Optimal Index Calculation Method and Device for Remote Sensing Temporal Data Synthesis Interval

Abstract

The present invention discloses a method and device for calculating the optimal index of the synthesis interval of remote sensing time series data, belonging to the technical field of remote sensing data processing. The method includes calculating the proportion of valid observations PVO, where the PVO represents the proportion of valid observations in the synthesized time series to all observations under a given synthesis interval; calculating the information loss ILC, where the ILC represents the ratio of the number of valid observations in the synthesized time series to the number of valid observations in the original time series; calculating the optimal synthesis interval index OCII, where the OCII is represented by the difference between the normalized forms of the PVO and the ILC; and selecting the optimal synthesis interval according to the value of the OCII, where the optimal synthesis interval is the synthesis interval when the OCII reaches the maximum value. The present invention can well quantitatively consider the balance between the positive effect and the negative effect of the increase in the synthesis interval, so as to provide the best synthesis interval suggestion and achieve a better effect of time series synthesis.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing data processing, and particularly relates to a method and device for calculating the optimal index of the synthesis interval of remote sensing time series data. Background Art

[0002] Remote sensing time series data plays an important role in monitoring surface dynamic changes. However, due to cloud contamination and other interference factors, the actually acquired time series data often has problems of discontinuity and inconsistent time intervals. In the prior art, time synthesis methods are widely used to generate standard multi-day composite time series products, such as MODIS 16-day normalized difference vegetation index (NDVI) products and Copernicus Global Land Service (CGLS) leaf area index (LAI) products. The process of remote sensing time series synthesis is to gradually move forward through a fixed window, and select the cloud-free and best-quality observation value within each time interval as the composite data. However, existing methods usually determine the synthesis interval based on empirical knowledge and lack quantitative indicators to optimize the selection of the synthesis interval.

[0003] In the related technologies of existing time series synthesis, the most common one is the synthesis method, that is, when there are multiple cloud-free original observations, the method of selecting all observation combinations of each pixel within each synthesis interval as the best observation value as the composite data. Methods such as Maximum Value Composites (MVC), Best Available Pixel algorithm, synthesis based on the maximum ratio of near-infrared to blue bands (MAX-RNB), and moving median synthesis are used to select the best observation value, but determining the synthesis interval is still a major challenge. For example, MVC is a commonly used method in remote sensing time series synthesis, and its core idea is to select the maximum value (such as NDVI or reflectance) of each pixel within a certain time window as the composite value. This method assumes that the maximum value can best represent the best state of surface vegetation and can usually effectively reduce the influence of noise such as clouds and shadows. Maximum value synthesis is simple and efficient, suitable for vegetation monitoring and phenological analysis, but may be sensitive to extreme values, easily affected by different interval lengths, and needs to be combined with quality control. The MAX-RNB synthesis algorithm selects the value with the largest ratio of near-infrared to blue reflectance within each synthesis window, improving the sensitivity to noise such as clouds and being less affected by data noise (clouds, cloud shadows, snow / ice).

[0004] At the same time, existing research usually uses a unified synthesis interval for time series synthesis without considering specific regions, and ignores the importance of selecting different synthesis intervals according to the characteristics of different regions, lacking quantitative indicators to determine the optimal synthesis interval, resulting in inconsistent quality of the synthesized time series data in different regions and seasons. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides a method and device for calculating the optimal index of the synthesis interval of remote sensing time series data, which realizes the quantitative determination of the optimal synthesis interval index (OCII) to maximize the proportion of valid observations (PVO) while ensuring high temporal resolution and minimize the information loss (ILC).

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A method for calculating the optimal index of the synthesis interval of remote sensing time series data, the method comprising:

[0008] Step 1, calculate the proportion of valid observations PVO, where PVO represents the proportion of valid observations in the synthesized time series to all observations under a given synthesis interval;

[0009] Step 2, calculate the information loss ILC, where ILC represents the ratio of the number of valid observations in the synthesized time series to the number of valid observations in the original time series;

[0010] Step 3, calculate the optimal synthesis interval index OCII, where OCII is represented by the difference between the normalized forms of PVO and ILC;

[0011] Step 4, select the optimal synthesis interval according to the value of the optimal synthesis interval index OCII, where the optimal synthesis interval is the synthesis interval when the optimal synthesis interval index OCII reaches the maximum value.

[0012] On the other hand, the present invention provides a device for calculating the optimal index of the synthesis interval of remote sensing time series data, comprising:

[0013] A PVO calculation module for calculating the proportion of valid observations PVO, where PVO represents the proportion of valid observations in the synthesized time series to all observations under a given synthesis interval;

[0014] An ILC calculation module for calculating the information loss ILC, where ILC represents the ratio of the number of valid observations in the synthesized time series to the number of valid observations in the original time series;

[0015] An OCII calculation module for calculating the optimal synthesis interval index OCII, where OCII is represented by the difference between the normalized forms of PVO and ILC;

[0016] An output module for selecting the optimal synthesis interval according to the value of the optimal synthesis interval index OCII, where the optimal synthesis interval is the synthesis interval when the optimal synthesis interval index OCII reaches the maximum value.

[0017] In a third aspect, the present invention provides an electronic device, including: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the aforementioned optimal index calculation method for the remote sensing time series data synthesis interval.

[0018] In a fourth aspect, the present invention provides a computer-readable storage medium, on which executable instructions are stored, and when the instructions are executed by a processor, the processor can be caused to implement the aforementioned optimal index calculation method for the remote sensing time series data synthesis interval.

[0019] The beneficial effects of the present invention are as follows:

[0020] Based on the fact that the increasing rates of PVO and ILC caused by the differences in image acquisition capabilities in different regions and seasons in the OCII provided by the present invention are different, these two parameters are used to numerically consider the differences in cloud cover and image acquisition capabilities in different regions and seasons, and can provide optimal synthesis interval suggestions for different regions. By measuring the relative changes between PVO and ILC, the balance between the positive effect (i.e., the increase in the proportion of effective observations) and the negative effect (i.e., the loss of time information) of the increase in the synthesis interval can be well quantitatively considered, so as to provide the best synthesis interval suggestion and achieve a better effect of time series synthesis. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is the schematic diagram of the optimal index calculation method for the remote sensing time series data synthesis interval of the present invention;

[0022] Figure 2 are the changes of normalized PVO, normalized ILC and OCII under different synthesis intervals;

[0023] Figure 3 are the changes of the crop classification F1-score at three test sites under different synthesis intervals;

[0024] Figure 4 are the classification maps at three test sites under different synthesis intervals, optimal intervals and large intervals. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The present invention will be further described below with reference to the drawings and embodiments.

[0026] As Figure 1As shown, the present invention proposes a method for calculating the optimal composite interval index of remote sensing time series data, which is used to determine the optimal composite interval of remote sensing time series data. By calculating the trade-off relationship between the proportion of valid observations (PVO) and the information loss of composite (ILC), the optimal composite interval (OCII) is determined. OCII is an index used to determine the composite interval of optical reflectance / vegetation index time series data (such as the most common 8 days, 10 days, 16 days, 30 days, or 60 days, etc.). It is assumed that a better composite interval should have a higher proportion of valid observations in the composite time series and a lower information loss due to the composite operation. According to this assumption, the method specifically includes:

[0027] Step 1. Calculate the proportion of valid observations (PVO): At the pixel scale, for the composite time series with a given composite interval, it is first necessary to calculate the proportion of valid observations in all observations (Proportion of Valid Observations, PVO) to represent the quality level of the composite time series. The proportion of valid observations of the i-th pixel in the composite time series is denoted as , that is, the ratio of the number of valid observations to the length of the composite time series:

[0028]

[0029] where L is the number of composite intervals in the length of the time series under the predefined composite interval, represents the number of valid observations of the i-th pixel in the time series under the predefined composite interval. It can be imagined that when the predefined composite interval is larger, a smaller number of composite intervals L may have a higher PVO value because it is easier to search for valid observations in a larger composite window ( Figure 1 middle view). For the study area, due to the random occurrence of clouds, the data availability of each pixel is not the same. To represent most areas, the values at the pixel level are averaged to the regional level to generate the regional-level PVO value:

[0030]

[0031] where N is the total number of pixels in the study area.

[0032] Step 2: Calculate the Information Loss of Composite (ILC); A wider composite window may lead to a reduction in the number of composite intervals in the length of the composite time series, indicating a decrease in temporal resolution and potential information loss (Information Loss of Composite, ILC). This information loss caused during the time series composition process is quantified by comparing the number of valid observations in the time series under predefined composite intervals with the number of valid observations in the original time series( Figure 1 End view). Denote the composite information loss of the i-th pixel as:

[0033]

[0034] where represents the number of valid observations in the original time series. Similarly, the values of all pixels in the study area are averaged to generate the regional-level ILC value.

[0035] Step 3: Calculate the Optimal Composite Interval Index (OCII); For a better composite interval, a higher proportion of valid observations (i.e., a larger PVO value) and less information loss (a smaller ILC value) are expected. However, these two aspects are competing and it is impossible to achieve the optimal result simultaneously. To consider the trade-off between PVO and ILC, within the regional scale, the optimal composite interval index OCII is calculated through the median composite algorithm:

[0036] (4)

[0037]

[0038] Here, the normalized forms of PVO and ILC are used ( and ), because the value ranges of these two components are different. Formula (5) represents the specific data normalization calculation method. PVO and ILC respectively represent the positive and negative effects of the increase in the composite interval. Therefore, as the composite interval increases, OCII first increases due to the greater influence of PVO and then decreases due to the greater influence of ILC. The highest value of OCII usually appears in the middle part of the predefined set of possible intervals.

[0039] Thus, it is ensured that all pixels in the study area have composite time series of consistent length, which is beneficial for subsequent standardized post-processing and analysis of remote sensing composite time series (such as classification, fitting, filtering, change detection, etc.). In practical applications, it overcomes the errors caused by spatial differences.

[0040] Preferably, other composite algorithms such as Maximum Value Composite (MVC) or Best Available Pixel algorithm can be used to replace the median composite algorithm.

[0041] Step 4. Select the optimal synthesis interval according to the value of OCII. The optimal synthesis interval is the synthesis interval when OCII reaches the maximum value. When subsequent remote sensing time series applications are carried out again according to the optimal synthesis interval index, better effects can be obtained. Taking the synthesis of remote sensing time series data for crop classification applications as an example, higher accuracy can be achieved. Due to the differences in image acquisition capabilities in different regions and seasons, the increasing rates of PVO and ILC are different, and the synthesis intervals corresponding to the top of the difference sequence (OCII) between the two are different.

[0042] For the remaining application scenarios, when using OCII to determine the optimal synthesis interval and more attention needs to be paid to PVO or ILC, the weight of ILC can be considered when calculating OCII or the weight of PVO , as follows:

[0043] (6)

[0044] For example, in applications such as land cover change detection or surface disturbance monitoring, a denser time series may be required, that is, larger; images with almost no clouds during the growing season of specific crops may be crucial for crop classification, that is, larger. The determination of the weight depends on the specific situation, such as the time scale of the disturbance or the crop calendar.

[0045] Examples

[0046] Since the dynamic changes of the optical signals of crops are relatively complex, it is more sensitive to the synthesis interval. The effectiveness of OCII was tested at three research sites mainly dominated by farmland. The main crops at the first site are corn and spring wheat, the second site mainly grows corn and soybeans, and the main spring crops at the third site are winter wheat and rapeseed. The actual data availability at different sites is different, representing the original data availability (number of cloud-free observations) from high to low; the dynamic complexity of the optical signals at each site is different, and is used in the example to illustrate the applicability of OCII under different sensitivity conditions.

[0047] PVO and ILC constitute two components of OCII. Figure 2Shows the change patterns of PVO, ILC, and OCII with increasing synthesis intervals at three test sites. At the first site, the PVO value rapidly reached a high value of 0.8 at a 16-day synthesis interval and then increased only slightly; however, the ILC value showed a different change pattern, continuously rising with the increasing synthesis interval until the 90-day interval. Therefore, for the Yili site with a relatively high proportion of sunny days, the peak of OCII occurred at a 16-day synthesis interval. At the second site, the PVO and ILC values increased relatively uniformly before the 25-day interval. After this point, the increase in ILC exceeded that of PVO, indicating that the peak of OCII occurred at the 25-day interval. At the third site, due to the limited availability of valid data, the PVO value continued to increase, while after the 30-day interval, the increase rate of ILC exceeded that of PVO. Therefore, the OCII value reached the highest at 30 days. For the synthesis of Sentinel-2 time series data, OCII provides a simple and operable criterion, suggesting a smaller synthesis interval (16 days) for the first site, while larger synthesis intervals (25 days and 30 days) for the second and third sites respectively.

[0048] For crop classification of synthetic time series data based on different intervals, a random forest (RF) model was used to conduct crop classification experiments based on synthetic time series. Considering the comparison of different synthesis intervals (i.e., 8 days, 10 days, 16 days, 20 days, 25 days, 30 days, 60 days, and 90 days), eight random forest classification scenarios were conducted based on the same crop samples at each test site. To avoid overfitting of the RF model, a sub-region to whole classification method was adopted, where the training samples were not randomly selected from the whole region but from a specific sub-region. Then, the trained classifier was applied to the whole region to obtain the crop map. Accuracy verification was conducted in the whole region. Since three sub-regions were selected at each site, a classification result could be obtained using each sub-region, and the final classification accuracy was calculated as the average of the F1-score of the accuracy metrics of the three classification results.

[0049] Figure 3 Shows the classification accuracy (F1-score) of the main crops at three test sites under different synthesis intervals. Figure 4The classification maps of three test sites are shown at small intervals (8 days), optimal intervals, and large intervals (90 days). The median composite algorithm and the MAX-RNB algorithm are used to generate composite values within each interval. When using the median composite algorithm, at the first site, the classification accuracy of the main crops (spring wheat and corn) is relatively high at the 16-day composite interval recommended by OCII, while it decreases significantly when the composite interval is greater than 20 - 25 days. At the second site, when using the 25-day composite interval recommended by OCII, the F1-score values of both corn and soybeans reach the highest. At the third site, when using the 30-day composite interval recommended by OCII, the classification accuracy of winter wheat and rapeseed is also relatively good. Similar results are obtained using the MAX-RNB composite algorithm ( Figure 3 right column). When the second and third sites use smaller composite intervals (e.g., 8 days) or larger composite intervals (>60 days), the classification accuracy is lower than that under the composite intervals recommended by OCII. Regarding the classification maps, salt noise appears in the classification map generated using 8-day composite images at the second site, while obvious omission errors exist in the classification map when using a 90-day composite interval. At the third site, when using a smaller composite interval (8 days) or a larger interval (90 days), rapeseed is misclassified as winter wheat. The classification map recommended by OCII is more consistent with the reference map ( Figure 4 ). Thus, it can be seen that OCII effectively balances the relationship between the proportion of effective observations and the loss of potential information, generating a synthetic time series suitable for practical applications.

[0050] On the other hand, the present invention provides a device for calculating the optimal index of the synthetic interval of remote sensing time series data. Each module included therein can implement each step of the foregoing method. Specifically, it includes:

[0051] A PVO calculation module for calculating the proportion of valid observations PVO, where the PVO represents the proportion of valid observations in the synthetic time series to all observations at a given synthetic interval;

[0052] An ILC calculation module for calculating the information loss ILC, where the ILC represents the ratio of the number of valid observations in the synthetic time series to the number of valid observations in the original time series;

[0053] An OCII calculation module for calculating the optimal synthetic interval index OCII, where the OCII is represented by the difference between the normalized forms of PVO and ILC;

[0054] An output module for selecting the optimal synthetic interval according to the value of the optimal synthetic interval index OCII, where the optimal synthetic interval is the synthetic interval when the optimal synthetic interval index OCII reaches the maximum value.

[0055] In a third aspect, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the optimal index calculation method for remote sensing time series data synthesis intervals described above.

[0056] In a fourth aspect, the present invention provides a computer-readable storage medium, having stored thereon executable instructions that, when executed by a processor, enable the processor to implement the optimal index calculation method for remote sensing time series data synthesis intervals described above.

[0057] The specific embodiments described above further elaborate on the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for calculating the optimal index of synthesis interval of remote sensing time series data, characterized in that: The method comprises: Step 1, calculate the effective observation ratio PVO, where PVO represents the ratio of effective observations to all observations in the synthetic time series under a given synthetic interval; Step 2: Calculate the information loss ILC, where ILC represents the ratio of the number of valid observations in the synthetic time series to the number of valid observations in the original time series. Step 3, calculating the optimal composite interval index OCII, wherein the OCII is represented by the difference between the normalized forms of PVO and ILC; Step 4: Select an optimal synthesis interval according to the value of the optimal synthesis interval index OCII, where the optimal synthesis interval is the synthesis interval when the optimal synthesis interval index OCII reaches a maximum value.

2. The method for calculating the optimal index of the synthesis interval of remote sensing time series data according to claim 1, characterized in that: The step 1 includes: the ratio of valid observation values ​​of the i-th pixel in the synthetic time series is recorded as : where L is the number of synthetic intervals in the length of the time series at a given synthetic interval, represents the number of valid observations of the i-th pixel in the time series at a given synthesis interval; The pixel level The values ​​are averaged at the regional level to generate regional PVO values: Where N is the total number of pixels in the study area.

3. The method for calculating the optimal index of the synthesis interval of remote sensing time series data according to claim 2, characterized in that: The step 2 includes the synthesis information loss of the i-th pixel It is expressed as: in, Represents the number of valid observations in the original time series; The pixel level The values ​​are averaged at the regional level to generate regional-level ILC values.

4. The method for calculating the optimal index of the synthesis interval of remote sensing time series data according to claim 1, characterized in that: The step 3 includes calculating the optimal composite interval index OCII by a median composite algorithm: （4） In the formula, They represent the normalized forms of the proportion of valid observations PVO and the information loss ILC respectively.

5. The method for calculating the optimal index of the synthesis interval of remote sensing time series data according to claim 1, characterized in that: The step 4 also includes assigning different weights to the normalized forms of PVO and ILC for calculation.

6. The method for calculating the optimal index of the synthesis interval of remote sensing time series data according to claim 1, characterized in that: The step 4 also includes that due to the differences in image acquisition capabilities in different regions and seasons, the PVO and ILC increase rates are different, and the synthesis intervals corresponding to the OCII are different.

7. The method for calculating the optimal index of the synthesis interval of remote sensing time series data according to claim 4, characterized in that: In step 3, the maximum value synthesis algorithm or the best available pixel algorithm may be used instead of the median synthesis algorithm.

8. A device for calculating the optimal index of synthesis interval of remote sensing time series data, characterized in that: include: A PVO calculation module is used to calculate the effective observation ratio PVO, where the PVO represents the ratio of effective observations to all observations in a synthetic time series at a given synthetic interval; An ILC calculation module, used for calculating the information loss ILC, wherein the ILC represents the ratio of the number of valid observations in the synthetic time series to the number of valid observations in the original time series; An OCII calculation module, configured to calculate an optimal composite interval index OCII, wherein the OCII is represented by a difference between normalized forms of PVO and ILC; The output module is used to select the optimal synthesis interval according to the value of the optimal synthesis interval index OCII, wherein the optimal synthesis interval is the synthesis interval when the optimal synthesis interval index OCII reaches a maximum value.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; Wherein, when one or more programs are executed by the one or more processors, the one or more processors implement a method for calculating an optimal index of a synthesis interval of remote sensing time series data as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: Executable instructions are stored thereon, and when the instructions are executed by the processor, the processor can implement the method for calculating the optimal index of the synthesis interval of remote sensing time series data as described in any one of claims 1-7.

Images