Method for monitoring withania frutescens based on unmanned aerial vehicle multispectral index rate of change

Abstract

The present invention relates to the field of vegetation index monitoring, and discloses a method for monitoring Solanum rostratum based on the change rate of drone multi-spectral indices, including obtaining multi-spectral image data of a target area in continuous time series; using an adaptive spectral calibration algorithm for the reconstructed time-series multi-spectral images to extract NDVI, GNDVI and red-edge vegetation index change layers, and obtaining an initial map of vegetation index change rates; aiming at the spectral confusion problem of Solanum rostratum at different growth stages and surrounding plants in terms of change rate performance; deeply fusing the initial map of vegetation index change rates with a growth fluctuation profile model to construct a change rate abnormal response factor layer and output candidate areas for suspected distribution of Solanum rostratum; combining the change rate evolution trend of the historical expansion path and the neighborhood growth consistency index, and dynamically generating a high-confidence identification map of Solanum rostratum. The present invention has the advantage of improving the discrimination accuracy of change rates.

Description

Technical Field

[0001] This invention relates to the field of vegetation index monitoring, specifically a method for monitoring Solanum nigrum based on the rate of change of multispectral indices from unmanned aerial vehicles (UAVs). Background Technology

[0002] Yellow-flowered Solanum is a highly resilient invasive plant, commonly found in grasslands, farmlands, and ecologically fragile areas of northern my country. Its rapid spread seriously threatens native vegetation and agricultural and pastoral ecological security. To achieve accurate identification and early warning of this harmful plant, a dynamic monitoring method based on vegetation index change rate has gradually emerged in recent years, utilizing drones equipped with multispectral cameras. By regularly collecting multi-temporal multispectral images and analyzing the changing trends of vegetation indices over time, it is helpful to identify the spatial expansion and growth characteristics of Yellow-flowered Solanum. However, in practical applications, especially in northern regions during the rapid growth period of Yellow-flowered Solanum from May to August each year, it coincides with frequent localized severe convective weather such as hail. Hail often causes large-scale mechanical damage to crops and weeds, leading to a sharp decline in vegetation indices in a short period of time, and even causing "false positive" signals in previously healthy vegetation that resemble the expansion areas of Yellow-flowered Solanum. This misjudgment caused by extreme weather interference seriously affects the monitoring accuracy based on vegetation index change rate.

[0003] Existing technologies often fail to effectively identify and eliminate areas of short-term vegetation damage caused by hailstorms during monitoring, and lack mechanisms for handling abnormal changes in vegetation indices under the influence of severe weather factors, leading to missed or false detections in the identification of Solanum nigrum. Therefore, a method for monitoring Solanum nigrum based on UAV multispectral vegetation index change rate is designed to improve the accuracy of change rate discrimination. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for monitoring the rate of change of the multispectral index of Solanum nigrum based on unmanned aerial vehicles (UAVs), which has the advantage of improving the accuracy of rate of change discrimination and solves the problems mentioned in the background technology.

[0005] To achieve the aforementioned goal of improving the accuracy of rate of change discrimination, this invention provides the following technical solution: a method for monitoring the rate of change of Solanum nigrum based on a UAV multispectral index, comprising the following steps:

[0006] Acquire multispectral image data of the target area in continuous time series, and reconstruct the spectral stability of the time series multispectral images based on the meteorological disturbance perception model;

[0007] An adaptive spectral calibration algorithm was used to extract NDVI, GNDVI and red-edge vegetation index change layers from the reconstructed temporal multispectral image. The vegetation change patches were then fused at multiple scales using a local spatiotemporal gradient enhancement module to obtain an initial map of vegetation index change rate.

[0008] To address the spectral confusion between Solanum nigrum and surrounding plants in terms of rate of change at different growth stages, a growth fluctuation profile model was constructed based on the morphological features of multidimensional change curves, and a dynamic feature matching algorithm based on edge fitting residuals was used for difference enhancement identification.

[0009] The initial map of vegetation index change rate was deeply fused with the growth fluctuation profile model to construct an anomaly response factor layer of change rate. Based on the vegetation index perturbation confidence propagation network, unstable areas were re-weighted and sampled to output candidate areas of suspected distribution of Solanum nigrum.

[0010] The candidate distribution areas of Solanum fulva were screened a second time based on the regional continuity discrimination mechanism. The evolution trend of the change rate of historical expansion path and the neighborhood growth consistency index were combined to dynamically generate a high-confidence identification map of Solanum fulva.

[0011] Preferably, the process of reconstructing the spectral stability of time-series multispectral images based on the meteorological disturbance sensing model is as follows:

[0012] Extreme weather events are constructed using data from external meteorological stations and illumination characteristics, reflectance changes, and cloud dynamics from multispectral images;

[0013] The spectral features of key frames affected by disturbances are reconstructed using time series interpolation and image inpainting techniques.

[0014] A spatiotemporal residual suppression mechanism is used to perform pixel-level correction on areas with severe local fluctuations, and consistency verification is performed on the reconstruction results.

[0015] Preferably, the process of extracting NDVI, GNDVI, and red-edge vegetation index change layers from the reconstructed temporal multispectral image using an adaptive spectral calibration algorithm is as follows:

[0016] Based on the dynamic changes in sensor characteristics and flight angle, spectral response correction is performed on each image frame using data from ground control points and reflectivity standard plates, and a local calibration model is constructed by combining stable background regions in the image.

[0017] Spectral similarity constraints are applied to pixels of the same ground features in a continuous image sequence. NDVI, GNDVI and red-edge exponential curves are dynamically fitted and smoothed to finally generate a layer with exponential variation.

[0018] Preferably, the process for obtaining the initial map of vegetation index change rate is as follows:

[0019] The NDVI, GNDVI and red-edge index layers at multiple time points are calculated pixel-by-pixel by differential calculation to construct a single-pixel temporal change rate curve;

[0020] A local spatiotemporal gradient enhancement module is introduced to analyze the spatial consistency and temporal similarity of the rate of change among neighboring pixels;

[0021] By combining the multi-scale image pyramid structure to fuse the change features at different patch scales, an initial map of vegetation index change rate is output.

[0022] Preferably, the process of constructing a growth fluctuation profile model based on the morphological characteristics of the multidimensional change curve is as follows:

[0023] Select the vegetation index time series of the target patch in the initial rate of change map and extract key dynamic features;

[0024] By jointly modeling the curves of NDVI, GNDVI, and red edge index, a multidimensional change profile vector describing the changes in growth rate, fluctuation frequency, and amplitude is constructed.

[0025] Clustering learning algorithms were used to classify and model the fluctuation characteristics of different plants, extract typical fluctuation templates of Solanum nigrum at each growth stage, and establish a standardized growth profile library.

[0026] Preferably, the difference enhancement recognition process based on the dynamic feature matching algorithm of edge fitting residuals is as follows:

[0027] By using the fitting error analysis method, the current time series curve is fitted with the standard fluctuation profile template by edge segment residuals to measure the difference in curve shape.

[0028] A dynamic window comparison mechanism is introduced to perform fine-grained similarity assessment on local perturbation points and identify potential plant species confusion areas;

[0029] By setting a multidimensional residual threshold, patch areas that are inconsistent with the growth trend of Solanum nigrum are filtered out, and the high-confidence fitted matching areas are output as the key target areas.

[0030] Preferably, the process of constructing the rate of change anomaly response factor layer is as follows:

[0031] The high-matching regions identified by difference enhancement are fused with the initial rate of change map, and the response intensity and salience of change in the whole map are statistically analyzed.

[0032] A pixel-level response factor layer is constructed based on the response intensity gradient and the change prominence coefficient to quantify the credibility of abnormal changes in each region;

[0033] By combining historical time series distribution and regional growth characteristics, a regional dynamic confidence boundary is set, and the expression form of the response factor is further adjusted to form a change rate anomaly response factor layer.

[0034] Preferably, the process for outputting candidate distribution regions of Solanum nigrum is as follows:

[0035] Based on the anomaly response factor layer of the rate of change, a region growing algorithm is performed on the regions in the layer whose response coefficients are higher than the threshold to expand the potential target boundary.

[0036] By combining a reweighted sampling mechanism, areas with abnormally high response are extracted in a focused manner, and the boundary morphology is corrected by combining a historical change trajectory model;

[0037] The confidence propagation network is used to re-score and classify regions containing unstable index responses, and output candidate regions for suspected distribution of Solanum nigrum.

[0038] Preferably, the process of dynamically generating a high-confidence recognition map of Solanum nigrum is as follows:

[0039] By comparing the suspected distribution candidate areas with the evolution path of Solanum nigrum in historical monitoring data, stable distribution core areas with high consistency were identified.

[0040] A regional continuity discrimination mechanism is introduced to apply consistency constraints to the distribution boundaries;

[0041] By combining the neighborhood growth consistency index, the synchronization of the index change trend within the target area is evaluated, and areas with matching fluctuation trends are screened.

[0042] By employing a multi-scale fusion strategy, the identification boundary is dynamically updated by integrating multi-temporal information, thereby generating a high-confidence identification map of Solanum nigrum.

[0043] Compared with existing technologies, this invention provides a method for monitoring Solanum nigrum based on the rate of change of multispectral index using a UAV, which has the following beneficial effects:

[0044] This invention provides a method for monitoring Solanum nigrum based on the multispectral index change rate of unmanned aerial vehicles (UAVs). This method integrates key technologies such as meteorological disturbance modeling, adaptive spectral calibration, multi-scale map construction, and dynamic difference identification in a multi-temporal, multispectral, and multi-dimensional feature space. It systematically solves technical challenges such as extreme weather interference, spectral confusion, and inaccurate extraction of vegetation change rates. By introducing a growth fluctuation profile model and a confidence propagation mechanism, it significantly improves the accuracy and stability of identifying Solanum nigrum at different growth stages and under complex ecological backgrounds. It effectively achieves intelligent identification and high-confidence dynamic updates of suspected distribution areas, enhancing the comprehensive performance of invasive species monitoring in terms of temporal adaptability, spatial accuracy, and intelligent processing capabilities. It possesses broad value for the application and promotion of ecological remote sensing. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the method of the present invention. Detailed Implementation

[0046] 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.

[0047] Example 1: Please refer to Figure 1 As shown in the embodiment of the present invention, a method for monitoring Solanum nigrum based on the multispectral index change rate of an unmanned aerial vehicle (UAV) includes the following steps:

[0048] S1: Acquire multispectral image data of the target area in continuous time series, and reconstruct the spectral stability of the time series multispectral images based on the meteorological disturbance perception model.

[0049] The process of reconstructing the spectral stability of time-series multispectral images based on the meteorological disturbance sensing model in S1 is as follows:

[0050] By utilizing external meteorological station data and illumination characteristics, reflectance changes, and cloud dynamics data from multispectral images, a perturbation classifier is constructed to identify extreme weather events such as rainfall, strong winds, hail, and heat waves.

[0051] Historical and real-time data from external meteorological stations covering the monitoring area are collected. Data types include, but are not limited to, meteorological parameters such as temperature, humidity, wind speed, wind direction, precipitation, and solar radiation intensity. Combined with the illumination characteristics, reflectivity fluctuations, and cloud movement paths and densities in the raw images acquired by UAVs equipped with multispectral sensors, a mapping relationship between image features and meteorological data is established. Based on supervised learning algorithms such as support vector machines or convolutional neural networks, a disturbance recognition model is trained to evaluate and classify the disturbance intensity of each frame of the image, identifying the affected areas and time windows of typical extreme weather events, including rainfall, strong winds, hail, and heat waves.

[0052] The spectral features of key frames affected by disturbances are reconstructed using time series interpolation and image inpainting techniques.

[0053] For image frames identified as being affected by disturbances, temporal interpolation algorithms such as linear regression interpolation, polynomial fitting interpolation, and spline interpolation are used to perform spectral compensation based on stable image features before and after the disturbance. For critical image frames with severe information loss, image restoration algorithms are combined to supplement multi-band spectral reflectance information of the damaged area by referencing adjacent pixels and intra-frame redundant information in the spatial domain. During the interpolation and restoration process, a regularization term is introduced to control spectral oscillations in frequently changing areas, ensuring that the interpolated image has good temporal continuity and physical rationality.

[0054] A spatiotemporal residual suppression mechanism is used to perform pixel-level correction on areas with severe local fluctuations, and consistency verification is performed on the reconstruction results.

[0055] Pixel-level difference analysis was performed on the interpolated and repaired images to calculate the residual value change curve of each pixel in the time series, identifying areas with local fluctuation anomalies. Combining the classification results of meteorological disturbances with spatial clustering algorithms, disturbance attribution analysis was performed on areas of severe fluctuations to identify non-biological interference areas that may be caused by sudden changes in illumination, cloud shadow interference, or local reflection anomalies. Residual smoothing processing was applied to the above areas, such as weighted moving average and principal component filtering, to suppress sudden fluctuations. Finally, the processed image sequence was statistically consistent with images from historical stable periods, including indicators such as inter-band correlation coefficient, mean drift, and standard deviation comparison, to ensure that the spectral reconstruction results meet the accuracy requirements for monitoring applications.

[0056] S2: The adaptive spectral calibration algorithm is used to extract the NDVI, GNDVI and red-edge vegetation index change layers from the reconstructed temporal multispectral image. The vegetation change patches are then fused at multiple scales using the local spatiotemporal gradient enhancement module to obtain the initial map of vegetation index change rate.

[0057] In S2, the process of extracting NDVI, GNDVI, and red-edge vegetation index change layers from the reconstructed temporal multispectral image using an adaptive spectral calibration algorithm is as follows:

[0058] Based on the dynamic changes in sensor characteristics and flight angle, spectral response correction is performed on each image frame using data from ground control points and reflectivity standard plates, and a local calibration model is constructed by combining stable background regions in the image.

[0059] For the multispectral sensors carried by UAVs, radiation response characteristic curves were established for different bands, considering the center wavelength, bandwidth, quantization accuracy, and field of view. Ground control points and standard reflectance white, gray, and blackboard images were collected during the flight mission to obtain standard reflectance values ​​under different lighting conditions, serving as absolute radiation calibration references. The influence of flight angle on incident angle and sensor response was analyzed, and an angle correction factor was introduced. Pixel-level normalization of the radiation response of each image frame was performed based on geometric attitude changes to improve the consistency of spectral response across different sorties. Spectrally stable background areas unaffected by seasonal interference, such as bare ground, roads, building roofs, or water bodies, were selected in the images, and their reflectance distribution changes across different image frames were statistically analyzed. Local calibration models were established using the spectral characteristics of these areas, and their time-series spectral curves were fitted through local regression analysis to generate calibration factor layers. For unstable regions, adaptive weight adjustments were made based on their relative relationship with stable regions in terms of band reflectance, achieving local correction between multispectral image frames.

[0060] Spectral similarity constraints are applied to the same ground object pixels in a continuous image sequence, and dynamic fitting and smoothing are performed on NDVI, GNDVI and red-edge exponential curves to finally generate a layer with exponential variation.

[0061] After spectral calibration, the reflectance values ​​of multiple time-series bands for the same ground feature pixel are constrained to ensure spectral consistency. This means that the fluctuations within a short period of time are limited to a set threshold, and obvious noise frames are removed to form a time-series vegetation index curve. Polynomial fitting filtering or Kalman filtering-based time-series smoothing methods are used on the above index curves to reduce curve jumps caused by noise or local disturbances. During the smoothing process, a typical index change rate model for the growing season is set to fit morphological guidance and enhance the biophysical interpretability of the curve. Finally, for each index type, a corresponding NDVI change layer, GNDVI change layer, and red-edge index change layer are generated as the core input for subsequent vegetation change patch detection and change rate extraction.

[0062] The process of obtaining the initial map of vegetation index change rate in S2 is as follows:

[0063] The NDVI, GNDVI and red-edge index layers at multiple time points are calculated pixel-by-pixel by differential calculation to construct a single-pixel temporal change rate curve;

[0064] At multiple consecutive time points, each moment corresponds to a set of spectrally calibrated NDVI, GNDVI, and red-edge vegetation index layers. For each pixel location, the vegetation index values ​​at two adjacent time points are compared, and the index change amplitude of the pixel between the two time points is estimated by the difference method. Using a time sliding window, the entire time series is traversed, and the changes in each period are calculated step by step, finally forming the NDVI change rate sequence, GNDVI change rate sequence, and red-edge vegetation index change rate sequence for each pixel in the entire observation period. After constructing the above three index change sequences, the change trend analysis can be performed separately by band, or they can be jointly constructed into a multi-dimensional change rate feature vector to comprehensively reflect the dynamic change characteristics of the pixel in different vegetation index dimensions, providing temporal change basic data support for subsequent spatial enhancement processing and target recognition.

[0065] A local spatiotemporal gradient enhancement module is introduced to analyze the spatial consistency and temporal similarity of the rate of change among neighboring pixels;

[0066] When analyzing the rate of change curve of each pixel, the rate of change information of its surrounding spatial neighborhood is incorporated. Statistical comparison is used to assess the consistency between the pixel and the average change trend of its neighborhood. If the change trend of a pixel is similar to that of other pixels in its neighborhood, it indicates high spatial consistency; otherwise, it may be affected by local disturbances or anomalous changes. Based on spatial consistency, temporal consistency characteristics are further analyzed. By comparing the current rate of change curve of a pixel with its typical change pattern in historical reference periods, it is determined whether the current change continues the previous growth trend. If the two are similar, it indicates strong temporal continuity and the change is relatively reliable; if the difference is significant, there may be staged anomalies or external interference. Finally, the consistency characteristics in both spatial and temporal dimensions are jointly modeled to construct a local spatiotemporal gradient response layer. This layer can effectively highlight areas that change drastically but exhibit consistency in both space and time, achieving an enhanced response to real vegetation changes while suppressing non-realistic variations caused by image noise, meteorological interference, etc., thus improving the accuracy and stability of subsequent change analysis.

[0067] By combining the multi-scale image pyramid structure to fuse the change features at different patch scales, an initial map of vegetation index change rate is output.

[0068] The rate of change layer is input into the image pyramid structure to construct an image pyramid from the original scale to the coarsened scale. Each layer obtains the rate of change layer at different spatial resolutions through downsampling.

[0069] In each scale layer, the scale response factor is used to evaluate the spatial connectivity, edge sharpness and rate of change of each patch, and to screen the effective patch regions.

[0070] A bottom-up feature fusion strategy is adopted to merge significantly changing regions at different scale levels layer by layer, thereby enhancing the comprehensive perception of continuous changes in large patches and sudden changes in small patches.

[0071] Finally, the initial map of vegetation index change rate after fusion is output. This map reflects the magnitude of index change at the pixel level and has spatial consistency and scale adaptability characteristics.

[0072] S3: To address the spectral confusion between Solanum nigrum and surrounding plants in terms of rate of change at different growth stages, a growth fluctuation profile model is constructed based on the morphological features of multidimensional change curves, and a dynamic feature matching algorithm based on edge fitting residuals is used for difference enhancement recognition.

[0073] The process of constructing the growth fluctuation profile model based on the morphological characteristics of the multidimensional change curve in S3 is as follows:

[0074] We selected the vegetation index time series of the target patches in the initial rate of change map and extracted key dynamic features such as slope abrupt change points, reversal nodes, and growth stagnation periods.

[0075] First, target patches with distinct dynamic characteristics were identified from the initial vegetation index change rate map. For these regions, NDVI, GNDVI, and red-edge index values ​​were extracted over continuous time series. By analyzing the trends in these series, a series of key nodes were identified, including rapid increases in the early stages of growth, slowdowns or plateaus in the later stages, reversals in periodic fluctuations, and potential sharp declines. These nodes reflect the rate changes and response characteristics of vegetation at different growth stages, forming the basis for morphological analysis.

[0076] By jointly modeling the curves of NDVI, GNDVI, and red edge index, a multidimensional change profile vector describing the changes in growth rate, fluctuation frequency, and amplitude is constructed.

[0077] This study jointly models the time-series curves of NDVI, GNDVI, and red-edged vegetation index, considering not only the changing trends of each individual index but also their synergistic changes. By analyzing the growth rate, frequency of change, and amplitude fluctuations of each index over different time periods, a comprehensive multidimensional change profile vector is constructed. This vector comprehensively reflects the growth rhythm and fluctuation patterns of vegetation within a specific time period, facilitating the differentiation between different plant species.

[0078] Clustering learning algorithms were used to classify and model the fluctuation characteristics of different plants, extract typical fluctuation templates of Solanum nigrum at each growth stage, and establish a standardized growth profile library.

[0079] After multidimensional feature extraction, an unsupervised clustering algorithm was used to classify and analyze the fluctuation curves of a large number of different plant samples. This process can group plants with similar growth behaviors into the same category, thereby uncovering the typical fluctuation characteristics of different plant species during their growth process. For Solanum nigrum, special attention was paid to its unique behaviors, such as high-frequency small-amplitude fluctuations, phased plateau stabilization periods, or sudden declines, exhibited during budding, rapid growth, expansion, and decline stages. Based on this, a standardized Solanum nigrum growth fluctuation profile template library was constructed. This library provides a basic reference for subsequent dynamic identification and differential enhancement.

[0080] The difference enhancement recognition process in S3 using the dynamic feature matching algorithm based on edge fitting residuals is as follows:

[0081] By using the fitting error analysis method, the current time series curve is fitted with the standard fluctuation profile template by edge segment residuals to measure the difference in curve shape.

[0082] The temporal variation curves of NDVI, GNDVI, and red edge index for each patch in the detection area are extracted, with a focus on the "edge segment" characteristics, namely the start and end stages of the curve throughout the growth cycle. These two stages typically contain key features of the species' growth initiation and decline. These edge segments are then fitted and compared with the corresponding standard template curves in the Solanum oxypetalum growth fluctuation profile template library, and the degree of deviation is measured by the residuals obtained after fitting. The key to this step is to use the fit of morphological trends as the criterion, rather than relying solely on numerical magnitude, thereby enabling a more accurate differentiation of dynamic differences between plant species.

[0083] A dynamic window comparison mechanism is introduced to perform fine-grained similarity assessment on local perturbation points and identify potential plant species confusion areas;

[0084] To address local fluctuations caused by environmental disturbances or observation errors, a dynamic window comparison mechanism is introduced. A sliding time window is set within a local region of the curve, comparing the microscopic morphological similarity between the template curve and the current curve within that window. This fine-grained sliding alignment analysis can identify pixels that, while exhibiting similar overall trends, show abnormal fluctuations in their local growth behavior. This is crucial for eliminating misidentifications caused by mixing of neighboring plants, overlapping of different vegetation types, or interference from remote sensing signals.

[0085] By setting a multidimensional residual threshold, patch areas that are inconsistent with the growth trend of Solanum nigrum are filtered out, and high-confidence fitted matching areas are output as key target areas.

[0086] Based on the fitting results of the first two steps, the residuals are decomposed into multiple dimensional indicators, such as maximum local difference, average overall deviation, and trend inflection point offset, to construct a multidimensional difference vector. Then, reasonable threshold rules are set to filter these indicators: patches whose residuals deviate significantly from the Solanum nigrum growth template are marked as "non-target areas" and removed; areas with high curve fitting, small residuals, and consistent fluctuation patterns are marked as "high-confidence target patches" and considered as possible candidate areas for Solanum nigrum distribution; the final output high-confidence matching areas are the key detection targets after difference enhancement identification, providing core input for subsequent response factor construction and region resampling.

[0087] S4: Deeply integrate the initial map of vegetation index change rate with the growth fluctuation profile model to construct a change rate anomaly response factor layer, and re-weight sample unstable areas based on the vegetation index perturbation confidence propagation network to output candidate areas of suspected distribution of Solanum nigrum.

[0088] The process of constructing the rate of change anomaly response factor layer in S4 is as follows:

[0089] The high-matching regions identified by difference enhancement are fused with the initial rate of change map, and their response intensity and salience of change in the whole map are statistically analyzed.

[0090] After completing the dynamic feature matching based on edge fitting residuals, a set of "high confidence matching regions" has been obtained, which may correspond to the growth activity patches of Solanum nigrum. Spatial overlap analysis is performed on these high confidence patches and the initially generated vegetation index change rate map. By statistically analyzing their change intensity, gradient mutation frequency and morphological persistence in each vegetation index change channel, the dynamic response features they exhibit in the whole map are extracted.

[0091] A pixel-level response factor layer is constructed based on the response intensity gradient and the change prominence coefficient to quantify the credibility of abnormal changes in each region;

[0092] Based on the above, the response factor value is calculated according to the spatial and temporal distribution of the response intensity of the patch to which each pixel belongs. This value comprehensively considers the following factors: rate of change amplitude: that is, the comprehensive intensity of the pixel's fluctuation in multiple exponential dimensions; degree of prominence of change gradient: measuring the abrupt contrast of the target area relative to its spatial neighborhood; frequency consistency index: judging whether the response continues to appear in multiple time periods to avoid misleading short-term anomalies; these indicators are uniformly encoded into a response factor value to represent the "anomaly credibility" exhibited by each pixel at the rate of change level.

[0093] By combining historical time series distribution and regional growth characteristics, a regional dynamic confidence boundary is set, and the expression form of the response factor is further adjusted to form a change rate anomaly response factor layer.

[0094] To improve the model's stability and generalization ability, historical time-series imagery and regional growth characteristics are incorporated for background correction. Specifically, this includes: referencing historical annual or seasonal distribution maps of *Solanum nigrum* growth to analyze whether similar response traces exist in the target area; integrating regional niche characteristics, such as soil type, landform location, and vegetation disturbance frequency, to ecologically constrain the spatial expression of response factors; constructing a dynamic confidence boundary based on regional habitat characteristics to re-normalize and suppress edge response values, avoiding misclassification of low-confidence areas as abnormal growth hotspots; and after the above multi-level fusion and correction, a "rate of change anomaly response factor layer" with continuous numerical expression is formed. This layer comprehensively reflects whether each pixel has strong, stable, and reliable dynamic change characteristics in the current time series, ultimately serving as the core input for subsequent reweighted sampling and extraction of suspected distribution areas.

[0095] The process of outputting the suspected distribution candidate region of Solanum nigrum in S4 is as follows:

[0096] Based on the anomaly response factor layer of the rate of change, a region growing algorithm is performed on the regions in the layer whose response coefficients are higher than the threshold to expand the potential target boundary.

[0097] Using the anomaly response factor layer constructed in the previous step, regions with response factor values ​​higher than a set threshold are selected. These regions exhibit significant dynamic changes and can be considered potential target regions. A region growing algorithm is then applied to the selected anomaly response regions. Starting from pixels with high response factors, the algorithm gradually expands towards neighboring regions based on the similarity of spatially adjacent pixels, forming a continuous boundary region that may include the distribution of *Solanum nigrum*. During the expansion process, the algorithm gradually adjusts the expansion range based on the intensity of the response factors and the consistency with the neighborhood, ensuring that expansion only reaches regions with highly similar rates of change and avoiding the accidental inclusion of irrelevant regions.

[0098] By combining a reweighted sampling mechanism, areas with abnormally high response are extracted in a focused manner, and their boundary morphology is corrected by combining historical change trajectory models;

[0099] For identified areas with abnormally high response, a reweighted sampling mechanism is introduced to facilitate focused extraction within these areas. Reweighted sampling: For regions with high response factors, a reweighting strategy based on local change patterns is employed to enhance their representativeness in spatiotemporal change trends. Through multiple sampling and weight adjustments, the mechanism identifies regions with more stable change patterns, further improving the reliability of candidate regions. Boundary correction: Combining historical change trajectory models, morphological correction is performed on the boundaries of these abnormally high-response regions. Historical models provide a reference to correct boundary errors caused by real-time data noise or other perturbations, ensuring that the morphology of the target region more closely matches the actual vegetation expansion trend.

[0100] The confidence propagation network is used to re-score and classify the regions containing unstable index responses, and output a layer of candidate regions for suspected distribution of Solanum cyrtonema.

[0101] Confidence Propagation Network: This network model dynamically analyzes the distribution of abnormal rates of change within a region, re-scoring each region using temporal variations and spatial structural features. By weighting the results of multiple propagations, the confidence level for unstable regions can be improved, further enhancing the accuracy of candidate region identification. By classifying and filtering the scored regions, areas with low scores and unstable changes are excluded, ensuring that the output candidate regions have a high probability of being distributed in *Solanum lyratum*. After these steps, a layer of suspected *Solanum lyratum* distribution candidate regions is finally output. This layer contains all *Solanum lyratum* growth patches verified to have a high probability of distribution, providing a foundation for subsequent high-confidence identification and verification.

[0102] S5: The candidate distribution areas of Solanum fulva are screened a second time based on the regional continuity discrimination mechanism. The evolution trend of the change rate of historical expansion path and the neighborhood growth consistency index are combined to dynamically generate a high-confidence identification map of Solanum fulva.

[0103] The process of dynamically generating the high-confidence recognition map of Solanum nigrum in S5 is as follows:

[0104] By comparing the suspected distribution candidate areas with the evolution path of Solanum nigrum in historical monitoring data, stable distribution core areas with high consistency were identified.

[0105] By analyzing the distribution trends of Solanum fulvidraco in historical monitoring data, we determined its evolutionary path at different time points, including expansion, contraction, and stabilization patterns. By calculating the consistency between the current candidate regions and historical evolutionary paths, we identified those regions that highly match the historical distribution patterns and determined these regions as "stable distribution core regions." These stable regions usually have relatively consistent spatiotemporal distribution characteristics and are areas where Solanum fulvidraco species have long-term stable existence, further verifying the credibility of the candidate regions.

[0106] A regional continuity discrimination mechanism is introduced to apply consistency constraints to the distribution boundary and eliminate discontinuous structures such as breaks, jumps, and isolated pixels.

[0107] The region continuity discrimination mechanism is used to determine whether a region is continuous by geometric connectivity and the spatial relationship of pixels within the region. For regions that are unnatural or do not conform to the plant growth pattern, such as those with broken or skipped distribution boundaries or isolated pixels, they are removed by region consistency test to ensure that the retained regions are coherent and reasonable.

[0108] By combining the neighborhood growth consistency index, the synchronization of the index change trend within the target area is evaluated, and areas with matching fluctuation trends are screened.

[0109] To further improve recognition accuracy, a neighborhood growth consistency index is introduced to ensure that the target area exhibits a consistent trend of change in both time and space. The neighborhood growth consistency index assesses the spatial and temporal stability of a region by calculating the consistency of its growth trends among multiple pixels within the target area and its neighborhood. For example, if the NDVI or GNDVI trend of a region shows high similarity to that of adjacent regions, it indicates that these regions have consistent growth trends. Combined with a spatiotemporal gradient enhancement method, the exponential change trends of each region are evaluated for synchronicity, ensuring that the fluctuation trends of the selected regions exhibit natural consistency in both time and space. Finally, regions with matching fluctuation trends are selected: through this evaluation process, regions with inconsistent trends with their surrounding areas or excessive fluctuations are eliminated, leaving stable target regions with matching fluctuation trends.

[0110] By employing a multi-scale fusion strategy, the identification boundary is dynamically updated by integrating multi-temporal information, thereby generating an identification map of Solanum nigrum.

[0111] By employing a multi-scale image pyramid structure and combining vegetation index changes at different scales, multi-level variation features are extracted. This enables the model to identify the distribution characteristics of *Solanum lyratum* at different resolutions. By combining data from multiple time points, the boundaries of the identified regions are dynamically updated and adjusted, which helps to eliminate errors caused by data from a single time point and enhances the accuracy of the *Solanum lyratum* distribution boundaries. By integrating spatiotemporal variation information and fusing layers from multiple time phases, the identification boundaries are dynamically adjusted and optimized, resulting in a final map with higher timeliness and stability. Finally, by combining the results from the above steps and using dynamically updated identification boundaries and the identification of high-confidence regions, a final high-confidence identification map of *Solanum lyratum* is generated.

[0112] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0113] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring Solanum nigrum based on the rate of change of multispectral index using a UAV, characterized in that, Includes the following steps: Acquire multispectral image data of the target area in continuous time series, and reconstruct the spectral stability of the time series multispectral images based on the meteorological disturbance perception model; The process of reconstructing the spectral stability of time-series multispectral images based on the meteorological disturbance sensing model is as follows: Extreme weather events are constructed using data from external meteorological stations and illumination characteristics, reflectance changes, and cloud dynamics from multispectral images; The spectral features of key frames affected by disturbances are reconstructed using time series interpolation and image inpainting techniques. A spatiotemporal residual suppression mechanism is used to perform pixel-level correction on areas with severe local fluctuations, and consistency verification is performed on the reconstruction results. An adaptive spectral calibration algorithm was used to extract NDVI, GNDVI and red-edge vegetation index change layers from the reconstructed temporal multispectral image. The vegetation change patches were then fused at multiple scales using a local spatiotemporal gradient enhancement module to obtain an initial map of vegetation index change rate. The process of obtaining the initial map of vegetation index change rate is as follows: The NDVI, GNDVI and red-edge index layers at multiple time points are calculated pixel-by-pixel by differential calculation to construct a single-pixel temporal change rate curve; A local spatiotemporal gradient enhancement module is introduced to analyze the spatial consistency and temporal similarity of the rate of change among neighboring pixels; By combining the multi-scale image pyramid structure to fuse the change features at different patch scales, an initial map of vegetation index change rate is output. To address the spectral confusion between Solanum nigrum and surrounding plants in terms of rate of change at different growth stages, a growth fluctuation profile model was constructed based on the morphological features of multidimensional change curves, and a dynamic feature matching algorithm based on edge fitting residuals was used for difference enhancement identification. The initial map of vegetation index change rate was deeply fused with the growth fluctuation profile model to construct an anomaly response factor layer of change rate. Based on the vegetation index perturbation confidence propagation network, unstable areas were re-weighted and sampled to output candidate areas of suspected distribution of Solanum nigrum. The candidate distribution areas of Solanum fulva were screened a second time based on the regional continuity discrimination mechanism. The evolution trend of the change rate of historical expansion path and the neighborhood growth consistency index were combined to dynamically generate a high-confidence identification map of Solanum fulva.

2. The method for monitoring Solanum nigrum based on the multispectral index change rate of an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The process of extracting NDVI, GNDVI, and red-edge vegetation index change layers from the reconstructed temporal multispectral image using an adaptive spectral calibration algorithm is as follows: Based on the dynamic changes in sensor characteristics and flight angle, spectral response correction is performed on each image frame using data from ground control points and reflectivity standard plates, and a local calibration model is constructed by combining stable background regions in the image. Spectral similarity constraints are applied to pixels of the same ground features in a continuous image sequence. NDVI, GNDVI and red-edge exponential curves are dynamically fitted and smoothed to finally generate a layer with exponential variation.

3. The method for monitoring Solanum nigrum based on the multispectral index change rate of an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The process of constructing a growth fluctuation profile model based on the morphological characteristics of multidimensional variation curves is as follows: Select the vegetation index time series of the target patch in the initial rate of change map and extract key dynamic features; By jointly modeling the curves of NDVI, GNDVI, and red edge index, a multidimensional change profile vector describing the changes in growth rate, fluctuation frequency, and amplitude is constructed. Clustering learning algorithms were used to classify and model the fluctuation characteristics of different plants, extract typical fluctuation templates of Solanum nigrum at each growth stage, and establish a standardized growth profile library.

4. The method for monitoring Solanum nigrum based on the multispectral index change rate of an unmanned aerial vehicle according to claim 3, characterized in that, The process of difference enhancement recognition based on dynamic feature matching algorithm of edge fitting residual is as follows: By using the fitting error analysis method, the current time series curve is fitted with the standard fluctuation profile template by edge segment residuals to measure the difference in curve shape. A dynamic window comparison mechanism is introduced to perform fine-grained similarity assessment on local perturbation points and identify potential plant species confusion areas; By setting a multidimensional residual threshold, patch areas that are inconsistent with the growth trend of Solanum nigrum are filtered out, and the high-confidence fitted matching areas are output as the key target areas.

5. The method for monitoring Solanum nigrum based on the multispectral index change rate of an unmanned aerial vehicle according to claim 4, characterized in that, The process of constructing the rate of change anomaly response factor layer is as follows: The high-matching regions identified by difference enhancement are fused with the initial rate of change map, and the response intensity and salience of change in the whole map are statistically analyzed. A pixel-level response factor layer is constructed based on the response intensity gradient and the change prominence coefficient to quantify the credibility of abnormal changes in each region; By combining historical time series distribution and regional growth characteristics, a regional dynamic confidence boundary is set, and the expression form of the response factor is further adjusted to form a change rate anomaly response factor layer.

6. The method for monitoring Solanum nigrum based on the multispectral index change rate of an unmanned aerial vehicle according to claim 5, characterized in that, The process of outputting the suspected distribution candidate regions of Solanum nigrum is as follows: Based on the anomaly response factor layer of the rate of change, a region growing algorithm is performed on the regions in the layer whose response coefficients are higher than the threshold to expand the potential target boundary. By combining a reweighted sampling mechanism, areas with abnormally high response are extracted in a focused manner, and the boundary morphology is corrected by combining a historical change trajectory model; The confidence propagation network is used to re-score and classify regions containing unstable index responses, and output candidate regions for suspected distribution of Solanum nigrum.

7. The method for monitoring Solanum nigrum based on the multispectral index change rate of an unmanned aerial vehicle according to claim 6, characterized in that, The process of dynamically generating a high-confidence recognition map of Solanum nigrum is as follows: By comparing the suspected distribution candidate areas with the evolution path of Solanum nigrum in historical monitoring data, stable distribution core areas with high consistency were identified. A regional continuity discrimination mechanism is introduced to apply consistency constraints to the distribution boundaries; By combining the neighborhood growth consistency index, the synchronization of the index change trend within the target area is evaluated, and areas with matching fluctuation trends are screened. By employing a multi-scale fusion strategy, the identification boundary is dynamically updated by integrating multi-temporal information, thereby generating a high-confidence identification map of Solanum nigrum.

Images