-infrared bands of the reference reflectance spectrum set to determine scattering attenuation distribution; and performing reflectance compensation using the scattering attenuation distribution combined with the canopy shadow correction factors to form the true surface reflectance spectrum. The index extraction module is used to determine the spectral response range of tree species by locating red-edge features based on the true surface reflectance spectrum, and to perform band ratio analysis using the spectral response range of tree species to obtain spectral index components. This includes: mapping the spectral response range of tree species to red and near-infrared bands to construct a response intensity comparison table; performing band difference analysis on the response intensity comparison table to obtain band response differences; extracting red-edge slope features and near-infrared plateau features based on the band response differences; performing coniferous and broad-leaved forest type difference adaptation on the red-edge slope features and near-infrared plateau features to form spectral index components; and forming a vegetation reflectance spectral index based on the spectral index components and the benchmark reflectance spectrum set. The change detection module is used to determine the spectral variation amplitude by performing multi-temporal spectral difference detection based on the vegetation reflectance spectral index, and to determine the change confidence interval after phenological interference removal based on the spectral variation amplitude. This includes: segmenting the spectral variation amplitude into seasonal periods to obtain phenological change intervals; extracting the normal phenological fluctuation range within the phenological change intervals to generate a phenological baseline; removing the change amount within the phenological baseline from the spectral variation amplitude to obtain non-phenological variation amounts; determining the change confidence interval after phenological interference removal based on the non-phenological variation amounts; and performing abnormal spectral screening based on the change confidence intervals to form a spectral change feature set. The results output module is used to perform time-series correlation analysis between the true surface reflectance spectrum and the vegetation reflectance spectral index to obtain the spectral evolution coefficient, correct the spectral change feature set based on the spectral evolution coefficient to form corrected spectral features, and label the corrected spectral features with forest stand health level to output the forest cover change measurement results.

Images