Ecological protection red line area change detection method based on artificial intelligence

Abstract

The invention discloses an ecological protection red line area change detection method based on artificial intelligence, and relates to the technical field of artificial intelligence, and the method comprises the steps: 1, converting remote sensing data at each time into directly comparable earth surface reflectivity; 2, splitting a mixed reflection spectrum in the remote sensing data into vegetation reflectivity spectrum data and soil reflectivity spectrum data through a pixel decomposition method; 3, establishing a deep learning model, automatically extracting change characteristics of the ecological protection area by taking the difference between the vegetation reflectivity spectral data and the soil reflectivity spectral data of the double-time-phase data as input, and outputting a change probability; 4, according to the sensitivity degree of the ecological protection red line area, setting a self-adaptive change detection threshold value; and determining a detection result according to the change probability and the adaptive change detection threshold. According to the method, the problems of large mixed pixel error, poor threshold fixity and the like in the traditional change detection are effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically to a method for detecting changes in the ecological protection red line area based on artificial intelligence. Background Art

[0002] Ecological protection red line areas usually have extremely high ecological function values and are the carriers of core ecological services such as water conservation, biodiversity protection, windbreak and sand fixation, and maintaining regional climate balance. Therefore, timely and accurate monitoring of their dynamic changes has become a key link in ensuring ecological security. Remote sensing technology, with its characteristics of large scale, periodicity, non-contact, and high spatial resolution, has been widely used in the dynamic monitoring of ecological protection red line areas. Currently, a large number of research and technical means rely on remote sensing images to identify ecological changes, such as the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Change Vector Analysis (CVA), and Image Differencing based on image differences. These traditional remote sensing analysis methods mainly identify surface changes through the spectral differences between remote sensing images at two time points. However, such methods still have significant problems in the monitoring of ecological changes in red line areas.

[0003] First of all, the existing technology has insufficient processing of mixed pixels, resulting in large errors in change identification. Red line areas are often distributed in complex terrain and landform areas such as mountains, hills, and wetlands. The pixels in remote sensing images are usually the mixed responses of various ground objects such as vegetation, bare soil, and water bodies. A single vegetation index or spectral difference cannot accurately distinguish weak ecological disturbances from natural background changes, and it is easy to produce "false changes" or "missed detections". Although some studies have introduced spectral mixture analysis (such as Linear Spectral Unmixing) to deconstruct pixels, these models are mostly limited to experimental verification under ideal conditions and lack the ability to real-time model the changing proportion of dynamic vegetation cover, making it difficult to adapt to the actual scenario of highly heterogeneous ecosystems.

[0004] Secondly, fixed thresholds or manual experience-based judgment criteria limit the accuracy and robustness of change recognition. Traditional change detection methods usually set a unified change threshold for spectral changes or vegetation index changes. Once the threshold is crossed, it is considered that a change has occurred. This method ignores the natural volatility, seasonal periodicity, and local outlier disturbances within the ecosystem. For example, in the forest edge zone of the red line area, non-structural changes such as leaf color transformation and light occlusion may occur with the change of seasons. These changes may show significant fluctuations in NDVI and are easily misidentified as ecological degradation. On the contrary, the slight disturbances in the initial stage of land reclamation may not be captured due to the small amplitude of coverage change. Current research has attempted to introduce dynamic thresholds or fuzzy logic rules to improve adaptability, but there are still limitations in terms of spatial scale, temporal consistency, and automated model deployment.

[0005] Thirdly, the lack of modeling of the spatial continuity of ecosystem changes leads to the lack of ecological interpretability of change recognition results. Ecosystem changes often have obvious spatial expansion characteristics. For example, forest degradation, water body shrinkage, and grassland fragmentation are not isolated events but show regional patch transfer or boundary movement. However, existing remote sensing change detection methods generally adopt pixel-level binary judgment, do not fully introduce spatial neighborhood information, and determine the change state only based on the spectral change of a single point, ignoring the overall consistency of the neighborhood structure. This is extremely likely to cause a speckled detection map or discontinuous boundary recognition, thus affecting the identification and management response of policymakers to ecological risk areas. Summary of the Invention

[0006] To solve the above technical problems, a change detection method for the ecological protection red line area based on artificial intelligence is provided, which can achieve high-precision intelligent recognition of small and critical ecological changes within the ecological protection red line area. This method effectively overcomes the problems existing in traditional change detection, such as large mixed pixel errors, poor threshold fixity, lack of spatial information, and lack of ecological interpretability of the model, and significantly improves the sensitivity, stability, and ecological interpretability of change detection.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A change detection method for the ecological protection red line area based on artificial intelligence, comprising: Step 1: Perform atmospheric correction and radiometric correction on the remote sensing data at each time to convert the remote sensing data at each time into directly comparable surface reflectance; Step 2: By means of pixel decomposition method, split the mixed reflection spectrum in the remote sensing data into vegetation reflectance spectrum data and soil reflectance spectrum data; both the vegetation reflectance spectrum data and the soil reflectance spectrum data are dual-temporal data; Step 3: Establish a deep learning model, using the difference between the vegetation reflectance spectral data and the soil reflectance spectral data of the dual-temporal data as the input, extract the change characteristics of the ecological protection area, and output the change probability; Step 4: Set an adaptive change detection threshold according to the sensitivity of the ecological protection red line area; determine the detection result according to the change probability and the adaptive change detection threshold.

[0008] Further, in step 1, the surface reflectance at time is: Where: represents the surface reflectance of pixel at time ; represents the radiance received by pixel at time ; represents the distance between the earth and the sun; represents the solar irradiance at time ; represents the solar zenith angle at time ; is the abscissa of the pixel; is the ordinate of the pixel.

[0009] Further, the radiance is obtained through the following process: Install a radiation detection instrument on the remote sensing sensor; the radiation detection instrument includes a multispectral sensor and an imaging spectrometer; when the remote sensing sensor passes over the ground, the sunlight reflected by the ground travels through the atmosphere and reaches the photosensitive element of the remote sensing sensor, and the remote sensing sensor records the intensity of the electromagnetic radiation energy received in different bands and stores it in the form of digital gray values; convert the digital gray values recorded by the remote sensing sensor into radiance with physical meaning through radiometric calibration processing.

[0010] Further, the process of converting the digital gray values recorded by the remote sensing sensor into radiance with physical meaning includes: Obtain the digital gray value calibration parameters, including gain and offset, and convert them through the radiance formula to obtain the radiance value of each pixel in the remote sensing data at a specific band. The radiance formula is: Radiance = Gain * Digital gray value + Offset, and the unit of radiance is W·m⁻²·sr⁻¹·μm⁻¹.

[0011] Further, in step 2, through the pixel decomposition method, the mixed reflectance spectrum in the remote sensing data is split into vegetation reflectance spectral data and soil reflectance spectral data as: Where: is time the vegetation coverage area ratio of the lower pixel ; is time the vegetation reflectance spectral data of the lower pixel ; is time the soil reflectance spectral data of the lower pixel ;

[0012] Furthermore, in step 3, the output change probability is: where: represents the change probability of the lower pixel at time ; represents the sigmoid activation function; represents the number of layers of the deep learning network, with a value greater than or equal to 6; represents the weight of the th layer; represents the non-linear activation function of the th layer; and represent two different times; is time the surface reflectance of the lower pixel at time is time the surface reflectance of the lower pixel at time is time the vegetation coverage area ratio of the lower pixel at time is time the vegetation coverage area ratio of the lower pixel at time is time the vegetation reflectance spectral data of the lower pixel at time is time the soil reflectance spectral data of the lower pixel at time is time the soil reflectance spectral data of the lower pixel at time

[0013] Furthermore, ; when is greater than or equal to 1 and less than 3, is the rectified linear unit function; when is greater than or equal to 3 and less than 5, is the exponential linear unit function; when Greater than or equal to 5 and less than When is the hyperbolic tangent function.

[0014] Furthermore, in step 4, set the adaptive change detection threshold as: Where: is the detection reference value, and its value range is a real number greater than 0 and less than 1. The larger it is, the greater the sensitivity of the ecological protection red line area.

[0015] Furthermore, in step 4, according to the change probability and the adaptive change detection threshold, the detection result is determined as: Where: represents the detection result of the pixel at time , 1 indicates change, and 0 indicates no change; represents the spatial neighborhood of the pixel ; represents the indicator function, which is 1 when the condition is satisfied and 0 otherwise; represents the number of pixels in the neighborhood; represents the neighborhood consistency threshold, and its value range is 0 - 1; and respectively represent the abscissa and ordinate of other pixels in the spatial neighborhood of the pixel ; represents the time at which the pixel changes; represents the time at which the pixel has the adaptive change detection threshold.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: It realizes high-precision identification and dynamic monitoring of ecological disturbances in the ecological protection red line area. Compared with the prior art that only relies on methods such as vegetation index changes or fixed thresholds, the present invention has significant beneficial effects: First, through physical radiance inversion and standardized reflectance calculation, the consistency and comparability of multi-temporal remote sensing data are ensured, laying a solid foundation for subsequent modeling; Second, by introducing a vegetation-soil decomposition model, the ecological information in the mixed pixels is effectively stripped, enhancing the sensitivity of the model to weak changes; Third, the deep neural network constructed by combining multi-layer activation functions and band-sensitive weights can automatically extract change features and generate a change probability output with physical interpretation; Finally, through the exponential adaptive threshold and spatial neighborhood consistency determination mechanism, the stability and ecological interpretability of the detection results are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic flowchart of the method for detecting changes in the ecological protection red line area based on artificial intelligence proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variants.

[0019] Referring to Figure 1 As shown, the method for detecting changes in the ecological protection red line area based on artificial intelligence in the embodiments of the present invention includes: Step 1: Perform atmospheric correction and radiometric correction on the remote sensing data at each time to convert the remote sensing data at each time into directly comparable surface reflectance. Step 1 is the basic link of the entire method system. Its core goal is to convert the remote sensing observation data obtained at different times into surface reflectance that can be compared across time, so as to provide input data with high consistency and high physical authenticity for subsequent change feature extraction and deep learning model analysis. In the original acquisition process of remote sensing images, due to the complex changes of observation conditions, solar irradiance, atmospheric composition, and sensor response, the reflection information of the same ground object shows significant differences at different times and under different climate conditions. Therefore, through precise atmospheric correction and radiometric calibration processes, the influence of these non-ground object attributes must be removed to restore its true surface radiation characteristics, so that the artificial intelligence model can scientifically and reliably learn and determine the changes in the ecological protection red line area.

[0020] In the present invention, the acquisition of surface reflectance starts from the original images recorded by remote sensing sensors. These images are presented in the form of digital gray values, and the gray value of each pixel actually corresponds to the radiation energy received by the pixel in a certain wavelength band. However, since the gray value itself does not have a direct physical meaning, it is necessary to first convert it into radiance with actual energy meaning through a radiometric calibration process. This process depends on the calibration parameters of the remote sensing sensor, including gain and bias, and realizes the conversion between gray value and radiance through linear mapping. This physical modeling process ensures the physical authenticity of the data and provides a reliable basis for subsequent reflectance calculation.

[0021] Furthermore, considering that the surface radiation signal is affected by multiple scattering and absorption of atmospheric gases, aerosols, water vapor and other components during the transmission process, the present invention introduces an atmospheric correction mechanism and comprehensively corrects it using key parameters such as solar zenith angle, Earth-Sun distance, and solar irradiance in the observation wavelength band. Through the derived reflectance expression, a quantitative relationship is established between the original radiance and the observation geometric conditions and solar energy distribution parameters, so as to obtain the standardized surface reflectance. This reflectance has been stripped of the interference caused by the observation time and climate conditions, so that remote sensing data from different time phases can be regarded as physical data under the same evaluation benchmark by the artificial intelligence model, making change detection possible.

[0022] It should be emphasized that the surface reflectance in the present invention is not only an intermediate product of data preprocessing, but also plays a key role as an information carrier in the whole ecological change detection. During the long-term monitoring of the ecological protection red line area, due to the interweaving of various factors such as vegetation growth cycle, climate evolution and human intervention, the surface reflection characteristics in the area will change subtly and complexly. Without accurate atmospheric correction and radiation standardization steps, these changes will be masked by observation errors, and it is difficult for the artificial intelligence model to learn effective change rules from them. Through the reflectance conversion method proposed by the present invention, not only the true spectral characteristics of ground objects are retained, but also the interference of time noise on model training is effectively reduced, and the sensitivity and robustness of change detection are improved.

[0023] Step 2: Through the pixel decomposition method, split the mixed reflection spectrum in the remote sensing data into vegetation reflectance spectrum data and soil reflectance spectrum data; both the vegetation reflectance spectrum data and the soil reflectance spectrum data are dual-temporal data; The core of step 2 lies in the inversion and decomposition of the mixed spectrum of each pixel in the remote sensing image, so as to transform the observed data into two components: vegetation reflectance and soil reflectance with ecological and physical indication significance. This decomposition is not only a key link in data preprocessing, but also a structural basis for supporting the subsequent deep learning model to accurately identify changes in the ecological protection red line area. Due to the limitations of the spatial resolution of remote sensing imaging, the ground objects corresponding to a pixel are often a mixture of multiple ground object types. For example, some bare soil areas covered with vegetation, desert areas scattered with shrubs, or buffer zones where weeds and buildings coexist. In this case, the original surface reflectance value cannot reflect the true changes of a single ecological factor. Therefore, it is necessary to use the mixed pixel decomposition technology to reconstruct the remote sensing observation value into a structural expression composed of weighted combinations of basic components.

[0024] On this basis, the present invention adopts a linearly mixed model with physical interpretability to decompose the overall reflectance of each pixel at any time into a linear superposition result jointly composed of a certain proportion of vegetation reflectance and soil reflectance. This model assumes that at the medium-scale resolution, the spectral responses of vegetation and soil as the dominant ecological factors can respectively represent the two basic components of green coverage and non-green background in the pixel. By introducing the vegetation coverage ratio as a weighting factor, each reflectance value not only carries spectral information but also integrates the structural information of the ground object distribution. The introduction of this method not only improves the discrimination of internal changes in ecological factors but also establishes a mapping relationship between the ecological structure and the spectral response at the data level, enabling the subsequent model to rely on more structurally discriminative feature data when learning and judging changes in the ecological red line area.

[0025] The essence of this step is a decoupling process that refines the macroscopic remote sensing signal into microscopic ecological indicators, which not only solves the inherent problems of "fuzziness" and "mixture" of remote sensing pixels but also provides a sparser and clearer input channel for the deep learning network. Especially in the red line area, this kind of area is often the spatial zone where the ecosystem changes are the most sensitive and variable. Its vegetation coverage and soil exposure degree fluctuate rapidly with seasonal, climate disturbances or human development. Therefore, without an accurate pixel decomposition process, it is very easy to generate "false changes" or "false alarms" in the model, thereby weakening the response ability of the intelligent system to real changes. Through the pixel spectral deconstruction mechanism adopted in the present invention, the model can focus on two clear change indicators during the learning process - the double-temporal difference between vegetation reflectance and soil reflectance - thus effectively avoiding the interference of complex ground object backgrounds on the classification boundary and improving the credibility of dynamic monitoring in the ecological red line area.

[0026] In addition, the decomposition results output by this step also have a high degree of physical consistency, making it easier for the subsequent neural network to converge during the training phase and better fitting the non-linear evolution process of the ecosystem. Since the deconstruction model adopted in the present invention takes into account both the coverage ratio of the spatial structure and the intensity characteristics of the spectral response in the modeling, its output not only serves the change detection itself, but also provides a data basis for the temporal evolution analysis of the ecological state, the evaluation of vegetation degradation, and the derivation of the soil drought index. This characteristic reflects the generality and expansibility of this method in the intelligent analysis of ecological remote sensing. To sum up, step 2 is not only a technical processing process, but also a key bridge for converting the original data into structural ecological variables and constructing a more expressive feature space for the artificial intelligence model.

[0027] Step 3: Establish a deep learning model, using the difference between the vegetation reflectance spectral data and the soil reflectance spectral data of the two-temporal-phase data as the input, extract the change characteristics of the ecological protection area, and output the change probability; Step 3 is the core of the entire technical system. Its fundamental task is to automatically extract the change characteristics of the ecological protection red line area and output the change probability at the pixel level based on the difference between the two-temporal-phase vegetation and soil reflectance spectral data obtained in the previous two steps by constructing a deep learning model. Compared with the traditional change detection method that only relies on the static threshold determination of single-band difference or index combination, the present invention fully introduces the temporal perception and non-linear feature learning ability of artificial intelligence, enabling the system to identify weak and critical ecological dynamics in complex surface change scenarios. Especially in the highly sensitive area of the ecological protection red line, its ability to identify small-scale disturbances and high-frequency perturbations is significantly better than the traditional method, reflecting the breakthrough value of intelligent monitoring.

[0028] The design of the deep learning model in the present invention is closely constructed around the physical essence of ecological changes. The input end collects the spectral differences between the vegetation and soil after pixel decomposition between the two temporal phases, and these differences not only contain the change information of the reflectance characteristics of the ground objects, but also imply the evolution trend of the surface coverage structure. Through multi-layer non-linear transformation, the model maps such original spectral differences into a feature space with higher discrimination, making the ecological disturbances present a more stable and easily recognizable feature form in this space. Especially in the deep structure, the model introduces a combination of multiple types of activation functions, such as rectified linear unit, exponential linear unit, and hyperbolic tangent function, etc., to simulate the response modes of ecological system changes at different scales and intensities, thereby improving the recognition ability of weak changes and composite changes. Correspondingly, the non-uniformity of the weight design of each layer is adopted, and the dominant weight of the high-level abstract features in the final determination is gradually enhanced through the hierarchical attenuation mechanism, making the model pay more attention to the deep structural differences hidden in the temporal evolution, rather than just the drastic fluctuations of the surface spectrum.

[0029] In the present invention, the output of the change probability is not just a linear function of a certain significant difference, but a non-linear output after the entire deep neural network comprehensively judges the input information. This way endows the model with the fault tolerance ability to noise perturbations and the enhanced sensitivity to local dynamics. For example, although there is little change in the surface reflectance in some areas, the vegetation coverage ratio fluctuates slightly and the soil moisture state is slightly adjusted. The evolution of this ecological state is a typical change with high sensitivity, which is easily overlooked by traditional methods. The deep model constructed in the present invention precisely expresses it as a high-probability output through multi-layer feature combination and weighting mechanism, reflecting the high sensitivity of the intelligent system to the changes in the ecological security boundary.

[0030] Furthermore, the model not only captures local features, but also introduces the ability of spatial similarity modeling through the convolution kernel weight and activation function adjustment mechanism, that is, the similarity or difference in the reflectance structure between pixels will also affect the change probability of the current pixel. This mechanism effectively alleviates the "island effect" and "boundary blur" in remote sensing data and improves the coherence and spatial consistency of the detection results. At the same time, the change probability output by the model is not a binary result, but a change trend index in a continuous value range, providing a mathematical basis and semantic interpretation for subsequent adaptive threshold judgment and spatial neighborhood constraint.

[0031] Step 4: Set an adaptive change detection threshold according to the sensitivity of the ecological protection red line area; determine the detection result according to the change probability and the adaptive change detection threshold.

[0032] Step 4 undertakes the key responsibility of making the final change recognition decision from the model output result. Its core lies in designing an adaptive change detection threshold mechanism highly adapted to ecological sensitivity, and combining the spatial neighborhood consistency strategy to accurately judge the change probability of each pixel. This process is not only an effective interpretation and physical constraint on the output of the deep learning model, but also an important guarantee for achieving high accuracy and high stability recognition in the ecological red line protection system. Since the ecological protection red line area often has strong spatial heterogeneity and temporal sensitivity, the fixed threshold method is often difficult to adapt to the complex evolution rhythm of different ecosystems. The present invention realizes the intelligent adjustment ability that adapts to local conditions in spatial distribution and adapts to the times in temporal changes by introducing an exponential threshold function dynamically coupled with the vegetation spectral characteristics, enabling the system to have stronger ecological adaptability and intelligent judgment ability.

[0033] In the specific implementation process, the present invention compares the change probability output by the model with an adaptive threshold to determine whether an ecological change has occurred in a certain pixel. This threshold is not a fixed value set artificially, but is dynamically generated according to the change amplitude of the vegetation reflectance between two temporal phases of the pixel. When the vegetation reflectance in a certain area changes drastically, it may mean a real ecological disturbance, and the system will automatically lower the judgment threshold and enhance the sensitivity; on the contrary, when the change is weak, the threshold will be increased accordingly to avoid the interference of accidental noise on the change determination. This exponential adjustment mechanism enables the system to have an amplification mechanism for weak changes at the red line boundary, while showing a fault tolerance characteristic for slight fluctuations in non-critical areas, effectively avoiding the dual risks of false alarms and missed judgments. Especially in some areas with complex landforms, frequent ecological fluctuations but not caused by human intervention, this mechanism can fully release the change perception ability of the artificial intelligence model while maintaining the physical rationality and ecological interpretability of the overall system judgment.

[0034] More importantly, the present invention does not limit the change determination to the single-pixel dimension, but further introduces a spatial neighborhood consistency judgment strategy, and uses the comparison results of the change probabilities and thresholds of the pixels in the neighborhood for integrated judgment. This strategy stems from the continuous feature of ecological changes in space, that is, real ecological disturbances often do not be limited to a single isolated pixel, but occur and expand at the regional scale in the form of groups or patches. By constructing a local neighborhood system centered on the current pixel and calculating and determining the proportion of changed pixels in the neighborhood, the present invention effectively eliminates the "noise change" phenomenon caused by isolated pixel errors, and also improves the spatial coherence and ecological structure interpretability of the change results. The introduction of the neighborhood consistency threshold not only enhances the stability of the determination, but also endows the system with a certain degree of "group judgment ability", which is difficult to compare with the traditional change detection method based on single-pixel logical judgment.

[0035] In addition, the change determination logic itself adopts a strict logical condition combination in mathematical form, making the final output result have replicability and high controllability. The change determination not only requires that the change probability of the current pixel is higher than the adaptive threshold, but also requires that the proportion of pixels in the neighborhood that meet the same conditions exceeds a certain set threshold. This "probability + structure" type determination mechanism reflects the full mobilization of the present invention's multi-scale structure perception ability of the ecological system driven by artificial intelligence, ensuring that the change determination is not only accurate, but also has ecological coherence. This spatially collaborative determination method is especially suitable for special ecological units such as the ecological protection red line area with clear boundaries, clear functions, and significant change consequences, enabling the system to maintain stability and discrimination ability in dynamic changes.

[0036] Furthermore, in step 1, the surface reflectance of the ground surface at time is: Wherein: represents time the surface reflectance of the lower pixel ; represents time the radiance of the pixel received at ; represents the distance between the Earth and the Sun; represents time the solar irradiance at represents time the solar zenith angle at is the abscissa of the pixel; is the ordinate of the pixel.

[0037] The adopted reflectance calculation model is a physical derivation model with radiance as the core input variable, and its goal is to convert the amount of electromagnetic radiation received by the remote sensor in the sky (i.e., radiance ) into the true surface reflectance of the corresponding ground object at the observation moment . This conversion process is based on the accurate modeling of solar irradiance, solar zenith angle, and the Earth-Sun distance. Specifically, the radiant energy reflected by the ground object after being irradiated by the sun first propagates through the atmosphere and is recorded by the detector at the lens of the remote sensor. This process is affected by the coupling of various geometric and radiation parameters. To eliminate these external interference factors and restore the reflection ability of the ground object itself in a specific band, an inversion mechanism needs to be introduced to normalize the received brightness value with the solar irradiation intensity and irradiation angle, etc.

[0038] Among them, is a factor introduced due to the common use of the Lambertian reflection model in the remote sensing system to average the angular distribution of radiant energy, ensuring the conservation of energy in the calculation process. The square term of the Earth-Sun distance is used to correct the change in energy density caused by the length of the irradiation path between the sun and the Earth. Considering that the orbit of the Earth around the sun is elliptical and its distance has periodic changes, if not corrected, it will cause errors in the solar energy input in winter and summer seasons, thus affecting the authenticity of the seasonal change characteristics of vegetation; this is particularly crucial in the ecological red line area because the small periodic characteristics of vegetation changes are often the early signals of ecological disturbances.

[0039] And , that is, the solar irradiance at time , is the solar energy input value per unit area defined in the band, and it jointly determines the final remotely sensed received brightness with the reflection characteristics of the ground object. By normalizing the radiance to the unit incident energy, the inconsistency of the solar input energy is eliminated, enhancing the comparability between time series. As for This item is used to correct the projection effect caused by the solar zenith angle. The larger the solar zenith angle, the closer the sun is to the horizon, the longer its irradiation path in the atmosphere, and the smaller the energy per unit area actually projected onto the ground; when the zenith angle approaches zero, that is, the sun is close to directly above, the irradiation intensity is the largest. Dividing the brightness by this cosine value is actually restoring the reflected amount of the obliquely illuminated ground surface to the standard reflectance under direct illumination conditions. This design enables the data obtained at different times and different observation angles to be unified in physical meaning, ensuring that the artificial intelligence system can process highly consistent data during the training and prediction stages.

[0040] Combined with the objective of the present invention, that is, to automatically and accurately identify the changes in the ecological protection red line area, the above reflectance calculation model has extremely high practical adaptability and engineering value. On the one hand, it provides the physical property values of ground objects that can be directly compared across time, making it possible for dual-temporal difference analysis, which is a prerequisite for subsequent pixel decomposition and change feature extraction. On the other hand, this calculation process fully suppresses the non-ground object interference factors in ecological changes and improves the sensitivity of the model to real ecological disturbances. For example, on the same plot, if there are significant differences in the solar zenith angle due to different imaging times of remote sensing images, without geometric normalization processing, the reflectance differences in the same vegetation area may be completely caused by the observation conditions, resulting in the model misidentifying it as an ecological change; by introducing and and other correction factors, this error can be eliminated, thereby improving the authenticity of change detection and the ecological interpretability of the determination results.

[0041] Furthermore, the radiance is obtained through the following process: Install a radiation detection instrument on the remote sensing sensor; the radiation detection instrument includes a multispectral sensor and an imaging spectrometer; when the remote sensing sensor passes over the ground, the sunlight reflected by the ground surface travels through the atmosphere and reaches the photosensitive element of the remote sensing sensor. The remote sensing sensor records the intensity of the electromagnetic radiation energy received in different bands and stores it in the form of digital gray values; the digital gray values recorded by the remote sensing sensor are converted into radiance with physical meaning through radiometric calibration processing.

[0042] In actual operation, the present invention realizes the sub-band observation of the surface reflected light at multiple specific bands by carrying a high-precision radiation detection device on a remote sensing platform, including a multispectral sensor and an imaging spectrometer. These detectors can separate the reflected solar radiation signals from the surface into electromagnetic energy flows of different bands and transmit them to the photosensitive element in the form of electrical signals for recording. During the process of the remote sensing platform passing over the surface, the sunlight reflected by the ground objects does not propagate directly to the sensor, but needs to pass through the atmosphere, a complex radiation transmission medium. The aerosols, water vapor, ozone, and various gas components in the atmosphere will have multiple effects on the radiation of different bands, such as absorption, scattering, and reflection, resulting in changes in the intensity and spectrum of the electromagnetic signals finally reaching the sensor. Therefore, the signals received by the remote sensing photosensitive device are not exactly the same as the original reflection characteristics of the ground objects, but a composite signal modulated by the atmospheric path. To enable the artificial intelligence model to learn the true characteristics of ecological state changes, it is necessary to connect the original remote sensing data with the true reflection behavior of the surface through the physical quantity of radiance.

[0043] The remote sensing images recorded by the sensor are presented in the form of pixel gray values. These gray values are the digital expressions after the analog-to-digital conversion of electronic signals, and there is no direct corresponding relationship between their numerical ranges and the actual radiation energy. Therefore, the present invention introduces a radiometric calibration mechanism to convert the gray value of each band into a radiance value with physical significance. This conversion process depends on the previously obtained calibration parameters, usually including the gain and offset of each band, which together define the linear mapping relationship between the gray value and the true brightness. Specifically, the system determines the response curve of each band under different gray inputs through calibration experiments, and then constructs a gray-brightness conversion function, so that the finally obtained radiance has the unit of W·m⁻²·sr⁻¹·μm⁻¹, with strict physical dimensions and comparability. Through this conversion process, the present invention ensures the true comparability of the brightness values of each pixel in the remote sensing data, providing stable input for the accurate identification of the spectral characteristics of ground objects and the spatial consistency analysis of ecological indicators.

[0044] More critically, the process of obtaining this radiance not only serves a single link in reflectance calculation but also serves as the basic data structure running through the entire change detection system. Its consistency and stability at different time nodes directly determine the credibility of the differences between the two-temporal data. In the dynamic monitoring of the ecological protection red line area, it is often necessary to accurately perceive extremely weak phenomena such as vegetation degradation, increased soil exposure, and the spread of local disturbances. These phenomena often do not cause obvious changes in the macroscopic structure of the image but can only be captured through minute fluctuations in pixel brightness values. Therefore, only by precisely converting the original grayscale values into radiance and correcting them through subsequent reflectance models can input data for identifying such "hidden changes" be provided to the deep learning model, enabling the model to have the discrimination ability when facing subtle but crucial disturbances in the ecosystem.

[0045] Furthermore, the process of converting the digital grayscale values recorded by the remote sensing sensor into radiance with physical meaning through radiometric calibration processing includes: obtaining the calibration parameters for the digital grayscale values, including gain and bias, and converting them through the radiance formula to obtain the radiance value of each pixel in the remote sensing data at a specific band. The radiance formula is: Radiance = Gain * Digital Grayscale Value + Bias, and the unit of radiance is W·m⁻²·sr⁻¹·μm⁻¹.

[0046] In the "Method for Detecting Changes in the Ecological Protection Red Line Area Based on Artificial Intelligence" of the present invention, to ensure the physical accuracy and ecological interpretability of remote sensing data in ecological change identification, a crucial step is to convert the digital grayscale values recorded in the remote sensing image into radiance values with physical meaning. This conversion process is not only a data format processing but also a scientific modeling of mapping the electrical signals obtained from the sensing device to the radiation attributes of ground objects. Its core principle stems from the theory of remote sensing radiometric calibration and is a crucial link in the remote sensing physical modeling system. Especially for an environmentally sensitive area such as the ecological protection red line area, any minor vegetation disturbance or surface exposure change may affect the stability of the overall ecosystem. Only a change detection mechanism based on real radiation energy values can achieve high-precision identification and high-credibility judgment in the artificial intelligence system.

[0047] During the process of earth observation by remote sensing sensors, each pixel recorded actually corresponds to the intensity of the electromagnetic radiation received by the sensor within a specific wavelength band. However, the sensor itself does not directly record the energy value. Instead, it converts the electromagnetic signal into voltage, and then converts it into digital gray values for storage by an analog-to-digital converter. Although these gray values are represented as integers in the range of 0 to 255 (or higher bits, such as 10-bit, 12-bit, or even 16-bit gray levels) on the image, they have no direct physical meaning and cannot reflect the true level of the energy reflected by the ground object. Therefore, before any remote sensing analysis with ecological and physical significance is carried out, the gray values must be converted into actual radiance first. Based on this core requirement, the present invention constructs a systematic and accurate radiometric calibration processing module.

[0048] The core of this conversion process lies in obtaining the response characteristics of the remote sensing device for each wavelength band, that is, obtaining two calibration parameters, gain and offset, through experiments or data provided by the manufacturer. Gain represents the amount of energy increase corresponding to a unit gray value and is a proportionality coefficient; offset represents the system background output at zero gray level, which is mainly used to compensate for the influence of sensor non-linear response and electronic noise. Therefore, for the gray value of each pixel , in a specific wavelength band , its radiance can be converted through the following linear relationship: ; where represents the gain of wavelength band , and represents the offset. The physical meaning of this formula is that it converts the remote sensing signal from "relative expression" to "absolute energy", enabling the pixel brightness value to have a strict energy measurement unit - watts per square meter per steradian per micrometer (W·m⁻²·sr⁻¹·μm⁻¹), thus providing solid data support for subsequent calculation of surface reflectivity.

[0049] Furthermore, in step 2, through the pixel decomposition method, the mixed reflectance spectrum in the remote sensing data is split into vegetation reflectance spectrum data and soil reflectance spectrum data as follows: where: is the proportion of the vegetation-covered area of pixel at time ; is the vegetation reflectance spectrum data of pixel at time ; is the soil reflectance spectrum data of pixel at time .

[0050] In the specific implementation process, this decomposition method is essentially based on an assumption: the observed spectral response can be regarded as a weighted superposition of two main land cover types, namely vegetation and soil, at the pixel scale. Different land covers have relatively stable characteristic peaks and reflectance intervals in the spectral space. Therefore, as long as an appropriate coverage factor can be found, the mixed spectrum observed once can be decomposed into a linear combination of several components. The most crucial concept here is the vegetation coverage area ratio, which indicates the distribution degree of green plants covered by a certain pixel at the observation moment, thereby quantitatively distinguishing the spectral mixture in a physical sense. When this ratio is higher, it means that the vegetation reflectance of the pixel has a greater impact on the overall spectrum; on the contrary, if this ratio is lower, the soil reflectance dominates. Through this explicit weight design, the present invention provides an ecological interpretation for the observed spectrum of each pixel: it can not only identify the spatio-temporal evolution of the vegetation condition, but also reveal various change forms of soil exposure or non-green coverage, thus making the observation of the ecological protection red line area more targeted and flexible.

[0051] Compared with traditional single threshold or fixed index, this mixed pixel decomposition method has higher adaptability. Because the areas within the ecological protection red line are not necessarily covered by dense and pure vegetation, but also include some shrubs, grasslands, desert margins and even human disturbance areas, where there may be a small amount of vegetation and bare soil or sandy surface within the same pixel range at the same time. If only relying on simple vegetation indices (such as the common normalized difference vegetation index) to judge, it often overestimates or underestimates the true coverage of the ecosystem and is difficult to reflect local subtle differences. By means of this linear decomposition, the present invention decomposes the surface reflectance at each moment into two main components, namely vegetation and soil, and simultaneously estimates the corresponding coverage ratios, thus accurately capturing the diversity and differences of the surface reflectance characteristics and ecological structures. After the decomposition is completed, the vegetation reflectance becomes an important indicator to measure the change of green coverage, while the soil reflectance is used to evaluate the impact brought by the change of bare surface or background interference. This "two-factor decomposition" has significant advantages in the monitoring of the ecological protection red line area.

[0052] More notably, the results output by this decomposition method are not only meaningful at the pixel scale, but can form a continuous ecological information distribution at a larger spatial scale. When processing the entire remote sensing image, each pixel will obtain the corresponding vegetation coverage ratio and the reflectance values of vegetation and soil. Combining with high-resolution geographical information, a quantitative description of the vegetation-soil structure in the ecological red line area can be formed. This can not only assist the artificial intelligence model in discriminating the change trend, but also provide a visual basis for decision-makers or ecologists to deeply understand the state of the ecosystem. For example, in which areas the vegetation coverage is gradually decreasing, in which places the increase in soil reflectance means an increase in the risk of drought or desertification, and whether there is an expansion of bare land caused by external construction, etc., can all be obtained through the analysis of the decomposed data for early warning and precise intervention.

[0053] In addition, this spectral inversion for mixed pixels is closely connected with the subsequent deep learning process in the present invention, and the two promote each other. The decomposed vegetation and soil reflectance components will be input into the deep network in the next step to form a sparser and more ecologically structured feature space, so that the neural network is no longer interfered by a large amount of mixed information or noise when discriminating ecological changes. Correspondingly, the network can also more accurately learn the temporal patterns and ecological driving factors behind the changes in surface cover. This logical chain of "decompose first, then learn" greatly reduces the convergence difficulties that the model may encounter when facing non-pure pixels and significantly improves the detection sensitivity to weak changes and early micro-disturbances. For some gradually occurring ecological degradation phenomena, even if the overall spectral change degree is not significant, but the vegetation coverage is slowly declining, such subtle trends can often be captured in the decomposition results of the present invention and intelligent judgments with high confidence can be made accordingly.

[0054] From a more macroscopic perspective, this pixel decomposition method based on linear combination also provides a general idea for multi-source remote sensing data, cross-temporal comparison, and cross-regional ecological monitoring. In the dynamic management of the ecological protection red line, it is often necessary to integrate datasets with different platforms, different bands, and different time resolutions. The spatial resolution and spectral resolution of these remote sensing data may vary, but as long as appropriate coverage factors and basic reflectance templates can be found, this decomposition process can be applied to unify the data caliber and generate comparable vegetation and soil reflectance information. In this way, the artificial intelligence model can also avoid the deviation caused by inconsistent bands or different degrees of pixel mixing during the subsequent training and application processes, and better realize the integrated analysis of cross-scale and cross-platform ecological data.

[0055] Furthermore, in step 3, the output change probability is: Where: represents time The lower pixel change probability; represents the sigmoid activation function; represents the number of layers of the deep learning network, with a value greater than or equal to 6; represents the weight of the layer; represents the non-linear activation function of the and represent two different times; represents the time of the lower pixel surface reflectance; represents the time of the lower pixel surface reflectance; is the time of the lower pixel proportion of vegetation coverage area; is the time of the lower pixel proportion of vegetation coverage area; is the time of the lower pixel vegetation reflectance spectral data; is the time of the lower pixel soil reflectance spectral data; is the time of the lower pixel soil reflectance spectral data.

[0056] In the formula, the change probability is the determination result of the change confidence of the pixel at time The value range is between 0 and 1, reflecting a probabilistic estimation of the ecological disturbance intensity. This output is completed through a composite structure, and the outermost layer uses the sigmoid activation function , which is often used to compress continuous values between 0 and 1, suitable for expressing the probability judgment of "whether to change", and also makes the model have good gradient propagation characteristics, facilitating the convergence and generalization of the network. The input received by the sigmoid function is not a simple difference term, but a weighted combination result after the action of multiple non-linear functions, specifically defined by this structure.

[0057] The non-linear mapping function of each layer represents the Layer non - linear transformation, whose design follows the modeling principle of neural network's response to multi - scale ecological changes. As the number of layers deepens, these transformation functions evolve from initially extracting local boundary and difference features to gradually becoming encoders of structural hierarchical features. For example, in the initial several layers of the model, the rectified linear unit (ReLU) function may be adopted to capture local significant change regions; the exponential linear unit (ELU) is used in the middle layers to enhance the response ability to weak gradual changes; while the hyperbolic tangent function (tanh) is used in the high - level layers to strengthen the representation ability of temporal non - linear evolution features. This layer - by - layer combination of activation functions and the attenuation factor of the weight coefficient is combined (such as ), enabling the model to not only retain the sensitivity of the shallow layer to the original change intensity but also emphasize the dominant position of the deep - layer structure in ecological pattern judgment, thus taking into account both local differences and overall structural evolution in change detection.

[0058] In terms of input feature construction, the present invention combines three types of key ecological change indicators and inputs them into the network: the absolute difference of the original spectral reflectance , the coupling term of vegetation cover change and vegetation reflectance change , and the soil reflectance change term . These three features are combined with weight coefficients of 0.2, 0.5, and 0.3 respectively, reflecting the hierarchical design of the model's sensitivity to different change types. The difference in the original reflectance mainly reflects the direct change in the overall energy response of pixels and is an important feature for capturing macroscopic disturbances; the change in vegetation cover combined with the reflectance difference focuses more on the health status and distribution structure changes of the green ecological system and is the most direct ecological indicator within the red line area; the soil reflectance term takes into account the dynamic changes of the background bare land and is of great significance for detecting non - vegetation changes such as human disturbances, bare soil expansion, and soil drought. The combination of these three forms a multi - dimensional modeling of the ecological system change mechanism and also reflects the precise matching made by the present invention between feature selection and ecological response.

[0059] In addition, to further improve the model's ability to understand band differences, the present invention also introduces band - sensitive weights into the network structure, including , , which are respectively used to adjust the sensitivity of the th layer to the difference in the original reflectance, the vegetation change term, and the soil change term at the th band. This design enables the deep - learning model to automatically learn the response feature differences of ground object changes at different wavelengths, thereby dynamically adjusting the importance of feature channels during the information fusion process. For example, in the near - infrared band, the reflectance difference of vegetation is the most significant, so the weight will automatically increase, while in the mid - infrared or short - wave infrared bands, soil moisture and bare state are more sensitive, The weight may be higher. This mechanism essentially constructs a "band attention mechanism", which is a deep fusion strategy between remote sensing data and deep networks at the ecological understanding level.

[0060] Therefore, this change probability output model is not only a mathematical mapping function, but also a composite ecological change judgment system that integrates remote sensing spectral knowledge, ecological structure changes, and artificial intelligence feature extraction mechanisms. By systematically modeling spectral differences, structural evolution, and spatial responses, and combining the strong expression ability of multi-layer neural networks, the present invention constructs a probability estimation mechanism with ecological interpretability and intelligent discrimination ability. It can not only capture significant changes in the ecological protection red line area, but also stably output change probabilities in complex scenarios such as coexistence of multiple weak perturbations, coordinated changes between vegetation and soil, and blurred spectral contrast, providing a solid data foundation and model guarantee for the intelligent supervision of the ecological red line.

[0061] Furthermore, ; when is greater than or equal to 1 and less than 3, is the rectified linear unit function; when is greater than or equal to 3 and less than 5, is the exponential linear unit function; when is greater than or equal to 5 and less than , is the hyperbolic tangent function.

[0062] Specifically, for the weight assignment of the th layer in the network, the present invention adopts an adaptive decreasing design method, that is, the weight of each layer is , where represents the total number of network layers. This design retains the contribution of the deep structure to change determination while appropriately suppressing the noise interference that may be introduced by over-response in the shallow layer. Since ecological changes often have implicit structures and slow evolution trends in space and time scales, shallow models are prone to misjudgment when facing apparent changes (such as sudden spectral perturbations, image noise), while deep networks can extract long-term stable evolution patterns through more complex non-linear transformations. Therefore, the present invention makes the weight increase linearly with the layer, while remaining overall stable due to the denominator being , making the deep output more influential but not overly strengthened, thus achieving a deep balance of change perception.

[0063] More importantly, in the design of the activation function , the present invention does not adopt a fixed structure, but sets a segmented non-linear transformation mechanism according to the functional positioning of different layers: when At the initial layer of the network, the rectified linear unit (ReLU) is used as the activation function. Its core advantage lies in its ability to quickly respond to strong mutations in input features, making it suitable for capturing direct and significant perturbation features such as original spectral changes and sharp fluctuations in coverage ratios. The model response at this stage tends to "capture", that is, it sensitively screens potential change factors and transmits rich local information to subsequent layers.

[0064] When the number of layers enters the middle stage, that is , the model activation function switches to the exponential linear unit (ELU), which is a function structure with strong non-linear response ability. It is particularly suitable for processing ecological data with certain slow-varying characteristics but where the change signal may be negative. ELU has a smooth decreasing property when the output is negative, and can make a delicate response to the "micro-perturbation" change characteristics commonly seen in the ecological red line, such as slight degradation of vegetation cover and slow change of soil moisture. This enables the model not to discard these weak ecological signals as noise, while maintaining the balance of the input distribution, thereby enhancing the model's recognition ability for transitional regions.

[0065] In the deep stage of the network, that is , the present invention selects the hyperbolic tangent function (tanh) as the activation mechanism. This is because the tanh function has good bidirectional symmetry and smoothness in the [-1,1] interval, and is particularly suitable for expressing the long-term trends of "positive and negative evolution" in ecosystems. For example, the gradual growth and degradation of vegetation cover, and the slow contraction and expansion of wetlands, can be characterized by the positive and negative output directions and amplitudes of tanh. In addition, the saturation property of tanh helps the model suppress extreme input values, preventing non-ecological jumps in change probabilities when affected by a single strong interference factor, making the model more robust in reflecting the ecological evolution trends at the regional scale.

[0066] This segmented activation strategy combines the response advantages of different functions in the feature space, enabling the model to form a progressive expression framework of "change reception - change understanding - change determination" throughout the network hierarchy. The shallow ReLU quickly captures possible changes, the middle-layer ELU eases the transition and extracts details, and the deep-layer tanh makes comprehensive judgments and spatial structure modeling. This design not only improves the model's adaptability to change intensity and change structure, but also makes the final output change probability interpretable in mathematical logic and has practical value in ecological significance.

[0067] Furthermore, in step 4, the adaptive change detection threshold is set as: Where: is the detection reference value, and its value range is a real number greater than 0 and less than 1. The larger it is, the greater the sensitivity of the ecological protection red line area.

[0068] The design logic of this structure is very clear: when the change range of the vegetation reflectance of a pixel between two time points is large, it indicates that significant ecological evolution may have occurred in this area. At this time, the fractional term in the exponential function is large, resulting in a significant attenuation of the entire threshold That is to say, the system becomes more "tolerant" of the detection results. As long as the change probability output by the model is slightly higher than this lower threshold, it can be determined that a change has occurred, ensuring that major disturbances are not missed. When the vegetation reflectance changes very little between two time points, the system considers this area to be relatively stable and there is no significant disturbance. At this time, the attenuation amplitude of the threshold is extremely small and basically remains at nearby, and the detection standard becomes "strict" to avoid misjudging normal fluctuations. This mechanism is particularly important in the ecological red line area because its goal is not only to identify severe damage, but also to include early warning of minor ecological trends and avoidance of false alarms. Therefore, it must have dynamic adaptability in the judgment standard.

[0069] The parameter in the formula is defined as the "detection reference value", which is a quantitative expression of the sensitivity of this area to ecological changes. Its value ranges from (0,1). The larger the value, the stronger the sensitivity of the system to changes, the lower the tolerance, and it is easy to determine changes; on the contrary, the smaller the value, it means that the system needs a higher significance of the change signal to respond in this area. The setting of this parameter can be assigned in combination with the management levels of different ecological function zones within the ecological protection red line area. For example, within the core protection area, can take 0.9 or even higher to enhance the model's response to minor changes; while within the buffer zone or restoration zone, this value can be appropriately reduced to prevent over-response to natural ecological seasonal fluctuations. This flexible configuration of the reference value also endows the present invention with high scalability and management integration capabilities, making it not only applicable to automatic discrimination, but also facilitating the connection with ecological policy standards to achieve the deep integration of technology and system.

[0070] In addition, the introduction of the exponential function itself also highly depicts the non-linear characteristics of the change response mechanism. During the ecological evolution process, the surface reflectance, especially the vegetation reflectance, does not grow or decay linearly, but often evolves in a stage-by-stage and critical manner. Small fluctuations do not necessarily represent trends, but once a certain change threshold is exceeded, a chain effect may occur. Therefore, by realizing the rapid attenuation of the threshold in an exponential form, the model actually implies an "ecological critical response mechanism", that is, once the change amplitude exceeds the threshold of the self-stabilizing ability of the ecosystem, the detection mechanism will quickly relax the judgment threshold to ensure the efficient identification of potential ecological turning events.

[0071] More importantly, the adaptive detection threshold mechanism is not only used for single-point determination at the pixel level, but also provides the basic support for subsequent spatial neighborhood consistency constraints. In the red line area where the spatial distribution is uneven and local disturbances occur frequently, whether a change occurs depends not only on the comparison between the probability of the pixel itself and the threshold, but also on the common evolution trend of the surrounding pixels. If the threshold of each pixel is automatically adjusted according to its own ecological state change, then the entire spatial determination system has a highly individualized but coordinated response judgment logic, realizing "multi-threshold collaborative determination" in the spatial sense. This mechanism overcomes the difficult problem of the traditional method being difficult to balance between spatial consistency and change sensitivity, and is the key breakthrough of the present invention in the direction of intelligent change detection.

[0072] Further, in step 4, according to the change probability and the adaptive change detection threshold, the detection result is determined as: Where: Denotes the pixel at time The detection result, 1 indicates change, and 0 indicates no change; Denotes the spatial neighborhood of pixel ; Denotes the indicator function, which is 1 when the condition is true and 0 otherwise; Denotes the number of pixels in the neighborhood; Denotes the neighborhood consistency threshold, with a value range of 0 - 1; And Denote the abscissa and ordinate of other pixels in the spatial neighborhood of pixel respectively; Denotes the time The change probability of pixel at time ; Denotes the time The adaptive change detection threshold of pixel

[0073] The core structure of the determination formula is a piecewise function in the form of a conditional expression, and its output result Denotes the final detection state of pixel at time , with a value of 1 indicating being determined as "changed" and a value of 0 indicating "not changed". First, the system will judge whether the change probability of the current pixel is higher than its corresponding adaptive change detection threshold This judgment reflects the confidence level of the system in the ecological state change of the current pixel between two time phases. Only when this value significantly exceeds the threshold can it indicate that there is a real structural change trend at this point. However, to avoid "false changes" caused by factors such as local noise, data loss, or abnormal single-point reflectivity, the present invention further introduces a spatial neighborhood consistency judgment as an auxiliary criterion.

[0074] Specifically, for each pixel, its spatial neighborhood is a set composed of multiple adjacent pixels, usually existing in the form of a sliding window such as 3×3 or 5×5 in a two-dimensional image. The system will count how many pixels in this neighborhood range also have a change probability higher than its corresponding adaptive threshold This judgment is achieved through an indicator function When a certain pixel in the neighborhood satisfies the condition of "change probability higher than the threshold", its corresponding value is 1, otherwise it is 0. Subsequently, the indicator function values of all neighborhood pixels are summed to obtain the number of pixels determined to have changed in the current neighborhood. This number will be compared with a threshold where is the preset neighborhood consistency threshold, is the total number of neighborhood pixels. If the number of satisfied pixels exceeds or equals this ratio threshold, it indicates that the change is not an isolated phenomenon but shows spatial continuity. Finally, the system will determine the current pixel

[0075] as having changed (i.e., the output is 1); otherwise, it is determined as unchanged (the output is 0).

[0075] The design of this dual judgment mechanism essentially embeds the idea of spatial consistency into the change judgment to enhance the system's discrimination ability for real disturbance signals in a complex ecological background. In the ecological protection red line area, real ecological changes often do not occur only in an isolated pixel but show continuous, clustered, or boundary-expanding evolution trends. For example, phenomena such as vegetation degradation, land reclamation, or water body reduction have obvious spatial expansion characteristics in remote sensing images, while "false changes" such as natural noise, local shadows, and imaging errors often show random and isolated pixel response characteristics. Therefore, only when most pixels in a local area jointly meet the change conditions is it ecologically reasonable.

[0076] At the same time, this judgment structure also enables the system to have strong adjustment ability and adaptability. By adjusting the size of the consistency threshold , the tolerance degree of the system to local changes can be controlled. A higher value (such as 0.8 or 0.9) requires that the vast majority of pixels in the neighborhood show consistent changes, which is applicable to highly sensitive ecological areas and emphasizes false alarm control; while a lower Values (such as 0.3 or 0.5) allow the change determination to hold even in cases of partial local inconsistency, and are applicable to transitional zones of change boundaries or areas with frequent human disturbances. Through this flexible setting, the system can not only improve the accuracy of change determination as a whole, but also adapt to different management zones, functional areas or natural geomorphic features within the ecological red line area according to local conditions, realizing an intelligent discrimination strategy with spatial differentiation.

[0077] In addition, this spatial neighborhood consistency mechanism helps to improve the spatial coherence and structural stability of the model output results, avoiding problems such as "speckled" change patterns or fragmented edges in the final change image, making the change detection results more geographically interpretable and more suitable for subsequent ecological statistical analysis, trend simulation and protection intervention. Combining with the aforementioned adaptive change detection threshold mechanism, the present invention constructs a change detection determination system composed of three-layer logic of probability output - threshold adjustment - spatial constraint, forming an intelligent identification mechanism for the whole chain, multi-scale and multi-dimension of ecological disturbance events from perceiving the amplitude of ecological changes, judging their credibility, to verifying their spatial structure consistency.

[0078] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection required by the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based ecological protection red line area change detection method, characterized in that: include: Step 1: Perform atmospheric correction and radiation correction on the remote sensing data at each time to convert the remote sensing data at each time into directly comparable surface reflectance; Step 2: splitting the mixed reflectance spectrum in the remote sensing data into vegetation reflectance spectrum data and soil reflectance spectrum data by pixel decomposition method; both the vegetation reflectance spectrum data and the soil reflectance spectrum data are dual-phase data; Step 3: Establish a deep learning model, take the difference between vegetation reflectance spectral data and soil reflectance spectral data of dual-phase data as input, extract the change characteristics of ecological protection areas, and output the change probability; Step 4: Set the adaptive change detection threshold according to the sensitivity of the ecological protection red line area; determine the detection result according to the change probability and the adaptive change detection threshold.

2. The method for detecting changes in ecological protection red line areas based on artificial intelligence according to claim 1, characterized in that: In step 1, at time The surface reflectivity at time is: in: Indicates time Lower pixel Surface reflectivity; Indicates time The pixels received The radiance of Indicates the distance between the earth and the sun; Indicates time Solar irradiance under Indicates time The solar zenith angle below; is the horizontal coordinate of the pixel; is the vertical coordinate of the pixel.

3. The method for detecting changes in ecological protection red line areas based on artificial intelligence according to claim 2, characterized in that: The radiation brightness is obtained through the following process: a radiation detection instrument is installed on the remote sensing sensor; the radiation detection instrument includes a multi-spectral sensor and an imaging spectrometer; when the remote sensing sensor passes over the ground, the sunlight reflected from the surface propagates through the atmosphere to reach the photosensitive element of the remote sensing sensor, and the remote sensing sensor records the electromagnetic radiation energy intensity received in different bands and stores it in the form of digital grayscale values; the digital grayscale values ​​recorded by the remote sensing sensor are converted into radiation brightness with physical meaning through radiation calibration processing.

4. The method for detecting changes in ecological protection red line areas based on artificial intelligence according to claim 3, characterized in that: The process of converting the digital grayscale value recorded by the remote sensing sensor into a radiation brightness with physical meaning through radiation calibration processing includes: obtaining the digital grayscale value calibration parameters, including gain and bias, converting through the radiation brightness formula, and obtaining the radiation brightness value of each pixel in the remote sensing data in a specific band. The radiation brightness formula is: radiation brightness = gain * digital grayscale value + bias. The unit of radiation brightness is W·m⁻²·sr⁻¹·μm⁻¹.

5. The method for detecting changes in ecological protection red line areas based on artificial intelligence according to claim 4, characterized in that: In step 2, the mixed reflectance spectrum in the remote sensing data is split into vegetation reflectance spectrum data and soil reflectance spectrum data by pixel decomposition method: in: For time Lower pixel The proportion of vegetation cover area; For time Lower pixel Spectral data of vegetation reflectance; For time Lower pixel Soil reflectance spectral data.

6. The method for detecting changes in ecological protection red line areas based on artificial intelligence according to claim 5, characterized in that: In step 3, the output change probability is: in: Indicates time Lower pixel The probability of change; Represents the sigmoid activation function; Indicates the number of layers of the deep learning network, the value is greater than or equal to 6; Indicates The weight of the layer; Indicates Non-linear activation function of the layer; and Indicates two different times; Indicates time Lower pixel Surface reflectivity; Indicates time Lower pixel Surface reflectivity; For time Lower pixel The proportion of vegetation cover area; For time Lower pixel The proportion of vegetation cover area; For time Lower pixel Spectral data of vegetation reflectance; For time Lower pixel Soil reflectance spectral data; For time Lower pixel Soil reflectance spectral data.

7. The method for detecting changes in ecological protection red line areas based on artificial intelligence according to claim 6, characterized in that: ;when When greater than or equal to 1 and less than 3, is a linear rectification function; when When greater than or equal to 3 and less than 5, is an exponential linear function; when Greater than or equal to 5 and less than hour, is the hyperbolic tangent function.

8. The method for detecting changes in ecological protection red line areas based on artificial intelligence according to claim 7, characterized in that: In step 4, set the adaptive change detection threshold for: in: is the detection reference value, and its value range is a real number greater than 0 and less than 1. The larger it is, the more sensitive the ecological protection red line area is.

9. The method for detecting changes in ecological protection red line areas based on artificial intelligence according to claim 8, characterized in that: In step 4, according to the change probability and the adaptive change detection threshold, the detection result is determined as: in: Expressed as time Lower pixel The test result, 1 means change, 0 means no change; Represents pixel The spatial neighborhood of Represents an indicator function, which is 1 when the condition is met, otherwise it is 0; Represents the number of pixels in the neighborhood; Indicates the neighborhood consistency threshold, the value range is 0-1; and Represents pixels The horizontal and vertical coordinates of other pixels in the spatial neighborhood of ; Indicates time Lower pixel The probability of change; Indicates time Lower pixel Adaptive change detection threshold.

Images