Quantitative retrieval method and system of environmental parameters based on spectral morphological features

Abstract

The present application relates to remote sensing technology field, specifically for the environmental parameter quantitative inversion method and system based on spectral morphological characteristics, including: obtaining the reflectance spectral data of target object, identifying the absorption valley in the reflectance spectral data and determining the valley bottom minimum reflectance point and valley bottom reflectance value, carrying out convex envelope line fitting to the absorption valley corresponding wave band range, obtaining the envelope line reflectance of valley bottom wave band position as the valley outside reference reflectance, and calculating the valley bottom depth, taking the valley bottom minimum reflectance point as the starting point, scanning the reflectance data points of adjacent wave band to the left side wave band and right side wave band respectively, the present application can more accurately find out the real left and right extension range of absorption valley by calculating the slope of left and right sides of absorption valley respectively and generating independent dynamic threshold, and after several iterations until the boundary is stable, the extracted valley depth and asymmetric width are closer to the actual situation of spectrum, and the extraction accuracy and reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of remote sensing technology, specifically to a method and system for quantitative inversion of environmental parameters based on spectral morphological characteristics. Background Technology

[0002] With the rapid development of remote sensing technology, hyperspectral remote sensing, due to its ability to provide continuous, narrow-band spectral information, has been widely used for the quantitative inversion of surface environmental parameters. Traditional hyperspectral environmental parameter inversion methods mainly rely on empirical statistical models, physical radiative transfer models, or simple spectral indices. While these methods have achieved rapid estimation of environmental parameters to some extent, they still have significant shortcomings in practical applications.

[0003] First, traditional methods often focus on the overall reflectance level of the spectrum or the ratio of specific bands, failing to adequately utilize the fine morphological features of absorption valleys in the spectral curve. Absorption valleys are the most characteristic regions in the spectrum of ground objects that reflect the composition and content of substances. Their depth, width, and asymmetry are often closely related to the target environmental parameters. However, existing technologies usually employ symmetry assumptions or envelope removal methods with fixed windows, making it difficult to accurately capture the true geometric shape of absorption valleys, resulting in low feature extraction accuracy.

[0004] In addition, existing absorption feature extraction methods mostly use fixed thresholds or global envelope fitting, ignoring the asymmetry of different absorption valleys on the left and right wingspans, as well as the influence of spectral noise and background interference. This can easily cause wingspan boundary positioning deviations, affecting the reliability of features, and consequently affecting the input quality of subsequent inversion models and the accuracy of the final inversion results. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for quantitative inversion of environmental parameters based on spectral morphological characteristics, comprising:

[0006] Acquire the reflectance spectral data of the target object, identify absorption valleys in the reflectance spectral data and determine the lowest reflectance point and valley bottom reflectance value; perform convex envelope fitting on the corresponding band range of the absorption valley to obtain the envelope reflectance of the valley bottom band position as the reference reflectance outside the valley, and calculate the valley bottom depth.

[0007] Starting from the lowest reflectance point at the bottom of the valley, the reflectance data points of adjacent bands are scanned independently to the left and right bands respectively. The proportion of the difference between each data point and the reflectance value at the bottom of the valley is calculated to obtain the left difference proportion sequence and the right difference proportion sequence.

[0008] Starting from the lowest reflectivity point at the bottom of the valley, a predetermined number of adjacent band points are taken to the left and right as initial scan segments. The average slope is calculated within each initial scan segment to obtain the initial average slope on the left and the initial average slope on the right.

[0009] Based on the normalized product of the valley depth with the reciprocal of the initial average slope on the left and the reciprocal of the initial average slope on the right, respectively, the left dynamic threshold and the right dynamic threshold are generated.

[0010] Find the data point position that first exceeds the left dynamic threshold in the left difference proportion sequence as the left wingspan boundary, and find the data point position that first exceeds the right dynamic threshold in the right difference proportion sequence as the right wingspan boundary to obtain the initial wingspan boundary;

[0011] Based on the reflectivity data within the initial wingspan boundary, the left and right average slopes are recalculated, the dynamic threshold is updated, and the boundary search steps are repeated until the boundary position converges and stabilizes, thus obtaining the final asymmetric wingspan.

[0012] Normalized absorption morphology features are calculated based on the final asymmetric wingspan and valley depth; these features are then input into a pre-trained inversion model to obtain quantitative inversion results of the target object's environmental parameters.

[0013] Preferably, identifying absorption valleys in reflectance spectral data and determining the lowest reflectance point and valley reflectance value at the bottom of the valley includes:

[0014] A hyperspectral imager is used to perform multi-band spectral measurements on the target object, and the original spectral signals of the target object in the visible to near-infrared band range are collected to obtain the original reflectance spectral data containing multiple bands.

[0015] The raw reflectance spectral data is smoothed, denoised, and radiometrically corrected to obtain preprocessed reflectance spectral data.

[0016] In the preprocessed reflectance spectral data, adjacent reflectance values ​​are compared band by band to identify continuous bands that first decrease and then increase in reflectance and meet preset depth and width thresholds as potential absorption valleys.

[0017] Within each potential absorption valley, the band point with the lowest reflectance value is identified as the valley bottom lowest reflectance point, and the reflectance value of that point is recorded as the valley bottom reflectance value.

[0018] Preferably, a convex envelope is fitted to the corresponding band range of the absorption valley to obtain the envelope reflectance at the valley bottom band position as the reference reflectance outside the valley, and the valley bottom depth is calculated, including:

[0019] Extend the bands to the left and right sides from the lowest reflectivity point at the bottom of the valley until a shoulder point is found where the reflectivity is higher than the reflectivity value at the bottom of the valley and the reflectivity of subsequent bands increases monotonically. Determine the band position corresponding to the shoulder point as the left and right boundaries of the absorption valley, thereby obtaining the band range corresponding to the absorption valley.

[0020] Extract all reflectance data points within the corresponding band range of the absorption valley as a point set;

[0021] Starting from the left and right boundary points of the point set, check the intermediate band points in turn. If the current point is below the straight line formed by connecting the existing envelope points, retain the existing envelope. If it is above the straight line, add the point to the envelope and remove the intermediate point whose convexity has been destroyed. Repeat this process until all points are below or on the envelope line, forming an upward convex envelope line.

[0022] On the convex envelope, the reflectance value of the envelope corresponding to the valley bottom band position is taken as the reference reflectance outside the valley.

[0023] Calculate the valley depth, where the valley depth is the difference between the reference reflectance outside the valley and the reflectance value at the valley bottom.

[0024] Preferably, starting from the lowest reflectance point at the bottom of the valley, reflectance data points in adjacent bands are independently scanned to the left and right bands respectively. The proportion of the difference between each data point and the valley reflectance value is calculated to obtain the left difference proportion sequence and the right difference proportion sequence, including:

[0025] The band number corresponding to the point with the lowest reflectivity at the bottom of the valley is determined as the scanning center.

[0026] Based on the scanning center, a preset number of band reflectance data are sequentially acquired along the direction of decreasing band number as the left scanning dataset, and a preset number of band reflectance data are sequentially acquired along the direction of increasing band number as the right scanning dataset.

[0027] Subtract the valley reflectance value from each reflectance data in the left scan dataset and the right scan dataset to obtain the left reflectance difference set and the right reflectance difference set;

[0028] Divide each difference in the set of reflectivity differences on the left and the set of reflectivity differences on the right by the reflectivity value at the bottom of the valley to obtain the relative proportion coefficient on the left and the relative proportion coefficient on the right.

[0029] According to the band number in order of distance from the scanning center from near to far, the relative proportion coefficients on the left and right sides are arranged respectively to obtain the left difference proportion sequence and the right difference proportion sequence.

[0030] Preferably, a predetermined number of adjacent band points are continuously selected from the lowest reflectivity point at the bottom of the valley to the left and to the right as initial scan segments. The average slope is calculated within each initial scan segment to obtain the initial average slope on the left and the initial average slope on the right, including:

[0031] Set a fixed number of band points as the initial scan window length;

[0032] Using the band position where the lowest reflectivity point at the bottom of the valley is located as the starting anchor point, a number of consecutive band data equal to the length of the initial scanning window are extracted along the direction of decreasing band number to form the initial scanning segment on the left.

[0033] A number of consecutive band data points, equal to the length of the initial scan window, are extracted along the direction of increasing band number to form the initial scan segment on the right.

[0034] In the initial scan segments on the left and right, the band points farthest from the starting anchor point are identified and used as the left edge point and the right edge point, respectively.

[0035] Calculate the difference in reflectivity and wavelength between the left edge point and the starting anchor point, and the difference in reflectivity and wavelength between the right edge point and the starting anchor point, respectively.

[0036] Divide the difference in reflectivity values ​​at the left edge points by the difference in their corresponding wavelength values ​​to obtain the initial average slope on the left.

[0037] Divide the difference in reflectivity values ​​at the right edge points by the difference in their corresponding wavelength values ​​to obtain the initial average slope on the right side.

[0038] Preferably, the left and right dynamic thresholds are generated based on the normalized products of the valley depth and the reciprocals of the initial average slopes on the left and right sides, respectively, including:

[0039] Obtain the initial average slope on the left and the initial average slope on the right respectively, and calculate the mathematical reciprocal of their values ​​to obtain the reciprocal of the slope on the left and the reciprocal of the slope on the right.

[0040] Multiply the reciprocal of the slope on the left and the reciprocal of the slope on the right by the valley depth to obtain the product of the slope on the left and the product of the slope on the right.

[0041] Normalization operations are applied to the left depth-slope product and the right depth-slope product respectively, and the product results are converted into numerical values ​​with the same dimensions as the difference proportion sequence to obtain the left dynamic threshold and the right dynamic threshold.

[0042] Preferably, the first data point in the left difference percentage sequence that first exceeds the left dynamic threshold is taken as the left wingspan boundary, and the first data point in the right difference percentage sequence that first exceeds the right dynamic threshold is taken as the right wingspan boundary, thus obtaining the initial wingspan boundary, including:

[0043] Read each percentage value in the left-hand difference percentage sequence in order of distance from the lowest reflectivity point at the bottom of the valley, from near to far.

[0044] Each percentage value read is compared with the dynamic threshold on the left to determine whether the current percentage value is greater than the dynamic threshold on the left.

[0045] When the current percentage value is determined to be greater than the left dynamic threshold for the first time, the reading stops and the band position corresponding to the percentage value is recorded as the left wingspan boundary.

[0046] Read each percentage value in the difference percentage sequence on the right in order of distance from the lowest reflectivity point at the bottom of the valley, from near to far.

[0047] Each percentage value read is compared with the dynamic threshold on the right to determine whether the current percentage value is greater than the dynamic threshold on the right.

[0048] When the current percentage value is determined to be greater than the dynamic threshold on the right for the first time, the reading stops and the band position corresponding to the percentage value is recorded as the right wingspan boundary.

[0049] The band range formed by the left and right wingspan boundaries is defined as the primary wingspan boundary.

[0050] Preferably, based on the reflectivity data within the initial wingspan boundary, the left and right average slopes are recalculated, the dynamic threshold is updated, and the boundary search step is repeated until the boundary position converges and stabilizes, resulting in the final asymmetric wingspan, including:

[0051] Use the band range determined by the initial wingspan boundary as the current reference range;

[0052] The average slope is recalculated using reflectance data within the current reference range, and updated left and right dynamic thresholds are generated accordingly.

[0053] Based on the updated left and right dynamic thresholds, the new left wingspan boundary and the new right wingspan boundary are obtained by searching again in the left difference proportion sequence and the right difference proportion sequence, respectively.

[0054] Compare the new left and right wingspan boundaries with the boundary positions of the current reference range. If the positions have changed, use the new left and right wingspan boundaries as the new current reference range and return to the step of generating the updated dynamic threshold.

[0055] If the position does not change, stop the iteration, determine the left and right wingspan boundaries at this time as the final boundaries, and calculate the wavelength difference between the two to obtain the final asymmetric wingspan.

[0056] Preferably, normalized absorption morphology features are calculated based on the final asymmetric wingspan and valley depth; these normalized absorption morphology features are then input into a pre-trained inversion model to obtain quantitative inversion results of the target object's environmental parameters, including:

[0057] By combining the valley depth with the final asymmetric wingspan, a numerical value that can characterize the geometry of the absorption valley is constructed, and the normalized absorption morphology features are obtained.

[0058] The pre-built and trained environmental parameter inversion model is invoked, and the model establishes a mapping relationship between spectral morphological features and environmental parameters through machine learning algorithms;

[0059] The normalized absorption morphology characteristics are used as input data and fed into the environmental parameter inversion model for calculation.

[0060] Obtain the output values ​​of the environmental parameter inversion model and determine these output values ​​as the quantitative inversion results of the environmental parameters of the target object.

[0061] A quantitative inversion system for environmental parameters based on spectral morphology features, applicable to the aforementioned quantitative inversion method for environmental parameters based on spectral morphology features, including:

[0062] The depth extraction module is used to acquire the reflectance spectral data of the target object, identify absorption valleys in the reflectance spectral data and determine the lowest reflectance point and valley bottom reflectance value; perform convex envelope fitting on the corresponding band range of the absorption valley to obtain the envelope reflectance of the valley bottom band position as the reference reflectance outside the valley, and calculate the valley bottom depth.

[0063] The sequence construction module is used to independently scan the reflectance data points of adjacent bands, starting from the lowest reflectance point at the bottom of the valley, and to calculate the proportion of the difference between each data point and the reflectance value at the bottom of the valley, so as to obtain the left difference proportion sequence and the right difference proportion sequence.

[0064] The slope initial calculation module is used to continuously take a preset number of adjacent band points from the lowest reflectivity point at the bottom of the valley to the left and right as initial scan segments, respectively, and calculate the average slope within each initial scan segment to obtain the initial average slope on the left and the initial average slope on the right.

[0065] The threshold generation module is used to generate dynamic thresholds for the left and right sides based on the normalized products of the valley depth and the reciprocal of the initial average slope on the left and the reciprocal of the initial average slope on the right, respectively.

[0066] The initial boundary positioning module is used to find the location of the data point that first exceeds the left dynamic threshold in the left difference proportion sequence as the left wingspan boundary, and to find the location of the data point that first exceeds the right dynamic threshold in the right difference proportion sequence as the right wingspan boundary, thus obtaining the initial wingspan boundary;

[0067] The iterative convergence module is used to recalculate the left and right average slopes based on the reflectivity data within the initial wingspan boundary, update the dynamic threshold, and repeat the boundary search steps until the boundary position converges and stabilizes, thus obtaining the final asymmetric wingspan.

[0068] The model inversion module is used to calculate the normalized absorption morphology features based on the final asymmetric wingspan and valley depth; the normalized absorption morphology features are input into the pre-trained inversion model to obtain the quantitative inversion results of the environmental parameters of the target object.

[0069] Compared with the prior art, the beneficial effects of the present invention are:

[0070] This invention calculates the slopes of the left and right sides of the absorption valley separately and generates independent dynamic thresholds. After several iterations until the boundary stabilizes, it can more accurately find the true left and right extension range of the absorption valley. In particular, it will not make mistakes when the shapes of the left and right sides are different. The valley depth and asymmetric width extracted in this way are closer to the actual situation of the spectrum than the traditional fixed threshold or symmetry assumption method, thus improving the extraction accuracy and reliability of absorption morphology features.

[0071] The dynamic threshold in this invention is generated based on the normalized relationship between valley depth and local slope, and through iterative updates, it can effectively resist spectral noise, background interference and differences in different land cover types, making the feature extraction process more robust, suitable for complex field environments and diverse target objects, and reducing feature extraction failures caused by spectral quality fluctuations.

[0072] The normalized absorption morphology features extracted by this invention can more accurately reflect the spectral response patterns of material composition and content. When these features are input into a pre-trained inversion model, the model can more easily learn the true relationship between the spectrum and environmental parameters, which can significantly improve the accuracy, stability, and generalization ability of quantitative inversion of environmental parameters and reduce inversion errors. Attached Figure Description

[0073] Figure 1 This is a schematic flowchart of the overall method in one embodiment of the present invention;

[0074] Figure 2 This is a schematic diagram of the overall system architecture in one embodiment of the present invention.

[0075] In the diagram: 1. Depth extraction module; 2. Sequence construction module; 3. Slope initial calculation module; 4. Threshold generation module; 5. Initial boundary localization module; 6. Iterative convergence module; 7. Model inversion module. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0077] Example 1, please refer to Figure 1 This invention provides a technical solution: a method for quantitative inversion of environmental parameters based on spectral morphology characteristics, comprising:

[0078] S1. Obtain the reflectance spectral data of the target object, identify the absorption valleys in the reflectance spectral data and determine the lowest reflectance point and valley bottom reflectance value; perform convex envelope fitting on the corresponding band range of the absorption valley to obtain the envelope reflectance of the valley bottom band position as the reference reflectance outside the valley, and calculate the valley bottom depth.

[0079] S2. Starting from the lowest reflectance point at the bottom of the valley, independently scan the reflectance data points of adjacent bands to the left and right bands respectively, calculate the proportion of the difference between each data point and the reflectance value at the bottom of the valley, and obtain the left difference proportion sequence and the right difference proportion sequence.

[0080] S3. Take a preset number of adjacent band points from the lowest reflectivity point at the bottom of the valley to the left and right as the initial scanning segments respectively, calculate the average slope in each initial scanning segment, and obtain the initial average slope on the left and the initial average slope on the right.

[0081] S4. Based on the normalized product of the valley depth with the reciprocal of the initial average slope on the left and the reciprocal of the initial average slope on the right, respectively, generate the dynamic threshold on the left and the dynamic threshold on the right.

[0082] S5. Find the data point position that first exceeds the left dynamic threshold in the left difference proportion sequence as the left wingspan boundary, and find the data point position that first exceeds the right dynamic threshold in the right difference proportion sequence as the right wingspan boundary to obtain the initial wingspan boundary.

[0083] S6. Based on the reflectivity data within the initial wingspan boundary, recalculate the left and right average slopes, update the dynamic threshold, and repeat the boundary search steps until the boundary position converges and stabilizes to obtain the final asymmetric wingspan.

[0084] S7. Calculate the normalized absorption morphology features based on the final asymmetric wingspan and valley depth; input the normalized absorption morphology features into the pre-trained inversion model to obtain the quantitative inversion results of the environmental parameters of the target object.

[0085] It should be noted that in the process of environmental parameter inversion, the basic spectral features must first be accurately extracted: reflectance spectral data represents the energy reflection of the target object in a continuous band, reflecting the physicochemical properties of the substance; absorption valleys are concave regions in the spectral curve formed by the absorption of energy by a specific substance, and their shape is directly related to the concentration of environmental parameters; the lowest reflectance point at the bottom of the valley represents the location of the strongest absorption; convex emboss fitting is to construct a "virtual baseline" to remove background interference, so that the calculated valley depth can more purely reflect the absorption intensity of the target substance and avoid errors caused by background tilt; the scan difference ratio sequence reflects The relative rate of change of the spectral curve from the valley bottom to both sides is scanned independently on the left and right sides because actual absorption valleys often exhibit asymmetrical shapes, and separate processing preserves richer morphological details; the initial average slope reflects the steepness near the valley bottom; a dynamic threshold is generated based on the normalized product of the valley bottom depth and the inverse of the slope, meaning that the system does not use a fixed standard to determine the valley boundary, but rather adaptively generates a judgment threshold based on the depth and steepness of each absorption valley; if the valley is deeper or the slope is gentler, the threshold will be automatically adjusted to cover a wider range; the purpose of this step is to ensure that the most reasonable wingspan boundary can be found for absorption valleys of different shapes;

[0086] The initial wingspan boundary is only a preliminary estimate based on local information and may not be accurate enough. The slope is recalculated and the threshold is updated based on the data within the initial boundary. This iterative convergence process is to correct any deviations that may exist in the initial scan segment. By continuously adjusting the reference range until the boundary position no longer changes, the final asymmetric wingspan width is obtained. This width value accurately describes the true coverage of the absorption feature on the wavelength axis. The normalized absorption morphology feature constructed based on the final asymmetric wingspan width and valley depth is a high-dimensional generalization of the absorption valley geometry. It eliminates the influence of dimensions and highlights the morphological essence. This feature is input into a pre-trained inversion model. The model learns the nonlinear mapping relationship between morphology and parameters and outputs the quantitative inversion results of environmental parameters of the target object (such as soil and vegetation). The purpose of this step is to use finely extracted morphological features to replace the traditional coarse spectral indices, thereby significantly improving the accuracy and stability of environmental parameter inversion.

[0087] In an optional embodiment, identifying absorption valleys in reflectance spectral data and determining the lowest reflectance point and valley reflectance value at the valley bottom includes:

[0088] A hyperspectral imager is used to perform multi-band spectral measurements on the target object, and the original spectral signals of the target object in the visible to near-infrared band range are collected to obtain the original reflectance spectral data containing multiple bands.

[0089] It should be noted that hyperspectral imaging technology can acquire spectral information of a target object in a continuous and narrow band, forming a complete spectral curve, which is the data basis for identifying subtle absorption features. The raw spectral signal usually exists in the form of digital quantized values, reflecting the light intensity signal received by the sensor. The visible to near-infrared band range usually covers the spectral response range of most key components of soil, vegetation, and other land features. For example, using a hyperspectral imager with a spectral range of 400nm to 1000nm and a spectral resolution of 2.5nm, scanning a piece of farmland soil can collect the raw light intensity values ​​of the soil in 240 continuous bands. These values ​​are arranged in order of wavelength to form the raw reflectance spectral data.

[0090] The raw reflectance spectral data is smoothed, denoised, and radiometrically corrected to obtain preprocessed reflectance spectral data.

[0091] It should be noted that the raw data often contains sensor dark current noise, random noise, and atmospheric scattering, which will cause feature extraction errors if used directly. Smoothing and denoising are used to remove spurious signals, making the spectral curve smoother and facilitating subsequent slope and morphology calculations. Radiometric correction converts the raw digital quantization values ​​into physically meaningful reflectance values, i.e., the ratio of reflected energy to incident energy, ensuring data comparability. For example, the Savitzky-Golay filtering algorithm is used, with a window size of 5 bands and a polynomial order of 2, to smooth the raw data to eliminate high-frequency noise. Then, using the reflectance data of a standard whiteboard as a reference, the smoothed digital quantization value is divided by the value of the corresponding band on the whiteboard to calculate the reflectance data in the range of 0 to 1, such as a reflectance calculation result of 0.25 at a certain band.

[0092] In the preprocessed reflectance spectral data, adjacent reflectance values ​​are compared band by band to identify continuous bands that first decrease and then increase in reflectance and meet preset depth and width thresholds as potential absorption valleys.

[0093] It should be noted that the absorption valley is geometrically represented as a concave region of the spectral curve; a simple "decline followed by a rise" is easily affected by minor noise, so a threshold limitation must be introduced; a preset depth threshold is used to eliminate false valleys that are too shallow, and a preset width threshold is used to eliminate spurious signals that are too narrow; only when the sum of the number of consecutive decreasing and rising bands is greater than the preset width threshold, and the decrease in the valley bottom relative to the starting point is greater than the preset depth threshold, is it determined to be an effective absorption valley; for example, if the preset depth threshold is set to 0.02 and the preset width threshold is 10nm; in the wavelength range of 600nm to 700nm, the reflectivity decreases from 0.40 to 0.15 and then rises to 0.35, with a decrease of 0.25 greater than 0.02, and the span of 100nm is greater than 10nm, the system determines that this continuous band from 600nm to 700nm constitutes a potential absorption valley;

[0094] Within each potential absorption valley, the band point with the lowest reflectance value is identified as the lowest reflectance point at the bottom of the valley, and the reflectance value at that point is recorded as the reflectance value at the bottom of the valley.

[0095] It should be noted that the valley bottom is the location of the strongest absorption in the absorption valley, and it is also the benchmark anchor point for subsequent construction of the envelope and calculation of asymmetric width and depth. The determined valley bottom point must be the global minimum within this local range to ensure the uniqueness and accuracy of feature extraction. Recording the specific value and wavelength position of this point is the starting condition for subsequent algorithm iteration calculations. For example, in the potential absorption valley dataset of 600nm to 700nm identified above, the system compares the reflectance values ​​of all band points and finds that the reflectance value of 0.15 at a wavelength of 660nm is the minimum value in this range, which is lower than 0.16 at the adjacent 658nm and 0.17 at 662nm. Therefore, 660nm is determined as the valley bottom band position, and 0.15 is recorded as the valley bottom reflectance value.

[0096] In an optional embodiment, a convex envelope is fitted to the band range corresponding to the absorption valley to obtain the envelope reflectance at the valley bottom band position as the reference reflectance outside the valley, and the valley bottom depth is calculated, including:

[0097] Extend the bands to the left and right sides from the lowest reflectivity point at the bottom of the valley until a shoulder point is found where the reflectivity is higher than the reflectivity value at the bottom of the valley and the reflectivity of subsequent bands increases monotonically. Determine the band position corresponding to the shoulder point as the left and right boundaries of the absorption valley, thereby obtaining the band range corresponding to the absorption valley.

[0098] It should be noted that determining the left and right boundaries of the absorption valley is to delineate the effective area for feature extraction and eliminate interference from irrelevant bands. The shoulder points usually correspond to the inflection points or local maxima of the spectral curve on both sides of the absorption valley, representing the start and end positions of the absorption features. Monotonicity judgment can prevent small noise fluctuations within the valley from being misjudged as boundaries, ensuring that the selected range completely covers the entire concave shape. For example, after determining 660nm as the valley bottom, searching to the left reveals that the reflectivity continuously increases to 0.30 when the wavelength decreases to 640nm, and then begins to decrease further to the left, so 640nm is determined as the left boundary; similarly, searching to the right reveals that the reflectivity reaches 0.32 and an inflection point appears at 680nm, so 680nm is determined as the right boundary, thus obtaining the band range corresponding to the absorption valley from 640nm to 680nm.

[0099] Extract all reflectance data points within the corresponding band range of the absorption valley as a point set;

[0100] It should be noted that this step discretizes the continuous spectral curve into a computer-processable two-dimensional coordinate point set, where the horizontal axis represents wavelength and the vertical axis represents reflectance. This point set contains all the morphological details inside the absorption valleys and serves as the data basis for subsequently constructing the geometric envelope. For example, the system reads out all the data sampled every 2nm in the range of 640nm to 680nm, obtaining wavelength-reflectance pairs of 21 band points including the start point, valley bottom point, and end point, forming the data point set to be processed.

[0101] Starting from the left and right boundary points of the point set, check the intermediate band points in turn. If the current point is below the straight line formed by connecting the existing envelope points, retain the existing envelope. If it is above the straight line, add the point to the envelope and remove the intermediate point whose convexity has been destroyed. Repeat this process until all points are below or on the envelope line, forming an upward convex envelope line.

[0102] It should be noted that the convex envelope algorithm is a geometric approximation method designed to simulate the background spectral continuum assuming no absorption features. This process is analogous to placing a taut rope over the spectral curve; only the prominent peaks support the rope, and these points are called envelope points. By eliminating points in the depressions, the constructed line best fits the top contour of the spectral curve, serving as a reference for removing background influences. For example, connecting the left boundary point A and the right boundary point B forms a straight line. If the reflectance of a certain band point C is higher than the value of the straight line at the corresponding wavelength, it indicates that point C is more prominent than the straight line. In this case, the system updates the envelope to two broken lines connecting point A and point C, and then connecting point C and point B, and continues to check other points, ultimately forming a broken line consisting of several line segments connected end-to-end that covers all data points as the convex envelope.

[0103] On the convex envelope, the reflectance value of the envelope corresponding to the valley bottom band position is taken as the reference reflectance outside the valley.

[0104] It should be noted that the valley point is usually located deep in the concave part of the spectral curve, not on the envelope. Therefore, its corresponding reference value needs to be obtained through calculation. Since the envelope is composed of line segments, the vertical coordinate value corresponding to the wavelength on the envelope can be calculated using linear interpolation based on the line segment interval where the valley wavelength is located. This value represents the theoretical background reflectance when no absorption occurs. For example, if the valley is located at 660 nm, and the envelope near 660 nm is a line segment connecting the envelope point at 650 nm (reflectance 0.31) and the envelope point at 670 nm (reflectance 0.33), the system calculates the value corresponding to 660 nm on this line segment as 0.32 based on linear proportions, and uses this 0.32 as the reference reflectance outside the valley at 660 nm.

[0105] Calculate the valley depth, where the valley depth is the difference between the reference reflectance outside the valley and the reflectance value at the valley bottom;

[0106] It should be noted that valley depth is a key physical quantity characterizing the absorption intensity of the target substance. It eliminates the influence of background lighting conditions and soil matrix brightness on the absolute value of reflectance. By subtracting the actual observed value from the baseline background value, the degree of spectral energy attenuation caused by the presence of a specific substance can be quantified. This depth value is usually positively correlated with the concentration of the substance. For example, the baseline reflectance outside the valley calculated above is 0.32, while the actual measured valley bottom reflectance value is 0.15. The system subtracts the two to calculate the valley bottom depth of the absorption valley as 0.17. This value will then be used as the input feature for constructing the environmental parameter inversion model.

[0107] In an optional embodiment, starting from the lowest reflectance point at the valley floor, reflectance data points in adjacent bands are independently scanned to the left and right bands respectively. The percentage difference between each data point and the valley floor reflectance value is calculated to obtain a left-side percentage difference sequence and a right-side percentage difference sequence, including:

[0108] The band number corresponding to the point with the lowest reflectivity at the bottom of the valley is determined as the scanning center.

[0109] It should be noted that the band number is the index of the spectrometer's detection unit in the spectral dimension, directly corresponding to the specific wavelength position; determining the scanning center is to establish a relative coordinate system, and all subsequent calculations will be carried out from this center as the origin, thereby realizing independent analysis of the left and right wings of the absorption valley; for example, in spectral data containing 200 bands, if the system identifies the point with the lowest reflectance at the bottom of the valley as being located in the 100th band, corresponding to a wavelength of 660nm, then the 100th band is marked as the scanning center;

[0110] Based on the scanning center, a preset number of band reflectance data are sequentially acquired along the direction of decreasing band number as the left scanning dataset, and a preset number of band reflectance data are sequentially acquired along the direction of increasing band number as the right scanning dataset.

[0111] It should be noted that the preset number defines the maximum window range for feature search. This range should be large enough to cover the expected maximum absorption valley width to prevent wingspan truncation due to an excessively small window. Acquiring data separately for the left and right sides is to capture the different trends of the spectral curve on both sides of the valley, which is the basis for handling asymmetric absorption valleys. For example, if the system sets the preset number of bands for a single-side scan to 50, then the reflectance data read in reverse order from band 99 to band 50 constitutes the left scan dataset, while the reflectance data read in forward order from band 101 to band 150 constitutes the right scan dataset.

[0112] Subtract the valley reflectance value from each reflectance data in the left scan dataset and the right scan dataset to obtain the left reflectance difference set and the right reflectance difference set;

[0113] It should be noted that the difference calculation is to eliminate the influence of the absolute magnitude of the background reflectance and focus the attention on the rise of the spectral curve relative to the valley. This step quantifies the degree to which each band point is higher than the valley in absolute value, representing the absolute morphological change of the absorption feature edge. For example, if the valley reflectance value is known to be 0.15 and the reflectance of an adjacent band point in the left scan dataset is 0.18, then the system subtracts 0.15 from 0.18 to obtain a reflectance difference of 0.03 for that point. Performing this operation on all points yields the difference set.

[0114] Divide each difference in the set of reflectivity differences on the left and the set of reflectivity differences on the right by the reflectivity value at the bottom of the valley to obtain the relative proportion coefficient on the left and the relative proportion coefficient on the right.

[0115] It should be noted that this step normalizes the data; converting the absolute difference into a proportional value relative to the valley depth eliminates the dimensional differences caused by the different reflectance bases of different target objects (e.g., light soil and dark soil), making absorption valleys of different intensities comparable in terms of steepness, thus making subsequent threshold judgments more robust; for example, the difference of 0.03 calculated above, divided by the valley reflectance value of 0.15, yields 0.20, which means that the reflectance at this point is 20% higher than that at the valley bottom, and this value of 0.20 is the corresponding relative proportion coefficient;

[0116] According to the band number in order of distance from the scanning center from near to far, the relative proportion coefficients on the left and right sides are arranged respectively to obtain the left difference proportion sequence and the right difference proportion sequence.

[0117] It is important to note that the order of the sequence is crucial, as it simulates the process of "walking" outward from the center of the valley. Arranging the sequence from closest to furthest ensures that subsequent algorithms, when searching for boundaries, always detect the point closest to the valley first, stopping once a threshold condition is met. This accurately locates the inner edge of the absorption valley and avoids misjudging distant noise points. For example, the proportion coefficient of the first adjacent point on the left (distance 11) is 0.05, the proportion coefficient of the second point (distance 2) is 0.12, and the proportion coefficient of the third point (distance 3) is 0.20. The system arranges these in order as [0.05, 0.12, 0.20, ...], which constitutes the left-side difference proportion sequence used for subsequent analysis.

[0118] In an optional embodiment, a predetermined number of adjacent band points are continuously selected from the lowest reflectivity point at the bottom of the valley to the left and to the right as initial scan segments. The average slope is calculated within each initial scan segment to obtain the initial average slope on the left and the initial average slope on the right, including:

[0119] Set a fixed number of band points as the initial scan window length;

[0120] It should be noted that the initial scan window length defines the size of the band range used to calculate the local slope; the choice of this length needs to strike a balance between noise resistance and morphological sensitivity. If the window is too short, the slope calculation is easily affected by single-point noise and will fluctuate drastically; if the window is too long, it may include too many gentle areas far from the valley bottom, making it impossible to accurately reflect the true steepness near the valley bottom. It is usually set according to the spectral resolution of the spectrometer and the average half-width of the target absorption characteristics. It is recommended to set the value to 2 to 3 times the number of bands corresponding to the spectral resolution. For example, for data with a spectral resolution of 2 nm, in order to capture the rapidly changing characteristics near the valley bottom, the system sets a fixed number of band points of 5 as the initial scan window length.

[0121] Using the band position where the lowest reflectivity point at the bottom of the valley is located as the starting anchor point, a number of consecutive band data equal to the length of the initial scanning window are intercepted along the direction of decreasing band number to form the initial scanning segment on the left; a number of consecutive band data equal to the length of the initial scanning window are intercepted along the direction of increasing band number to form the initial scanning segment on the right.

[0122] It should be noted that this step is to clarify the specific data subset used for slope calculation; extending to both sides with the valley bottom as the anchor point ensures that the calculated slope is the slope of the rising section immediately adjacent to the valley bottom, which can truly reflect the geometric opening shape of the absorption valley bottom; the left and right sides are intercepted separately so that the slopes on both sides can be calculated independently later, thus preserving asymmetry information; for example, if the lowest reflectivity point at the valley bottom is known to be located in band 100 (corresponding to a wavelength of 660nm), the system intercepts data from bands 99 to 95 in the left direction where the band number decreases, forming the initial scan segment on the left; at the same time, it intercepts data from bands 101 to 105 in the right direction where the band number increases, forming the initial scan segment on the right.

[0123] In the initial scan segments on the left and right, the band points farthest from the starting anchor point are identified and used as the left edge point and the right edge point, respectively.

[0124] It should be noted that the edge point is the end of the local window, and together with the starting anchor point (valley bottom), it forms the two endpoints defining the local average slope. Identifying the edge point is to determine the endpoint of the secant line. By calculating the slope of the line connecting these two endpoints, the average rate of change of the curve segment can be approximated. For example, in the initial scan segment on the left above, band 95 is the farthest from the starting point band 100 (wavelength 650nm), and it is marked as the left edge point; in the initial scan segment on the right, band 105 is the farthest from the starting point (wavelength 670nm), and it is marked as the right edge point.

[0125] Calculate the difference in reflectivity and wavelength between the left edge point and the starting anchor point, and the difference in reflectivity and wavelength between the right edge point and the starting anchor point, respectively.

[0126] It should be noted that this step prepares the numerator and denominator for slope calculation; the difference in reflectance values ​​represents the height of the spectral curve "climbing" within the scanning window (change on the vertical axis), and the difference in wavelength values ​​represents the corresponding wavelength span (change on the horizontal axis); the numerical difference usually refers to the absolute difference, focusing on the magnitude of the change rather than the directional sign, to quantify the steepness; for example, assuming the reflectance at the valley bottom (starting anchor point) is 0.15, the reflectance at the left edge point is 0.20, and the reflectance at the right edge point is 0.25; the system calculates the left reflectance difference as 0.20 minus 0.15, resulting in 0.05, and the left wavelength difference as 660nm minus 650nm, resulting in 10nm; similarly, the right reflectance difference is calculated as 0.25 minus 0.15, resulting in 0.10, and the right wavelength difference as 670nm minus 660nm, resulting in 10nm;

[0127] Divide the difference in reflectivity values ​​at the left edge points by the difference in their corresponding wavelength values ​​to obtain the initial average slope on the left.

[0128] It should be noted that the slope is a physical quantity that reflects the steepness of the spectral curve; the initial average slope on the left quantifies the degree of abruptness of the left wall of the absorption valley extending upwards from the valley floor; the larger this value, the more drastic the change in absorption characteristics on the left, and the closer the corresponding absorption boundary may be to the valley floor; for example, dividing the above-mentioned left reflectance difference of 0.05 by the left wavelength difference of 10 nm yields a value of 0.005, which is the initial average slope on the left, indicating that the reflectance on the left increases by an average of 0.005 per nanometer wavelength;

[0129] Divide the difference in reflectivity values ​​at the right edge points by the difference in their corresponding wavelength values ​​to obtain the initial average slope on the right side.

[0130] It should be noted that, similarly, the initial average slope on the right quantifies the steepness of the right side wall of the absorption valley; independently calculating the right slope allows the algorithm to distinguish between asymmetrical absorption patterns (such as a gentle left and a steep right); these slope values ​​will then directly participate in the generation of the dynamic threshold, determining the search slackness of the left and right wingspan boundaries; for example, dividing the above-mentioned right reflectivity difference of 0.10 by the right wavelength difference of 10nm yields a value of 0.01, which is the initial average slope on the right, indicating that the reflectivity on the right side increases by an average of 0.01 per nanometer wavelength, indicating that in this example, the right valley wall is steeper than the left.

[0131] In an optional embodiment, a left dynamic threshold and a right dynamic threshold are generated based on the normalized products of the valley depth and the reciprocals of the initial average slopes of the left and right sides, respectively, including:

[0132] Obtain the initial average slope on the left and the initial average slope on the right respectively, and calculate the mathematical reciprocal of their values ​​to obtain the reciprocal of the slope on the left and the reciprocal of the slope on the right.

[0133] It should be noted that calculating the reciprocal of the slope is to transform the indicator measuring steepness into an indicator measuring gentleness or potential width; the smaller the slope, the gentler the curve change, and the larger its reciprocal, the wider the flank may extend. This step transforms the derivative characteristics of the geometry into linear scaling characteristics, providing a basis for estimating the horizontal span of the absorption valley; for example, in the previous embodiment, the initial average slope of the left side was calculated to be 0.005, and the initial average slope of the right side was 0.01; the system calculates the reciprocal of the slope of the left side as 1 divided by 0.005 to get 200, and calculates the reciprocal of the slope of the right side as 1 divided by 0.01 to get 100;

[0134] Multiply the reciprocal of the slope on the left and the reciprocal of the slope on the right by the valley depth to obtain the product of the slope on the left and the product of the slope on the right.

[0135] It should be noted that the valley depth represents the intensity of absorption, while the inverse of the slope represents the breadth of absorption. The product of the two (i.e., depth divided by slope) is physically approximated by the width of the base of a characteristic triangle. It reflects the theoretically expected characteristic wingspan of the absorption valley at the current depth and steepness. The larger this product, the deeper and wider the absorption valley, and the higher the tolerance should be when searching the boundary. For example, given a valley depth of 0.17, multiplying it by the inverse of the slope on the left (200) gives a depth-slope product of 34; multiplying it by the inverse of the slope on the right (100) gives a depth-slope product of 17.

[0136] Normalization operations are applied to the left depth-slope product and the right depth-slope product respectively, and the product results are converted into numerical values ​​with the same dimensions as the difference proportion sequence to obtain the left dynamic threshold and the right dynamic threshold.

[0137] It should be noted that the product obtained in the previous step has the dimension of wavelength and must be converted. In this embodiment, a fixed empirical proportionality coefficient k=0.05 is used for conversion. This coefficient 0.05 is determined based on the statistical results of a large number of soil sample spectral data. Under this coefficient, the algorithm can optimally balance the sensitivity and stability of the wingspan boundary. The calculation formula is: dynamic threshold = depth slope product × 0.05. For example, for the left side, multiplying the product 34 by the fixed coefficient 0.05, the dynamic threshold for the left side is calculated to be 1.7. For the right side, multiplying the product 17 by the fixed coefficient 0.05, the dynamic threshold for the right side is calculated to be 0.85. These two determined values ​​(1.7 and 0.85) will be used as the only threshold values ​​for determining the left and right wingspan boundaries.

[0138] In an optional embodiment, the initial wingspan boundary is obtained by finding the location of the data point that first exceeds the left dynamic threshold in the left difference proportion sequence as the left wingspan boundary, and finding the location of the data point that first exceeds the right dynamic threshold in the right difference proportion sequence as the right wingspan boundary, including:

[0139] Read each percentage value in the left-hand difference percentage sequence in order of distance from the lowest reflectivity point at the bottom of the valley, from near to far.

[0140] Each percentage value read is compared with the dynamic threshold on the left to determine whether the current percentage value is greater than the dynamic threshold on the left.

[0141] When the current percentage value is determined to be greater than the left dynamic threshold for the first time, the reading stops and the band position corresponding to the percentage value is recorded as the left wingspan boundary.

[0142] It should be noted that the search process follows an inside-out principle, that is, scanning point by point from the center of the absorption valley towards the short-wavelength direction of the spectrum. This is to ensure that the effective boundary closest to the valley bottom is found, preventing misjudgment of the boundary due to accidental noise far from the absorption center or interference from other adjacent absorption features. "First exceedance" means that at this point, the relative increase in spectral reflectance relative to the valley bottom has reached the dynamic limit set according to the current depth and slope, indicating that the spectral curve has morphologically left the core influence region of the absorption valley and entered the background or transition region. The establishment of the left wingspan boundary provides the left-side coordinates for subsequent calculation of the asymmetric width. The system uses a benchmark; for example, given that the left dynamic threshold generated in the previous embodiment is 1.7, the values ​​in the left difference percentage sequence are arranged from near to far from the valley bottom as 0.5, 1.2, 1.8, 2.1, etc.; the system first reads the first value 0.5, compares it with 1.7, finds that 0.5 is less than 1.7, and continues reading; then it reads the second value 1.2, which is still less than 1.7, and continues reading; it reads the third value 1.8, finds that 1.8 is greater than 1.7, and meets the stopping condition; at this time, the system immediately stops scanning, queries the wavelength position corresponding to the value 1.8, which is 654nm, and records 654nm as the left wingspan boundary;

[0143] Read each percentage value in the difference percentage sequence on the right in order of distance from the lowest reflectivity point at the bottom of the valley, from near to far.

[0144] Each percentage value read is compared with the dynamic threshold on the right to determine whether the current percentage value is greater than the dynamic threshold on the right.

[0145] When the current percentage value is determined to be greater than the dynamic threshold on the right for the first time, the reading stops and the band position corresponding to the percentage value is recorded as the right wingspan boundary.

[0146] It should be noted that the search logic on the right side is completely independent and parallel to that on the left side, but it is based on a dynamic threshold specific to the right side. This independence is the core of this solution for handling asymmetric absorption valleys, because absorption valleys of actual ground features often exhibit a shape that is wider on the left and narrower on the right or vice versa. Due to the steeper slope on the right side, the generated threshold is lower, which makes the judgment condition of the right boundary more stringent. It usually finds the boundary point faster than on the gentler side, thus realistically restoring the geometric feature of the absorption valley's steep descent on the right side. For example, it is known that the dynamic threshold generated on the right side in the previous embodiment is 0.85, and the values ​​in the right difference percentage sequence are arranged from near to far from the valley bottom as 0.4, 0.9, 1.3, etc. The system first reads the first value 0.4, which is less than 0.85, and continues. It reads the second value 0.9 and finds that 0.9 is greater than 0.85, satisfying the stopping condition. At this time, the system stops scanning and queries the wavelength position corresponding to the value 0.9, which is 664nm. Therefore, 664nm is recorded as the right wingspan boundary.

[0147] The band range formed by the left and right wingspan boundaries is defined as the primary wingspan boundary.

[0148] It should be noted that the initial wingspan boundary is a closed spectral range defined by the two independently found wavelength positions mentioned above. This range is not the final result, but rather serves as the basis for the next round of iterative calculations. Data within this range is considered to be the most representative effective data segment of the current absorption valley core morphology. By defining this specific range, subsequent steps can recalculate the average slope only for the spectral data within this range, thereby eliminating interference from irrelevant data outside the range and making the slope calculation closer to the actual valley wall slope. For example, combining the above results, the system combines 654nm on the left and 664nm on the right to determine the band range of 654nm to 664nm as the initial wingspan boundary of the absorption valley. Compared to the initially broad scanning range, this initial boundary has preliminarily locked the effective outline of the absorption valley.

[0149] In an optional embodiment, the left and right average slopes are recalculated based on reflectivity data within the initial wingspan boundary, the dynamic threshold is updated, and the boundary search step is repeated until the boundary position converges and stabilizes, resulting in the final asymmetric wingspan, including:

[0150] Use the band range determined by the initial wingspan boundary as the current reference range;

[0151] It should be noted that the iterative algorithm requires an initial reference state. Although the initial wingspan boundary may be rough, it has already provided a range description that is closer to the real shape than a fixed window. Setting this range as the current reference range means that the subsequent calculations will trust and self-correct based on the spectral geometric features within this range. For example, the band interval formed by the left wingspan boundary of 654nm and the right wingspan boundary of 664nm determined in the previous embodiment is set as the current reference range for the first round of iteration.

[0152] The average slope is recalculated using reflectance data within the current reference range, and updated left and right dynamic thresholds are generated accordingly.

[0153] It should be noted that the initial slope is calculated based on a fixed small window, which may only reflect the gentle changes within a very small range at the bottom of the valley. The slope calculated based on the current reference range, however, is a "secant slope," which utilizes the overall span from the bottom of the valley to the current boundary, and can more accurately characterize the average steepness of the entire absorption valley wall. Regenerating the threshold using the updated slope demonstrates the algorithm's adaptability: if the actual valley wall is found to be steeper than expected, resulting in a larger slope, the calculated threshold will automatically decrease, thus tightening the boundary in the next round to prevent overflow. For example, using the current left boundary data at 654nm (reflectivity 0.27 higher than the valley bottom) and the valley bottom at 660nm, the new slope on the left is calculated as 0.27 divided by 6nm, yielding 0.045. Similarly, the new slope on the right is calculated as 0.135 divided by 4nm, yielding 0.03375. Compared to the initial slope, the new slope is significantly larger. Substituting these values ​​into the formula, the new threshold on the left is updated to approximately 0.19, and the new threshold on the right is updated to approximately 0.25.

[0154] Based on the updated left and right dynamic thresholds, the new left wingspan boundary and the new right wingspan boundary are obtained by searching again in the left difference proportion sequence and the right difference proportion sequence, respectively.

[0155] It should be noted that this step uses a revised new threshold to remeasure the spectral data; because the threshold has changed, the position where the stopping condition is met may also change accordingly; this simulates a dynamic approximation process: estimating the slope with a broad boundary, finding the slope to be very steep, so tightening the threshold to find a narrower boundary; or finding the slope to be very gentle, relaxing the threshold to find a wider boundary; for example, in the left difference percentage sequence [0.5, 1.2, 1.8, ...], using the new threshold 0.19 for searching; the first value 0.5 is greater than 0.19, so the search immediately stops at the first point, i.e., 658nm; similarly, using the new threshold 0.25 in the right sequence [0.4, 0.9, ...], the first value 0.4 is greater than 0.25, stopping at 662nm; thus obtaining the new left wingspan boundary 658nm and the new right wingspan boundary 662nm;

[0156] Compare the new left and right wingspan boundaries with the boundary positions of the current reference range. If the positions have changed, use the new left and right wingspan boundaries as the new current reference range and return to the step of generating the updated dynamic threshold.

[0157] It should be noted that a change in position indicates that the current morphological description (slope and boundary) has not yet reached a self-consistent state and needs further correction. Through repeated iterations, the boundary position will gradually find a balance between "morphological features" and "discrimination thresholds." To prevent infinite loop oscillations under noise interference, the system also sets a maximum number of iterations (e.g., 10 times) as a forced stopping condition. For example, a comparison reveals that the new left boundary of 658nm is inconsistent with the old 654nm, and the new right boundary of 662nm is inconsistent with the old 664nm, indicating that the boundaries are not yet stable. The system updates 658nm and 662nm to the new current reference range and recalculates the slope and threshold using the data in this new range, entering the next round of iteration.

[0158] If the position does not change, stop the iteration, determine the left and right wingspan boundaries at this time as the final boundaries, and calculate the wavelength difference between the two to obtain the final asymmetric wingspan.

[0159] It should be noted that convergence and stability mean that the algorithm has found the feature boundary that best matches the actual shape of the absorption valley. At this point, the threshold calculated by the slope points exactly to the current boundary position, reaching a mathematically fixed point. The final asymmetric wingspan is the most robust geometric description of the absorption features within this band, eliminating subjective errors caused by manually setting a fixed threshold. For example, suppose that after a subsequent iteration, the calculated boundary position remains at 658nm on the left and 662nm on the right, then the algorithm is considered to have converged. The system calculates the final asymmetric wingspan as 662nm minus 658nm, resulting in 4nm, and outputs this value as a key input feature for the subsequent inversion model.

[0160] In an optional embodiment, normalized absorption morphology features are calculated based on the final asymmetric wingspan and valley depth; these normalized absorption morphology features are input into a pre-trained inversion model to obtain quantitative inversion results of the target object's environmental parameters, including:

[0161] By combining the valley depth with the final asymmetric wingspan, a numerical value that can characterize the geometry of the absorption valley is constructed, and the normalized absorption morphology features are obtained.

[0162] It should be noted that the valley depth reflects the intensity of the strongest absorption point, while the asymmetric wingspan reflects the coverage of the absorption feature along the wavelength axis. The combination of the two (usually in product form or area approximation form) can approximate the spectral absorption area. This area physical quantity has a more stable linear or nonlinear correlation with the measured environmental parameters compared to depth or width alone. Normalization here refers to the fact that this feature is calculated based on the normalized spectral depth after removing the envelope, thus eliminating the interference of background illumination and terrain scattering, and purely reflecting the absorption characteristics of the material itself. For example, in the above embodiment, the obtained valley depth is 0.17, and the final asymmetric wingspan is 4nm. The system uses the rectangular area approximation formula for combined calculation, that is, calculates the product of depth and width, and obtains a normalized absorption morphology feature value of 0.68. This value of 0.68 is the quantified total absorption intensity index.

[0163] The pre-built and trained environmental parameter inversion model is invoked, and the model establishes a mapping relationship between spectral morphological features and environmental parameters through machine learning algorithms;

[0164] The normalized absorption morphology characteristics are used as input data and fed into the environmental parameter inversion model for calculation.

[0165] It should be noted that the environmental parameter inversion model needs to be constructed and trained before actual use. The specific implementation steps are as follows: Select a preset number (e.g., 500) of similar target objects as a sample set, use standard chemical analysis methods to determine the real environmental parameters (such as organic matter content) corresponding to each sample as label data, and simultaneously acquire its reflectance spectral data; then, using the methods described in steps S1 to S6 above, extract the normalized absorption morphology features (i.e., the product of valley depth and final asymmetric wingspan) of each sample one by one to construct an input feature vector set; based on this, use the radial basis function kernel support vector regression (SVR) algorithm to establish an initial model, set the penalty coefficient C=100 and the kernel function parameter γ=0.1, and combine the feature vector set with the label data. The data is divided into training and testing sets according to a preset ratio (e.g., 8:2). The model is iteratively trained using the training set data until the loss function converges, thereby establishing a stable nonlinear mapping relationship between spectral morphological features and environmental parameters. The trained model parameters are then fixed and saved for subsequent steps. In addition, input calculation refers to passing the feature values ​​calculated in the previous step to the algorithm engine according to the input format required by the model (e.g., vector or matrix form) to trigger the model's prediction and inference process. The system calls the SVR soil organic matter inversion model that has been trained and whose parameters are fixed (C=100, γ=0.1). The normalized absorption morphological feature value 0.68 is calculated and constructed as a feature vector [0.68], which is then input into the model.

[0166] Obtain the output values ​​of the environmental parameter inversion model, and determine these output values ​​as the quantitative inversion results of the environmental parameters of the target object;

[0167] It should be noted that the output value is an estimated value of environmental parameters predicted by the model based on the input features; this value has a clear physical dimension and directly represents the current environmental attribute state of the target object; this is the final output of the entire data processing flow and can be directly used for subsequent applications such as agricultural monitoring and environmental assessment; for example, after internal calculation, the SVR soil organic matter inversion model outputs a predicted value of 15.5; the system, combined with the preset unit definition, interprets this result as the organic matter content of the target soil being 15.5 g / kg, and outputs it to the user as the final quantitative inversion result.

[0168] Example 2, please refer to Figure 2 This invention provides a technical solution: a quantitative inversion system for environmental parameters based on spectral morphology characteristics, applicable to the aforementioned quantitative inversion method for environmental parameters based on spectral morphology characteristics, comprising:

[0169] The depth extraction module 1 is used to acquire the reflectance spectral data of the target object, identify absorption valleys in the reflectance spectral data and determine the lowest reflectance point and valley bottom reflectance value; perform convex envelope fitting on the corresponding band range of the absorption valley to obtain the envelope reflectance of the valley bottom band position as the reference reflectance outside the valley, and calculate the valley bottom depth.

[0170] Sequence construction module 2 is used to independently scan the reflectance data points of adjacent bands to the left and right bands, starting from the lowest reflectance point at the bottom of the valley, and calculate the proportion of the difference between each data point and the reflectance value at the bottom of the valley to obtain the left difference proportion sequence and the right difference proportion sequence.

[0171] The slope initial calculation module 3 is used to continuously take a preset number of adjacent band points from the lowest reflectivity point at the bottom of the valley to the left and right as initial scanning segments, respectively, and calculate the average slope in each initial scanning segment to obtain the initial average slope on the left and the initial average slope on the right.

[0172] Threshold generation module 4 is used to generate left dynamic threshold and right dynamic threshold based on the normalized product of valley depth and the reciprocal of the initial average slope on the left and the reciprocal of the initial average slope on the right, respectively.

[0173] The initial boundary positioning module 5 is used to find the position of the data point that first exceeds the left dynamic threshold in the left difference proportion sequence as the left wingspan boundary, and to find the position of the data point that first exceeds the right dynamic threshold in the right difference proportion sequence as the right wingspan boundary, so as to obtain the initial wingspan boundary.

[0174] Iterative convergence module 6 is used to recalculate the left and right average slopes based on the reflectivity data within the initial wingspan boundary, update the dynamic threshold, and repeat the boundary search steps until the boundary position converges and stabilizes, thus obtaining the final asymmetric wingspan.

[0175] Model inversion module 7 is used to calculate normalized absorption morphology features based on the final asymmetric wingspan and valley depth; the normalized absorption morphology features are input into the pre-trained inversion model to obtain quantitative inversion results of the environmental parameters of the target object.

[0176] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for quantitative retrieval of environmental parameters based on spectral morphological features, characterized in that, include: Acquire the reflectance spectral data of the target object, identify absorption valleys in the reflectance spectral data, and determine the lowest reflectance point and valley reflectance value at the bottom of the valley; A convex envelope is fitted to the corresponding band range of the absorption valley to obtain the envelope reflectance of the band position at the bottom of the valley as the reference reflectance outside the valley, and the depth of the valley bottom is calculated. Starting from the lowest reflectance point at the bottom of the valley, the reflectance data points of adjacent bands are scanned independently to the left and right bands respectively. The proportion of the difference between each data point and the reflectance value at the bottom of the valley is calculated to obtain the left difference proportion sequence and the right difference proportion sequence. Starting from the lowest reflectivity point at the bottom of the valley, a predetermined number of adjacent band points are taken to the left and right as initial scan segments. The average slope is calculated within each initial scan segment to obtain the initial average slope on the left and the initial average slope on the right. Based on the normalized product of the valley depth with the reciprocal of the initial average slope on the left and the reciprocal of the initial average slope on the right, respectively, the left dynamic threshold and the right dynamic threshold are generated. Find the data point position that first exceeds the left dynamic threshold in the left difference proportion sequence as the left wingspan boundary, and find the data point position that first exceeds the right dynamic threshold in the right difference proportion sequence as the right wingspan boundary to obtain the initial wingspan boundary; Based on the reflectivity data within the initial wingspan boundary, the left and right average slopes are recalculated, the dynamic threshold is updated, and the boundary search steps are repeated until the boundary position converges and stabilizes, thus obtaining the final asymmetric wingspan. Normalized absorption morphology features were calculated based on the final asymmetric wingspan and valley depth. By inputting the normalized absorption morphology features into the pre-trained inversion model, the quantitative inversion results of the environmental parameters of the target object are obtained.

2. The method according to claim 1, wherein, Identify absorption valleys in reflectance spectral data and determine the lowest reflectance point and valley reflectance value at the valley bottom, including: A hyperspectral imager is used to perform multi-band spectral measurements on the target object, and the original spectral signals of the target object in the visible to near-infrared band range are collected to obtain the original reflectance spectral data containing multiple bands. The raw reflectance spectral data is smoothed, denoised, and radiometrically corrected to obtain preprocessed reflectance spectral data. In the preprocessed reflectance spectral data, adjacent reflectance values ​​are compared band by band to identify continuous bands that first decrease and then increase in reflectance and meet preset depth and width thresholds as potential absorption valleys. Within each potential absorption valley, the band point with the lowest reflectance value is identified as the valley bottom lowest reflectance point, and the reflectance value of that point is recorded as the valley bottom reflectance value.

3. The method according to claim 2, wherein, A convex envelope is fitted to the corresponding band range of the absorption valley to obtain the envelope reflectance at the valley bottom band position as the reference reflectance outside the valley. The valley bottom depth is then calculated, including: Extend the bands to the left and right sides from the lowest reflectivity point at the bottom of the valley until a shoulder point is found where the reflectivity is higher than the reflectivity value at the bottom of the valley and the reflectivity of subsequent bands increases monotonically. Determine the band position corresponding to the shoulder point as the left and right boundaries of the absorption valley, thereby obtaining the band range corresponding to the absorption valley. Extract all reflectance data points within the corresponding band range of the absorption valley as a point set; Starting from the left and right boundary points of the point set, check the intermediate band points in turn. If the current point is below the straight line formed by connecting the existing envelope points, retain the existing envelope. If it is above the straight line, add the point to the envelope and remove the intermediate point whose convexity has been destroyed. Repeat this process until all points are below or on the envelope line, forming an upward convex envelope line. On the convex envelope, the reflectance value of the envelope corresponding to the valley bottom band position is taken as the reference reflectance outside the valley. Calculate the valley depth, where the valley depth is the difference between the reference reflectance outside the valley and the reflectance value at the valley bottom.

4. The method according to claim 3, wherein, Starting from the lowest reflectance point at the valley floor, reflectance data points in adjacent bands are independently scanned to the left and right bands respectively. The proportion of the difference between each data point and the valley floor reflectance value is calculated, resulting in a left-side difference proportion sequence and a right-side difference proportion sequence, including: The band number corresponding to the point with the lowest reflectivity at the bottom of the valley is determined as the scanning center. Based on the scanning center, a preset number of band reflectance data are sequentially acquired along the direction of decreasing band number as the left scanning dataset, and a preset number of band reflectance data are sequentially acquired along the direction of increasing band number as the right scanning dataset. Subtract the valley reflectance value from each reflectance data in the left scan dataset and the right scan dataset to obtain the left reflectance difference set and the right reflectance difference set; Divide each difference in the set of reflectivity differences on the left and the set of reflectivity differences on the right by the reflectivity value at the bottom of the valley to obtain the relative proportion coefficient on the left and the relative proportion coefficient on the right. According to the band number in order of distance from the scanning center from near to far, the relative proportion coefficients on the left and right sides are arranged respectively to obtain the left difference proportion sequence and the right difference proportion sequence.

5. The method according to claim 4, wherein, Starting from the lowest reflectivity point at the bottom of the valley, a predetermined number of adjacent band points are continuously selected to the left and right as initial scan segments. The average slope is calculated within each initial scan segment to obtain the initial average slope on the left and the initial average slope on the right, including: Set a fixed number of band points as the initial scan window length; Starting from the band position where the lowest reflectivity point at the bottom of the valley is located, take the number of consecutive band data equal to the length of the initial scanning window along the direction of decreasing band number to form the left initial scanning segment; take the number of consecutive band data equal to the length of the initial scanning window along the direction of increasing band number to form the right initial scanning segment. In the initial scan segments on the left and right, the band points farthest from the starting anchor point are identified and used as the left edge point and the right edge point, respectively. Calculate the difference in reflectivity and wavelength between the left edge point and the starting anchor point, and the difference in reflectivity and wavelength between the right edge point and the starting anchor point, respectively. Divide the difference in reflectivity values ​​at the left edge points by the difference in their corresponding wavelength values ​​to obtain the initial average slope on the left. Divide the difference in reflectivity values ​​at the right edge points by the difference in their corresponding wavelength values ​​to obtain the initial average slope on the right side.

6. The method according to claim 5, wherein, Based on the normalized products of the valley depth and the reciprocals of the initial average slopes on the left and right sides, respectively, dynamic thresholds for the left and right sides are generated, including: Obtain the initial average slope on the left and the initial average slope on the right respectively, and calculate the mathematical reciprocal of their values ​​to obtain the reciprocal of the slope on the left and the reciprocal of the slope on the right. Multiply the reciprocal of the slope on the left and the reciprocal of the slope on the right by the valley depth to obtain the product of the slope on the left and the product of the slope on the right. Normalization operations are applied to the left depth-slope product and the right depth-slope product respectively, and the product results are converted into numerical values ​​with the same dimensions as the difference proportion sequence to obtain the left dynamic threshold and the right dynamic threshold.

7. The method according to claim 6, wherein, The initial wingspan boundary is obtained by finding the data point position in the left difference proportion sequence that first exceeds the left dynamic threshold, and the data point position in the right difference proportion sequence that first exceeds the right dynamic threshold, which includes: Read each percentage value in the left-hand difference percentage sequence in order of distance from the lowest reflectivity point at the bottom of the valley, from near to far. Each percentage value read is compared with the dynamic threshold on the left to determine whether the current percentage value is greater than the dynamic threshold on the left. When the current percentage value is determined to be greater than the left dynamic threshold for the first time, the reading stops and the band position corresponding to the percentage value is recorded as the left wingspan boundary. Read each percentage value in the difference percentage sequence on the right in order of distance from the lowest reflectivity point at the bottom of the valley, from near to far. Each percentage value read is compared with the dynamic threshold on the right to determine whether the current percentage value is greater than the dynamic threshold on the right. When the current percentage value is determined to be greater than the dynamic threshold on the right for the first time, the reading stops and the band position corresponding to the percentage value is recorded as the right wingspan boundary. The band range formed by the left and right wingspan boundaries is defined as the primary wingspan boundary.

8. The method for quantitative inversion of environmental parameters based on spectral morphology characteristics according to claim 7, characterized in that, Based on the reflectivity data within the initial wingspan boundary, the left and right average slopes are recalculated, the dynamic threshold is updated, and the boundary search step is repeated until the boundary position converges and stabilizes, yielding the final asymmetric wingspan, including: Use the band range determined by the initial wingspan boundary as the current reference range; The average slope is recalculated using reflectance data within the current reference range, and updated left and right dynamic thresholds are generated accordingly. Based on the updated left and right dynamic thresholds, the new left wingspan boundary and the new right wingspan boundary are obtained by searching again in the left difference proportion sequence and the right difference proportion sequence, respectively. Compare the new left and right wingspan boundaries with the boundary positions of the current reference range. If the positions have changed, use the new left and right wingspan boundaries as the new current reference range and return to the step of generating the updated dynamic threshold. If the position does not change, stop the iteration, determine the left and right wingspan boundaries at this time as the final boundaries, and calculate the wavelength difference between the two to obtain the final asymmetric wingspan.

9. The method according to claim 8, wherein, Normalized absorption morphology features were calculated based on the final asymmetric wingspan and valley depth. By inputting normalized absorption morphology features into a pre-trained inversion model, quantitative inversion results of environmental parameters of the target object are obtained, including: By combining the valley depth with the final asymmetric wingspan, a numerical value that can characterize the geometry of the absorption valley is constructed, and the normalized absorption morphology features are obtained. The pre-built and trained environmental parameter inversion model is invoked, and the model establishes a mapping relationship between spectral morphological features and environmental parameters through machine learning algorithms; The normalized absorption morphology characteristics are used as input data and fed into the environmental parameter inversion model for calculation. Obtain the output values ​​of the environmental parameter inversion model and determine these output values ​​as the quantitative inversion results of the environmental parameters of the target object.

10. A system for quantitative retrieval of environmental parameters based on spectral morphological features, which is suitable for the method for quantitative retrieval of environmental parameters based on spectral morphological features according to any one of claims 1-9, characterized in that, include: The depth extraction module is used to acquire the reflectance spectral data of the target object, identify absorption valleys in the reflectance spectral data and determine the lowest reflectance point and valley bottom reflectance value; perform convex envelope fitting on the corresponding band range of the absorption valley to obtain the envelope reflectance of the valley bottom band position as the reference reflectance outside the valley, and calculate the valley bottom depth. The sequence construction module is used to independently scan the reflectance data points of adjacent bands, starting from the lowest reflectance point at the bottom of the valley, and to calculate the proportion of the difference between each data point and the reflectance value at the bottom of the valley, so as to obtain the left difference proportion sequence and the right difference proportion sequence. The slope initial calculation module is used to continuously take a preset number of adjacent band points from the lowest reflectivity point at the bottom of the valley to the left and right as initial scan segments, respectively, and calculate the average slope within each initial scan segment to obtain the initial average slope on the left and the initial average slope on the right. The threshold generation module is used to generate dynamic thresholds for the left and right sides based on the normalized products of the valley depth and the reciprocal of the initial average slope on the left and the reciprocal of the initial average slope on the right, respectively. The initial boundary positioning module is used to find the location of the data point that first exceeds the left dynamic threshold in the left difference proportion sequence as the left wingspan boundary, and to find the location of the data point that first exceeds the right dynamic threshold in the right difference proportion sequence as the right wingspan boundary, thus obtaining the initial wingspan boundary; The iterative convergence module is used to recalculate the left and right average slopes based on the reflectivity data within the initial wingspan boundary, update the dynamic threshold, and repeat the boundary search steps until the boundary position converges and stabilizes, thus obtaining the final asymmetric wingspan. The model inversion module is used to calculate normalized absorption morphology features based on the final asymmetric wingspan and valley depth. By inputting the normalized absorption morphology features into the pre-trained inversion model, the quantitative inversion results of the environmental parameters of the target object are obtained.

Images