A method and device for remote sensing instant detection of vegetation anomalies

Abstract

The present invention discloses a remote sensing instant detection method and device for vegetation anomalies, comprising: dividing the surface reflectance image of a target area within a study period into a historical period and a monitoring period, performing preprocessing on each period, synthesizing all data from each year and month in the historical period into monthly data for each year; calculating multiple vegetation indices pixel by pixel based on the preprocessed monthly data for each year and the image to be detected; performing two linear regressions on the vegetation index of each pixel's monthly data to sequentially remove noise and vegetation growth trend information; performing kernel density estimation on the historical residuals of the vegetation index of each pixel at each time point as sample data to construct a normal vegetation pattern for each pixel in each month; then finding the normal vegetation pattern for the corresponding month based on the month to which the image to be detected belongs; and determining, pixel by pixel, whether vegetation anomalies occur based on the predicted value of the vegetation index of each pixel in the corresponding month and the boundary of the normal vegetation pattern. The present invention can improve the universality of instant detection of vegetation anomalies.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a remote sensing instant detection method and device for vegetation anomaly, which can be applied to dynamic change monitoring research of vegetation remote sensing. Background Art

[0002] Forests and other vegetation, as a crucial component of Earth's surface, are crucial for the Earth's carbon and water cycles and energy exchange, playing an irreplaceable role in maintaining global ecological balance. Currently, vegetation loss caused by increasingly active human activities or natural factors such as deforestation, fires, and pests and diseases is considered vegetation anomaly. Real-time detection of vegetation anomalies, timely monitoring of changes in vegetation cover, and intervention in unsustainable human activities can help maintain ecosystem stability.

[0003] Vegetation anomaly detection falls under the category of change detection. Remote sensing data, due to its high frequency and large coverage, is the mainstream tool for change detection. Furthermore, a growing number of historical remote sensing images are playing a significant role in detecting and quantifying land cover changes. Over the past decade, numerous parametric and non-parametric vegetation anomaly detection algorithms have been developed using Landsat time series data, including the Landsat-based detection of Trends in Disturbance and Recovery (LandTrendr), the Breaks for Additive Seasonal and Trend (BFAST), the Continuous Change Detection and Classification (CCDC), and the Anomaly Vegetation Change Detection (AVOCADO). However, these algorithms primarily focus on detecting historical anomalies and lack immediacy.

[0004] To more quickly capture vegetation changes, the demand for real-time detection is increasing. Most existing real-time detection methods have been developed specifically for tropical deforestation and can be roughly divided into two categories. The first category is based on conditional probability. The main idea is to classify a vegetation characteristic value into past, present, and future time periods. For each time period, the conditional probability of the characteristic value being non-forest is calculated. If the probability at time t is greater than 0.5, the moment is considered a potential anomaly. The observations at time t-1, t, and the future t+i are then iterated through a Bayesian update calculation to confirm or reject the anomalous event at time t. The second category is based on vegetation phenological parameter models. The main idea is to use historical time series data of a vegetation characteristic, fit the normal phenology based on sine, cosine, or bilogistic functions, and predict the observed value at the time t to be detected. The deviation between the actual observation and the predicted value is used as a disturbance indicator.

[0005] However, the application scenarios of existing instant detection methods are relatively single, and there is no universal method that can be applied to different types of vegetation anomalies; the mathematical theory of the phenological parameter model algorithm is relatively complex, and the detection results are affected by the parameter fitting effect; in addition, in terms of the characteristics used in the algorithm, most methods only use a certain vegetation index and may ignore the information of other bands. Summary of the Invention

[0006] In view of this, the main purpose of the present invention is to provide a remote sensing instant detection method and device for vegetation anomalies, so as to improve the universality of instant detection of vegetation anomalies.

[0007] According to a first aspect of the present invention, a method for real-time remote sensing detection of vegetation anomalies is provided, comprising:

[0008] Collect all available Landsat series of surface reflectance images of the target area during the study period;

[0009] The collected surface reflectance images are divided into historical period and monitoring period. The historical period data are used to construct the normal vegetation pattern for each month, and the monitoring period data are used for the real-time detection of vegetation anomalies.

[0010] The historical data and the images to be detected during the monitoring period were preprocessed separately, and all the data of each year and month in the historical period were processed into monthly data of each year through mean synthesis;

[0011] Calculate multiple vegetation indices pixel by pixel for the pre-processed historical monthly data and the images to be detected;

[0012] The vegetation index of each pixel in each year and month is subjected to two time series linear regressions to remove noise and vegetation growth trend information. The historical residual of the difference between the actual value and the regression value of the vegetation index of each pixel at each historical time point is obtained based on the result of the second linear regression, and the predicted value of the vegetation index of each pixel in the corresponding month in the future is obtained.

[0013] The historical residuals of the vegetation index of each pixel are used as sample data for non-parametric kernel density estimation to construct the vegetation normal pattern of each pixel in each month, and the boundaries of the vegetation normal pattern of each pixel in each month are determined according to the set probability threshold.

[0014] According to the month to which the image to be detected belongs, the normal vegetation pattern of the corresponding month is found. Combined with the predicted value of the vegetation index of each pixel in the corresponding month in the future and the boundary of the normal vegetation pattern, it is judged pixel by pixel whether the vegetation in the target area is abnormal.

[0015] According to a second aspect of the present invention, there is provided a remote sensing instant detection device for vegetation anomalies, comprising:

[0016] An image collection unit is used to collect all available Landsat series surface reflectance images of the target area during the study period;

[0017] The period division unit is used to divide the collected surface reflectance images into historical periods and monitoring periods. The historical period data is used to construct the normal vegetation pattern for each month, and the monitoring period data is used for the real-time detection of vegetation anomalies.

[0018] The preprocessing unit is used to preprocess the historical period data and the images to be detected in the monitoring period respectively, and process all the data of each year and month in the historical period into the historical year-by-month data by means of mean synthesis;

[0019] The vegetation index calculation unit is used to calculate multiple vegetation indices for each pixel of the pre-processed historical monthly data and the image to be detected;

[0020] The time series processing unit is used to use two time series linear regressions on the vegetation index of each pixel in each historical year and month to remove noise and vegetation growth trend information in turn, and obtain the historical residual of the difference between the actual value and the regression value of the vegetation index of each pixel at each historical time point based on the second linear regression result, and obtain the predicted value of the vegetation index of each pixel in the corresponding month in the future;

[0021] A normal pattern construction unit is used to perform non-parametric kernel density estimation on the historical residuals of the vegetation index of each pixel as sample data to construct the vegetation normal pattern of each pixel in each month, and determine the boundary of the vegetation normal pattern of each pixel in each month according to a set probability threshold;

[0022] The anomaly discrimination unit is used to find the normal vegetation pattern of the corresponding month according to the month to which the image to be detected belongs, and to judge whether the vegetation in the target area is abnormal pixel by pixel based on the predicted value of the vegetation index of each pixel in the corresponding month in the future and the boundary of the normal vegetation pattern.

[0023] According to a third aspect of the present invention, an electronic device is provided, comprising: a processor and a memory storing computer executable instructions, wherein the executable instructions, when executed by the processor, implement the aforementioned method for real-time remote sensing detection of vegetation anomalies.

[0024] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores one or more programs, and when the one or more programs are executed by a processor, the aforementioned remote sensing instant detection method of vegetation anomalies is implemented.

[0025] The beneficial effects of the present invention are:

[0026] The present invention is aimed at various typical anomaly types of vegetation, and uses multiple vegetation indices that reflect the vegetation status from different aspects. Combined with non-parametric kernel density estimation, it can be universally applicable to the detection of different vegetation anomalies in different regions; and the entire anomaly detection process of the present invention is carried out pixel by pixel, and the vegetation normal model is constructed on a monthly basis and the pixel-by-pixel processing can further improve the universality of the present invention in performing anomaly detection at different time points and different spatial positions on the image, so that it is not limited to the region and vegetation type. Once there is an available image, the real-time detection of vegetation anomalies can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] By reading the detailed description of the preferred embodiment below, all advantages and benefits will become clear to those skilled in the art. The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention. In addition, the same reference symbols are used to represent the same components throughout the drawings. In the drawings:

[0028] Figure 1 A schematic diagram showing a flow chart of a method for real-time remote sensing detection of vegetation anomalies according to an embodiment of the present invention;

[0029] Figure 2 To correspond Figure 1 A flowchart of the remote sensing instant detection of vegetation anomalies according to the method shown;

[0030] Figure 3 A schematic diagram of two linear regressions of NDVI and NBR indices according to an embodiment of the present invention is shown;

[0031] FIG4 shows a schematic diagram of probability density after kernel density estimation according to an embodiment of the present invention;

[0032] FIG5 shows a schematic diagram of pixel abnormality discrimination according to an embodiment of the present invention;

[0033] Figure 6 A block diagram of a remote sensing instant detection device for vegetation anomalies according to an embodiment of the present invention is shown;

[0034] Figure 7 A block diagram of an electronic device according to an embodiment of the present invention is shown;

[0035] Figure 8 Landsat image data of the Colombian Amazon forest according to Example 1 of the present invention is shown;

[0036] Figure 9 It shows the shortwave infrared 2-band graph and the corresponding real-time detection result graph for each date to be detected in Example 1 of the present invention;

[0037] Figure 10 The Landsat image data of Daxing'anling in Genhe Town, Inner Mongolia, according to Example 2 of the present invention is shown;

[0038] Figure 11 The diagram shows the near-infrared bands for each date to be immediately detected and the corresponding immediately detected result diagrams according to the second embodiment of the present invention. DETAILED DESCRIPTION

[0039] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. These embodiments are provided to facilitate a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention may be implemented in various forms and should not be limited by the embodiments set forth herein.

[0040] Figure 1 The following is a flow chart showing a method for real-time remote sensing detection of vegetation anomalies according to an embodiment of the present invention. Figure 2 To correspond Figure 1 The process framework diagram of the remote sensing real-time detection of vegetation anomalies of the method shown. Figure 1 and Figure 2 The method of the embodiment of the present invention includes the following steps S110 to S160:

[0041] Step S110 : collecting all available Landsat series surface reflectance images of the target area within the study period.

[0042] Landsat (Land Resources Satellite) data is a key data source for studying land cover change across various types, due to its long-term continuous observation record, high spatial resolution, and free access. This study collected all available Landsat surface reflectance products for the target area during the study period, including those from Landsat 4, Landsat 5, Landsat 7, and Landsat 8.

[0043] Step S120 , dividing the collected surface reflectance images into a historical period and a monitoring period, wherein the historical period data is used to construct a normal vegetation pattern for each month, and the monitoring period data is used for real-time detection of vegetation anomalies.

[0044] The collected time series data of Landsat surface reflectance images are divided into a historical period and a monitoring period. The data in the historical period should be relatively stable without obvious anomalies and be used to construct the normal vegetation pattern for each month. The monitoring period usually starts in the year following the end of the historical period, and the data in the monitoring period are used for the immediate detection of vegetation anomalies.

[0045] Step S130 , pre-processing the historical period data and the images to be detected in the monitoring period, and processing all the data of each year and month in the historical period into monthly data of each year in the historical period by using a mean synthesis method.

[0046] Because the collected surface reflectance image data involves multiple satellite sensors with different band settings, and each image of the same area has certain differences in spatial position, all data need to be unified and spatially registered during the preprocessing step. The processed data has a total of seven bands: blue, green, red, near-infrared, shortwave infrared 1, shortwave infrared 2, and QA (Quality Assessment) quality control band, and have the same spatial range.

[0047] In addition, clouds, cloud shadows, snow accumulation, etc. will affect the true reflectivity of the surface, so they need to be masked in combination with the QA band provided by the surface reflectivity product to prevent them from participating in subsequent processing.

[0048] Both the historical data and the images to be detected during the monitoring period need to be preprocessed by band unification, spatial registration, and removal of clouds, cloud shadows, and snow. In addition, all the data of each year and month in the historical period are further processed into monthly data for each historical year through mean synthesis, which facilitates the subsequent construction of the normal vegetation pattern for each month.

[0049] Step S140 , calculating a plurality of vegetation indices for each pre-processed historical monthly data and each pixel of the image to be detected.

[0050] The vegetation index calculated in step S140 for each pixel of the pre-processed historical monthly data and the image to be detected includes but is not limited to: Normalized Difference Vegetation Index (NDVI) and Normalized Burn Ratio (NBR);

[0051] The NDVI index amplifies the reflectivity difference between the near-infrared band and the red band and can be used to indicate the degree of vegetation cover. The NBR index amplifies the reflectivity difference between the near-infrared band and the short-wave infrared band and can be used to indicate the degree of water shortage in vegetation leaves. These two indices reflect the status of vegetation from the perspectives of vegetation greenness and water content, respectively. The specific calculation formulas are as follows:

[0052]

[0053]

[0054] Among them 、 、 They are the reflectances of the near infrared, red and short-wave infrared bands respectively.

[0055] It is understood that in addition to calculating the two vegetation indices NDVI and NBR pixel by pixel, the present invention can also calculate other vegetation indices as needed, such as the Enhanced Vegetation Index EVI (Enhanced Vegetation Index) and the Moisture Stress Index MSI (Moisture Stress Index).

[0056] In step S150, two time series linear regressions are performed on the vegetation index of each pixel in each historical year and month data to remove noise and vegetation growth trend information in turn. Based on the result of the second linear regression, the historical residual of the difference between the actual value and the regression value of the vegetation index of each pixel at each historical time point is obtained, and the predicted value of the vegetation index of each pixel in the corresponding month in the future is obtained.

[0057] This step S150 is mainly aimed at historical period data, that is, the pre-processed historical monthly data of each year is used to remove noise and vegetation growth trend information in sequence by using two linear regressions month by month.

[0058] Because the QA band mask is not thorough, residual noise such as clouds and cloud shadows may still be present in the image, causing errors in the subsequent construction of the vegetation normal model and thus requiring removal. Furthermore, the vegetation normal model is based on time series data, which contains information on vegetation growth trends, which also needs to be removed. This shows that step S150 can also, to a certain extent, mitigate growth differences between different vegetation types, thereby increasing the universality of the present invention.

[0059] Cloud information often appears as a downward "spike" in the NDVI index time series. For each pixel's monthly data, the NDVI index is used to perform a first time series linear regression. If the actual value at a historical time point is lower than the regression value, and the difference between the two is greater than a threshold, the point is considered a noise point and needs to be removed. The NDVI and NBR indices involved in anomaly detection are then subjected to a second time series linear regression on the retained historical time points. By subtracting the actual value of each historical time point from the regression value, vegetation growth trend information can be removed.

[0060] Therefore, in step S150, two time series linear regressions are used to remove noise and vegetation growth trend information from the vegetation index of each pixel's historical monthly data. Specifically, the following steps may be performed:

[0061] For each pixel's historical monthly data, the NDVI index is used to perform the first time series linear regression. If the actual value at a certain time point is lower than the regression value, and the difference between the two is greater than the set elimination threshold, the time point is eliminated as a noise point, and the data at the corresponding time point in the NBR index is also eliminated.

[0062] A second time series linear regression was performed on the retained time points of the NDVI index and NBR index to obtain the difference between the actual value and the regression value at each time point to remove the vegetation growth trend information.

[0063] Figure 3 A schematic diagram of double linear regression of NDVI and NBR indices according to an embodiment of the present invention is shown. Figure 3 In the equation, "1st_regression" and "2nd_regression" represent the first linear regression and the second linear regression respectively, "data to be removed" represents the data to be removed, and "final data" represents the data to be retained. Figure 3 In the embodiment shown, the noise point removal threshold set in the first linear regression is 0.1, and the first time series linear regression is performed on the NDVI index. Data that does not meet the threshold is considered to be residual cloud information and is removed ( Figure 3 The dots in the figure above) are also removed. Figure 3 The dots in the figure below); the data retained for NDVI and NBR index ( Figure 3 A second time series linear regression is performed to remove trend information.

[0064] According to the second linear regression results, the historical residual of the difference between the actual value and the regression value of the vegetation index of each pixel at each historical time point can be obtained. The historical residual is further used to construct the normal vegetation model.

[0065] The results of the second linear regression also provide a predicted vegetation index value for each pixel in the corresponding month. This prediction bridges the gap between the historical period and the monitoring period. The second linear regression results provide a monthly trend in the vegetation index for each pixel. Based on this trend, the predicted index value for the month in which the image to be detected appears can be predicted. The residual difference between the predicted value and the actual index value of the image to be detected in that month is then determined. This residual difference is then used to determine whether vegetation anomalies are occurring.

[0066] In step S160 , the historical residual of the vegetation index of each pixel is used as sample data to perform non-parametric kernel density estimation to construct the normal vegetation pattern of each pixel in each month, and the boundary of the normal vegetation pattern of each pixel in each month is determined according to a set probability threshold.

[0067] In step S160, kernel density estimation is performed using the historical residuals of the vegetation index for each pixel obtained in the previous step as sample data to approximate the normal distribution pattern of the data for each pixel in each month. Kernel density estimation is a non-parametric estimation method that fully utilizes the data itself, avoiding subjective prior knowledge. It fits the distribution density function of the overall data based on the sample set, thus achieving the greatest approximation of the sample data compared to parametric estimation.

[0068] Kernel density estimation can be implemented by calling the python sklearn KernelDensity package. The principle is to center a kernel function with a bandwidth of h at each observation point and take the average of all kernel functions until the final probability density estimate is obtained. Specifically,

[0069] definition is the sample data, each sample contains the historical residuals of N vegetation indices involved in the construction:

[0070]

[0071] Then the N-dimensional Gaussian kernel density estimation expression is:

[0072]

[0073] in: is a point in the N-dimensional vegetation index space, is the kernel function bandwidth, represents the Gaussian kernel function, Represents N-dimensional space The distance from the sample point.

[0074] It should be noted that N is the number of vegetation indices involved in the normal model construction. If N is 1, the kernel density estimation is one-dimensional; if it is 2, it is two-dimensional, and so on. Each vegetation index is equivalent to a dimension, and in actual calculations, they are calculated together according to the formula for N-dimensional kernel density estimation. Furthermore, since the historical residuals of each vegetation index can be considered to be approximately Gaussian distributed, the present invention calculates each kernel function as if it were consistent, and the expressions for N-dimensional kernel density estimation are all Gaussian kernel functions.

[0075] The methods for determining bandwidth include empirical formula method, cross-validation method, plug-in method, etc., which are affected by the number of samples and the dimension of the kernel function. When the kernel density estimation is N-dimensional, the empirical formula can be used To determine the kernel function bandwidth. Compared with other bandwidth determination methods, this empirical formula method is simpler and more convenient to apply. In addition, the parameter coefficients of the empirical formula have been repeatedly tested, and the kernel density estimation effect is better.

[0076] Figure 4 shows a schematic diagram of the probability density after kernel density estimation of an embodiment of the present invention. Among them, Figure 4 (a) is a schematic diagram of the probability density after two-dimensional kernel density estimation, and Figure 4 (b) is the projection of Figure 4 (a) on the xoy plane. In Figure 4, "NDVI_Diff" and "NBR_Diff" respectively represent the historical residuals of NDVI and NBR indices after two linear regressions, and "Kernel Density Estimation" represents kernel density estimation. Obviously, through kernel density estimation, the probability density of the distribution area in the data set is higher. If the probability density surface is intercepted by a plane parallel to xoy, then the integral of the area above the cross section can be used to represent the normal probability, and based on this, the approximate range of the normal distribution pattern of vegetation can be extracted according to a certain probability threshold. At this time, the boundary of the normal pattern of vegetation is the probability density corresponding to the cross section. As shown in the following formula:

[0077]

[0078] in: is the boundary of the normal vegetation pattern, that is, the probability density corresponding to the cross section, which can be obtained by of The quantiles are calculated, is the area above the cross section. From the above formula, we can see that the probability that this area contains at least normal , once the probability threshold is set, it can be determined , thereby determining the boundary of the normal distribution of vegetation patterns. For example, when the probability threshold is 95%, it means that at least 95% of the points are normal, and the probability that the points outside this range are normal is less than 5%. At this time, the boundary is the probability density at the 5% quantile of the kernel density estimation result.

[0079] Step S170 , finding the normal vegetation pattern for the corresponding month according to the month to which the image to be detected belongs, combining the predicted value of the vegetation index of each pixel in the corresponding month in the future and the boundary of the normal vegetation pattern, and determining pixel by pixel whether the vegetation in the target area is abnormal.

[0080] In step S150, the predicted value of the vegetation index of each pixel in the corresponding month in the future is obtained. In step S160, the vegetation normal pattern and its boundary are constructed through kernel density estimation. In this step S170, the actual value of the vegetation index of the image to be detected is subtracted from the predicted value of the corresponding month for each pixel to obtain the predicted value residual. The predicted value residual is then substituted into the vegetation normal pattern of the corresponding month to obtain the corresponding probability density. By comparing it with the probability density of the corresponding boundary, it is determined whether the vegetation in the target area is abnormal.

[0081] This step S170 specifically includes: finding the normal vegetation pattern of the corresponding month according to the month to which the image to be detected belongs, substituting the predicted value residual of the difference between the actual value of the vegetation index of the image to be detected and the predicted value of the corresponding month into the normal vegetation pattern of the corresponding month for each pixel, and comparing the obtained probability density with the probability density of the corresponding boundary. If the probability density of a pixel of the image to be detected is less than the probability density of the corresponding boundary, it is judged that the vegetation in the area corresponding to the pixel is abnormal; otherwise, it is judged that the vegetation in the area corresponding to the pixel is normal.

[0082] Figure 5 shows a schematic diagram of pixel anomaly discrimination according to one embodiment of the present invention. Figures 5(a) and 5(b) illustrate anomaly discrimination for different pixels to be discriminated. The "samples" dots represent sample points participating in kernel density estimation, and the "new data" cross-shaped dots represent points to be discriminated for anomalies. The contour lines in Figures 5(a) and 5(b) represent the normal boundaries corresponding to at least 95%, 90%, 75%, and 50% normal probabilities, respectively. Clearly, if 95% is used as the probability threshold, the cross-shaped dots in Figure 5(a) are identified as normal, while the cross-shaped dots in Figure 5(b) are identified as anomalies.

[0083] In summary, the present invention collects all available Landsat series surface reflectance images of the target area within the study period, and divides the collected surface reflectance images into a historical period and a monitoring period, wherein the historical period data is in months, and is used for non-parametric kernel density estimation to construct the normal vegetation pattern of each month, and the monitoring period data is used for real-time detection of vegetation anomalies; then the historical period data and the images to be detected in the monitoring period are preprocessed respectively, and all the data of each year and month in the historical period are processed into historical monthly data of each year by mean synthesis; then, a plurality of vegetation indices are calculated pixel by pixel for the preprocessed historical monthly data of each year and the images to be detected, and these vegetation indices can reflect the vegetation status from different aspects; then the historical monthly data of each pixel are calculated The vegetation index is used to remove noise and vegetation growth trend information in turn by using two time series linear regressions. The historical residual of the difference between the actual value and the regression value of the vegetation index of each pixel at each historical time point is obtained according to the result of the second linear regression, as well as the predicted value of the vegetation index of each pixel in the corresponding month in the future. The historical residual of the vegetation index of each pixel is used as sample data for non-parametric kernel density estimation to construct the vegetation normal mode of each pixel in each month, and the boundary of the vegetation normal mode of each pixel in each month is determined according to the set probability threshold. Finally, the vegetation normal mode of the corresponding month is found according to the month to which the image to be detected belongs. Combined with the predicted value of the vegetation index of each pixel in the corresponding month in the future and the boundary of the vegetation normal mode, it is judged pixel by pixel whether the vegetation in the target area is abnormal.

[0084] It can be seen that the present invention is aimed at various typical anomaly types of vegetation, and uses a variety of vegetation indices that reflect the status of vegetation from different aspects. Combined with non-parametric kernel density estimation, it can be universally applicable to the detection of different vegetation anomalies in different regions; and the entire anomaly detection process of the present invention is carried out pixel by pixel, and the construction of vegetation normal mode on a monthly basis and pixel-by-pixel processing can further improve the universality of the present invention in performing anomaly detection at different time points and different spatial positions on the image, thereby realizing no restrictions on regions and vegetation types. Once there is an available image, vegetation anomalies can be detected immediately.

[0085] The present invention also provides a remote sensing instant detection device for vegetation anomalies, which is based on the same technical concept as the aforementioned remote sensing instant detection method for vegetation anomalies. Figure 6 A block diagram of a remote sensing instant detection device for vegetation anomalies according to an embodiment of the present invention is shown. Figure 6 The remote sensing instant detection device 600 for vegetation anomalies of the present invention comprises:

[0086] The image collection unit 610 is used to collect all available Landsat series surface reflectance images of the target area during the study period;

[0087] The period division unit 620 is used to divide the collected surface reflectance images into a historical period and a monitoring period, wherein the historical period data is used to construct the normal vegetation pattern of each month, and the monitoring period data is used for the real-time detection of vegetation anomalies;

[0088] The pre-processing unit 630 is used to pre-process the historical period data and the images to be detected in the monitoring period, and process all the data of each year and month in the historical period into the historical monthly data by using the mean synthesis method;

[0089] The vegetation index calculation unit 640 is used to calculate multiple vegetation indices for each pixel of the pre-processed historical monthly data and the image to be detected;

[0090] The time series processing unit 650 is configured to perform two time series linear regressions on the vegetation index of each pixel in each historical year and month to sequentially remove noise and vegetation growth trend information. Based on the second linear regression result, the historical residual of the difference between the actual value and the regression value of the vegetation index of each pixel at each historical time point is obtained, and the predicted value of the vegetation index of each pixel in the corresponding future month is obtained.

[0091] A normal pattern construction unit 660 is configured to perform non-parametric kernel density estimation using the historical residual of the vegetation index of each pixel as sample data to construct a normal vegetation pattern for each pixel in each month, and determine the boundary of the normal vegetation pattern for each pixel in each month according to a set probability threshold;

[0092] The abnormality judgment unit 670 is used to find the normal vegetation pattern of the corresponding month according to the month to which the image to be detected belongs, and combine the predicted value of the vegetation index of each pixel in the corresponding month in the future and the boundary of the normal vegetation pattern to judge whether the vegetation in the target area is abnormal pixel by pixel.

[0093] In one embodiment of the present invention, the vegetation index calculation unit 640 calculates the vegetation index for each pixel of the pre-processed historical monthly data and the image to be detected, including but not limited to: Normalized Difference Vegetation Index NDVI and Normalized Burn Ratio Index NBR; wherein,

[0094] The NDVI index amplifies the reflectivity difference between the near-infrared band and the red band, and can represent the degree of vegetation coverage. Its calculation formula is:

[0095]

[0096] The NBR index amplifies the reflectivity difference between the near-infrared band and the short-wave infrared band, and can characterize the degree of water shortage in vegetation leaves. Its calculation formula is:

[0097]

[0098] Among them 、 、 They are the reflectivity of the near infrared, red and short-wave infrared bands respectively.

[0099] In one embodiment of the present invention, the time series processing unit 650 performs two time series linear regressions on the vegetation index of each pixel in each month of each year to remove noise and vegetation growth trend information, specifically including:

[0100] For each pixel's historical monthly data, the NDVI index is used to perform the first time series linear regression. If the actual value at a certain time point is lower than the regression value, and the difference between the two is greater than the set elimination threshold, the time point is eliminated as a noise point, and the data at the corresponding time point in the NBR index is also eliminated.

[0101] A second time series linear regression was performed on the retained time points of the NDVI index and NBR index to obtain the difference between the actual value and the regression value at each time point to remove the vegetation growth trend information.

[0102] In one embodiment of the present invention, the normal pattern construction unit 660 uses the historical residual of the vegetation index of each pixel as sample data to perform non-parametric kernel density estimation to construct the vegetation normal pattern of each pixel in each month, specifically including:

[0103] definition is the sample data, each sample contains the historical residuals of N vegetation indices involved in the construction:

[0104]

[0105] Then the N-dimensional kernel density estimation expression is:

[0106]

[0107] in: is a point in the N-dimensional vegetation index space, is the kernel function bandwidth, represents the Gaussian kernel function, Represents N-dimensional space The distance from the sample point.

[0108] Among them, when the kernel density estimation is N-dimensional, the empirical formula is used To determine the kernel function bandwidth.

[0109] In one embodiment of the present invention, the normal pattern construction unit 660 determines the boundary of the vegetation normal pattern for each pixel in each month according to a set probability threshold, specifically including:

[0110] The boundary of the normal vegetation pattern for each pixel in each month is determined using the following formula:

[0111]

[0112] in: is the boundary of the normal vegetation pattern, that is, the probability density corresponding to the cross section, which can be obtained by of The quantiles are calculated, is the area above the cross section; from the above formula, we can see that the probability that this area contains at least normal , once the probability threshold is set, it can be determined , thereby determining the boundaries of the normal distribution pattern of vegetation.

[0113] In one embodiment of the present invention, the abnormality determination unit 670 is specifically configured to:

[0114] According to the month to which the image to be detected belongs, the normal vegetation pattern of the corresponding month is found. The predicted value residual of the difference between the actual value of the vegetation index of the image to be detected and the predicted value of the corresponding month is substituted into the normal vegetation pattern of the corresponding month for each pixel. The obtained probability density is compared with the probability density of the corresponding boundary. If the probability density of a pixel in the image to be detected is less than the probability density of the corresponding boundary, the vegetation in the area corresponding to the pixel is judged to be abnormal; otherwise, the vegetation in the area corresponding to the pixel is judged to be normal.

[0115] The present invention also provides an electronic device that shares the same technical concept as the aforementioned remote sensing instant detection method for vegetation anomalies. Figure 7 A block diagram of an electronic device according to an embodiment of the present invention is shown. Figure 7 At the hardware level, the electronic device of the present invention includes a processor and memory. Optionally, it also includes an interface module, a communication module, and the like. The memory may include internal memory, such as high-speed random-access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device. Of course, the electronic device may also include other hardware required for its services.

[0116] The processor, interface module, communication module, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus. Buses can be divided into address buses, data buses, control buses, etc. For ease of representation, Figure 7 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0117] Memory is used to store computer-executable instructions. The memory provides computer-executable instructions to the processor through an internal bus.

[0118] The processor executes the computer-executable instructions stored in the memory and is specifically used to implement the following operations:

[0119] Collect all available Landsat series of surface reflectance images of the target area during the study period;

[0120] The collected surface reflectance images are divided into historical period and monitoring period. The historical period data are used to construct the normal vegetation pattern for each month, and the monitoring period data are used for the real-time detection of vegetation anomalies.

[0121] The historical data and the images to be detected during the monitoring period were preprocessed separately, and all the data of each year and month in the historical period were processed into monthly data of each year through mean synthesis;

[0122] Calculate multiple vegetation indices pixel by pixel for the pre-processed historical monthly data and the images to be detected;

[0123] The vegetation index of each pixel in each year and month is subjected to two time series linear regressions to remove noise and vegetation growth trend information. The historical residual of the difference between the actual value and the regression value of the vegetation index of each pixel at each historical time point is obtained based on the result of the second linear regression, and the predicted value of the vegetation index of each pixel in the corresponding month in the future is obtained.

[0124] The historical residuals of the vegetation index of each pixel are used as sample data for non-parametric kernel density estimation to construct the vegetation normal pattern of each pixel in each month, and the boundaries of the vegetation normal pattern of each pixel in each month are determined according to the set probability threshold.

[0125] According to the month to which the image to be detected belongs, the normal vegetation pattern of the corresponding month is found. Combined with the predicted value of the vegetation index of each pixel in the corresponding month in the future and the boundary of the normal vegetation pattern, it is judged pixel by pixel whether the vegetation in the target area is abnormal.

[0126] The present invention Figure 6 The functions performed by the remote sensing instant detection device for vegetation anomalies disclosed in the illustrated embodiments can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, the steps of the aforementioned method can be performed by hardware integrated logic circuits within the processor or by software instructions. The aforementioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., for implementing or executing the methods, steps, and flowcharts disclosed in the various embodiments of the present invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the present invention can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules within the decoding processor. The software module can be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes each step of the method of the present invention.

[0127] The electronic device may also perform Figure 1 The steps of the remote sensing instant detection method for vegetation anomaly are implemented and realized Figure 2 The functions of the process framework diagram for real-time remote sensing detection of vegetation anomalies are not described in detail in the present invention.

[0128] The present invention also proposes a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by a processor, the various embodiments of the aforementioned remote sensing instant detection method for vegetation anomalies are implemented. The present invention will not go into details here.

[0129] Here are two application examples: Example 1

[0130] Example 1 is located in the Amazon forest of Colombia, with the approximate center latitude and longitude 1°56′9″N, 73°28′34″W. The main vegetation type is tropical rainforest, and the main abnormal type is deforestation. Figure 8 This figure shows Landsat image data of the Colombian Amazon rainforest according to Example 1 of the present invention. The historical period is 1985 to 2017, and the monitoring period is 2018. "LT04," "LT05," "LT07," and "LT08" represent Landsat 4, 5, 7, and 8, respectively. In this example, real-time detection experiments were conducted on dates with low cloud cover in 2018: January 29, 2018, February 22, 2018, March 26, 2018, July 16, 2018, and December 31, 2018.

[0131] Figure 9 The shortwave infrared 2-band graph and the corresponding real-time detection result graph for each date to be detected in Example 1 of the present invention are shown. Figure 9 (a-e) are the shortwave infrared 2-band images for each pending detection date. Figure 9 (f - j) are the real-time detection results for each date. There are two possible cases for background values in the result maps: one is that the pixel is a cloud pixel that has been removed and no longer participates in anomaly detection and discrimination; the other is that the historical time series data at this pixel is insufficient to perform two linear regressions and subsequent kernel density estimation, so it is also not involved in anomaly detection and discrimination.

[0132] The results of the real-time detection are generally consistent with those presented by the shortwave infrared band 2 image. Specifically, from January 29th to March 26th, the clearcut area expanded. By July 16th, shallow vegetation had appeared in some of the clearcut areas within six months due to secondary succession, which is no longer considered an anomaly according to the present method. On December 31st, a small area of clearcut expansion occurred within the original clearcut area, and shallow vegetation growth appeared in almost all of the previously clearcut areas.

[0133] Example 2

[0134] Example 2 is located in Daxinganling, Genhe Town, Inner Mongolia, with the approximate center latitude and longitude 51°15′33″N, 121°50′47″E. It mainly refers to the mixed forest types of Daxinganling larch, white birch and larch, etc. A forest fire occurred on May 5, 2003. Figure 10 The Landsat image data of the Greater Khingan Range in Genhe Town, Inner Mongolia according to Example 2 of the present invention are shown, wherein the historical period is from 1986 to 2002 and the monitoring period is 2003. The selected real-time detection dates are 20030126, 20030526, 20030611, and 20030814.

[0135] Figure 11 The diagram shows the near infrared bands of each date to be detected in Example 2 of the present invention and the corresponding instant detection result diagram, wherein Figure 11 (a-d) are the near-infrared band maps for each date of immediate detection. Figure 11 (e-h) are the real-time detection results corresponding to each date. The background value is the same as that described in the deforestation results of Example 1 above.

[0136] The results show that the immediate detection results are generally consistent with those presented by the near-infrared band. Specifically, on January 26, 2003, some areas were unable to participate in anomaly detection due to insufficient data, while the remaining areas were all judged to be normal, which is consistent with the actual situation. May 26 was the first Landsat image acquired after the fire. By comparing it with the near-infrared band image, although there were some missed detections and false detections, the method was able to detect the vast majority of fire anomalies. From June 11 to August 14, secondary succession appeared in some burn areas and increased over time, which was no longer detected as an anomaly by the method of this invention.

[0137] The above two application examples show that the method of the present invention has a certain universality for different types of anomalies in different regions, and can realize the real-time detection of vegetation anomalies when there are available images.

Claims

1. A remote sensing instant detection method for vegetation anomalies, characterized in that: include: Collect all available Landsat series of surface reflectance images of the target area during the study period; The collected surface reflectance images are divided into historical period and monitoring period. The historical period data are used to construct the normal vegetation pattern for each month, and the monitoring period data are used for the real-time detection of vegetation anomalies. The historical data and the images to be detected during the monitoring period were preprocessed separately, and all the data of each year and month in the historical period were processed into monthly data of each year through mean synthesis; Calculate multiple vegetation indices pixel by pixel for the pre-processed historical monthly data and the images to be detected; The vegetation index of each pixel in each year and month is subjected to two time series linear regressions to remove noise and vegetation growth trend information. The historical residual of the difference between the actual value and the regression value of the vegetation index of each pixel at each historical time point is obtained based on the result of the second linear regression, and the predicted value of the vegetation index of each pixel in the corresponding month in the future is obtained. The historical residuals of the vegetation index of each pixel are used as sample data for non-parametric kernel density estimation to construct the vegetation normal pattern of each pixel in each month, and the boundaries of the vegetation normal pattern of each pixel in each month are determined according to the set probability threshold. According to the month to which the image to be detected belongs, the normal vegetation pattern of the corresponding month is found. Combined with the predicted value of the vegetation index of each pixel in the corresponding month in the future and the boundary of the normal vegetation pattern, it is judged pixel by pixel whether the vegetation in the target area is abnormal.

2. The method according to claim 1, characterized in that The vegetation indices calculated for each pixel of the pre-processed historical monthly data and the image to be detected include but are not limited to: Normalized Difference Vegetation Index NDVI and Normalized Burn Ratio Index NBR; The NDVI index amplifies the reflectivity difference between the near-infrared band and the red band, and can represent the degree of vegetation coverage. Its calculation formula is: The NBR index amplifies the reflectivity difference between the near-infrared band and the short-wave infrared band, and can characterize the degree of water shortage in vegetation leaves. Its calculation formula is: Among them, R NIR 、R red 、R SWIR2 They are the reflectivity of the near infrared, red and short-wave infrared bands respectively.

3. The method according to claim 2, characterized in that The method of removing noise and vegetation growth trend information from the vegetation index of each pixel's historical monthly data by two time series linear regressions includes: For each pixel's historical monthly data, the first time series linear regression is performed using the NDVI index month by month. If the actual value at a certain time point is lower than the regression value, and the difference between the two is greater than the set elimination threshold, the time point is eliminated as a noise point, and the data at the corresponding time point in the NBR index is also eliminated. A second time series linear regression was performed on the retained time points of the NDVI index and NBR index to obtain the difference between the actual value and the regression value at each time point to remove the vegetation growth trend information.

4. The method according to claim 1, wherein The non-parametric kernel density estimation is performed using the historical residual of the vegetation index of each pixel as sample data to construct the normal vegetation pattern of each pixel in each month, including: Define X = (X1, X2, ..., X n ] T is the sample data, each sample contains the historical residuals of N vegetation indices involved in the construction: X i =(Index1 i ,Index2 i ,......,IndexN i ),i=1,......,n Then the N-dimensional kernel density estimation expression is: Where: x is a point in the N-dimensional vegetation index space, h is the kernel function bandwidth, K() represents the Gaussian kernel function, ||xX i || represents the distance between x and the sample point in N-dimensional space.

5. The method according to claim 4, characterized in that When the kernel density estimation is N-dimensional, the empirical formula h = 1 / (N+3)n is used. (-1 / (N+4)) To determine the kernel function bandwidth.

6. The method according to claim 4, characterized in that Determining the boundary of the normal vegetation pattern for each pixel in each month according to the set probability threshold includes: The boundary of the normal vegetation pattern for each pixel in each month is determined using the following formula: in: is the boundary of the normal vegetation pattern, that is, the probability density corresponding to the cross section, which can be obtained by of The quantiles are calculated, is the area contained above the cross section; From the above formula, we can see that the probability that the region contains at least normal Once the probability threshold is set, it can be determined This determines the boundaries of the normal distribution pattern of vegetation.

7. The method according to claim 1, characterized in that The method of finding the normal vegetation pattern for the corresponding month according to the month to which the image to be detected belongs, combining the predicted value of the vegetation index of each pixel in the corresponding month in the future and the boundary of the normal vegetation pattern, and determining whether the vegetation in the target area is abnormal pixel by pixel, includes: According to the month to which the image to be detected belongs, the normal vegetation pattern of the corresponding month is found. The predicted value residual of the difference between the actual value of the vegetation index of the image to be detected and the predicted value of the corresponding month is substituted into the normal vegetation pattern of the corresponding month for each pixel. The obtained probability density is compared with the probability density of the corresponding boundary. If the probability density of a pixel in the image to be detected is less than the probability density of the corresponding boundary, the vegetation in the area corresponding to the pixel is judged to be abnormal; otherwise, the vegetation in the area corresponding to the pixel is judged to be normal.

8. A remote sensing instant detection device for vegetation anomalies, characterized in that: include: An image collection unit is used to collect all available Landsat series surface reflectance images of the target area during the study period; The period division unit is used to divide the collected surface reflectance images into historical periods and monitoring periods. The historical period data is used to construct the normal vegetation pattern for each month, and the monitoring period data is used for the real-time detection of vegetation anomalies. The preprocessing unit is used to preprocess the historical period data and the images to be detected in the monitoring period respectively, and process all the data of each year and month in the historical period into the historical year-by-month data by means of mean synthesis; The vegetation index calculation unit is used to calculate multiple vegetation indices for each pixel of the pre-processed historical monthly data and the image to be detected; The time series processing unit is used to use two time series linear regressions on the vegetation index of each pixel in each historical year and month to remove noise and vegetation growth trend information in turn, and obtain the historical residual of the difference between the actual value and the regression value of the vegetation index of each pixel at each historical time point based on the second linear regression result, and obtain the predicted value of the vegetation index of each pixel in the corresponding month in the future; A normal pattern construction unit is used to perform non-parametric kernel density estimation on the historical residuals of the vegetation index of each pixel as sample data to construct the vegetation normal pattern of each pixel in each month, and determine the boundary of the vegetation normal pattern of each pixel in each month according to a set probability threshold; The anomaly discrimination unit is used to find the normal vegetation pattern of the corresponding month according to the month to which the image to be detected belongs, and to judge whether the vegetation in the target area is abnormal pixel by pixel based on the predicted value of the vegetation index of each pixel in the corresponding month in the future and the boundary of the normal vegetation pattern.

9. An electronic device comprising: A processor and a memory storing computer executable instructions, wherein the executable instructions, when executed by the processor, implement the remote sensing instant detection method for vegetation anomalies according to any one of claims 1 to 7.

10. A computer-readable storage medium storing one or more programs, wherein when executed by a processor, the one or more programs implement the method for real-time remote sensing detection of vegetation anomalies according to any one of claims 1 to 7.

Images