A geometric and radiometric compensation method applied to the extraction of cultivated land

Abstract

The application belongs to the field of image data processing, and particularly relates to a geometric and radiation compensation measurement method applied to farmland extraction, which comprises the following steps: obtaining remote sensing image data of a target area, pre-processing the remote sensing image, constructing a label, cutting and cropping, and dividing a training set and a test set; under the condition of fixed spatial resolution, the application establishes a unified compensation measurement mechanism between spatial resolution and spectral band number, realizes quantitative description of farmland extraction performance, solves the problem that the existing technology cannot uniformly measure the synergistic effect of spectral radiation information and spatial geometric information, and provides method support for two-dimensional quantitative analysis of farmland extraction precision; by constructing an Iso-IoU joint regression model, mathematical expression of the compensation relationship between spectral radiation information and spatial geometric information is realized, and the farmland extraction capability under different remote sensing configuration conditions can be quantitatively evaluated.

Description

Technical Field

[0001] This invention belongs to the field of image data processing, specifically a geometric and radiometric compensation measurement method applied to farmland extraction. Background Technology

[0002] Accurate farmland extraction is an important research direction in intelligent remote sensing interpretation, with wide applications in food security monitoring, agricultural resource surveys, land use analysis, and ecological environment assessment. With the development of remote sensing satellite observation technology and the continuous enrichment of high spatial resolution and hyperspectral resolution remote sensing data, deep learning-based farmland extraction methods are gradually becoming the mainstream technical approach.

[0003] The ability to identify land features in remote sensing imagery is typically influenced by both spectral and spatial information. Spectral radiometric information primarily reflects the radiation characteristics of land features, enhancing the class separability between different features; spatial geometric information determines the ability to express land feature boundaries, texture structures, and spatial details. Therefore, both spectral radiometric and spatial geometric information directly affect the accuracy of farmland extraction. Existing research typically focuses on the individual effects of spectral radiation information or spatial geometry information on farmland extraction performance, lacking a unified method for measuring the synergistic relationship between the two. Especially in practical remote sensing applications, when spatial geometry information degrades, there is currently a lack of effective quantitative mechanisms to describe whether extraction accuracy can be maintained by enhancing spectral radiation information.

[0004] Furthermore, different remote sensing sensors exhibit significant differences in spectral radiometric configuration and spatial geometric resolution, making it difficult for existing methods to provide a unified reference for remote sensing sensor design, data configuration optimization, and remote sensing missions under resource-constrained conditions.

[0005] Therefore, the present invention provides a geometric and radiation compensation measurement method for farmland extraction. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve its technical problem is: a geometric and radiation compensation measurement method for farmland extraction, which includes the following steps: S1. Acquire remote sensing image data of the target area, preprocess the remote sensing images, construct labels and crop slices, and divide the data into training and test sets. S2. Under fixed spatial resolution, construct remote sensing image data with different spectral band combinations, and extract farmland based on a deep learning model to obtain the cross-union ratio (CUNR) accuracy index under the corresponding spectral radiance dimension. IoU ; S3. Under the preset spectral band combination conditions, remote sensing image data with different spatial resolution degradation levels are constructed through multi-scale downsampling, and farmland is extracted based on a deep learning model to obtain the cross-union ratio (CUNR) accuracy index under the corresponding spatial geometric dimensions. IoU ; S4. Based on the results obtained in steps S2 and S3, the number of spectral bands... N band and spatial resolution degradation level L Construct equivalent intersection-union ratios with the variable as the independent variable. Iso-IoU Linear regression model; S5, Utilization Iso-IoU Linear regression models generate isometric maps, enabling a quantitative description of the compensation relationship between spectral radiation information and spatial geometric information.

[0008] Furthermore, in step S1, the remote sensing image includes a multispectral remote sensing image, which includes the red band, green band, blue band and near-infrared band. The label data was constructed through manual visual interpretation combined with verification.

[0009] Furthermore, in step S2, different spectral band combinations are formed by selecting different subsets of bands; Spectral band combinations include single-band combinations, dual-band combinations, three-band combinations, and multispectral combinations; the spectral band data input to the deep learning model is linearly scaled using a uniform normalization method.

[0010] Furthermore, in step S3, the original high-resolution image is set as the baseline, corresponding to a spatial resolution degradation level of [missing information]. L=0 ; Subsequently, an interpolation algorithm is used to downsample step by step to half resolution, corresponding to the spatial resolution degradation level. L=1 and a quarter resolution, corresponding to the level of spatial resolution degradation. L=2 Meanwhile, the nearest neighbor interpolation algorithm is used to synchronously resample the real label mask corresponding to the remote sensing image.

[0011] Furthermore, in step S4, the constructed equivalent intersection-union ratio Iso-IoU The linear regression model satisfies the following relationship: IoU=c 0 +c 1 ×N band +c 2 ×L； in, IoU This indicates the ratio of cultivated land extraction to land consolidation.N band This represents the number of input spectral bands, which is a spectral radiance dimension feature. L It represents the level of degradation of the current resolution relative to the original reference spatial resolution, and is a feature of spatial geometric dimension. c 0 For the fitting constant term; c 1 This represents the spectral gain coefficient, indicating the gain corresponding to adding one more spectral band. IoU Gain; c 2 The spatial sensitivity coefficient represents the loss of accuracy caused by spatial resolution degradation.

[0012] Furthermore, the level of spatial resolution degradation L Expressed as a dimensionless relative scaling factor, the calculation formula is: L= log 2 ( s ),in This represents the uniform downsampling scaling factor.

[0013] Furthermore, based on Iso-IoU Linear regression models, at a given target accuracy level IoU target and the current level of spatial resolution degradation L Under these conditions, the number of spectral bands required for compensation to maintain target extraction accuracy. N band The calculation formula is as follows: Furthermore, based on Iso-IoU Linear regression models, at a given target accuracy level IoU target and the current level of spatial resolution degradation L Under these conditions, the number of spectral bands required for compensation to maintain target extraction accuracy. N band The calculation formula is as follows: .

[0014] Furthermore, in step S5, Iso-IoU Contour maps are drawn on the basis of L x-axis N band In a coordinate system with the vertical axis as the ordinate, the slope of the contour lines represents the rate of substitution between spectral radiation information and spatial geometric information, which is used to provide a quantitative basis for selecting sensor configuration under constraints.

[0015] The beneficial effects of this invention are as follows: 1. The geometric and radiometric compensation measurement method for farmland extraction described in this invention establishes a unified compensation measurement mechanism between spatial resolution and the number of spectral bands, realizing a quantitative description of farmland extraction performance. This solves the problem that existing technologies cannot uniformly measure the synergistic effect of spectral radiometric information and spatial geometric information, providing methodological support for the two-dimensional quantitative analysis of farmland extraction accuracy. By constructing an Iso-IoU joint regression model, a mathematical expression of the compensation relationship between spectral radiometric information and spatial geometric information is realized, enabling quantitative evaluation of farmland extraction capabilities under different remote sensing configurations.

[0016] 2. The geometric and radiometric compensation measurement method for farmland extraction described in this invention can intuitively reflect the substitution relationship between the number of different spectral bands and spatial resolution through a spatial-spectral equivalence map, providing a theoretical basis for the configuration of remote sensing sensor parameters; by calculating the number of spectral bands to be supplemented, the accuracy of target farmland extraction is maintained, thereby improving the efficiency of remote sensing data resource allocation. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart from the present invention; Figure 2 This is a comparison chart of the accuracy of farmland extraction under different combinations of spectral bands in this invention; Figure 3 This is a visual comparison of farmland extraction results under different combinations of spectral bands in this invention; Figure 4 This is a visual comparison of farmland extraction results under different spatial resolution degradation levels in this invention; Figure 5 The spatial resolution degradation level in this invention L Number of spectral bands N band Linear fitting plot of the relationship between IoU and extraction accuracy; Figure 6 It is based on the present invention Iso-IoU Contour maps of spatial-spectral compensation relationships generated by the model. Detailed Implementation

[0019] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0020] like Figure 1 to Figure 6 As shown in the figure, an embodiment of the present invention provides a geometric and radiation compensation measurement method for farmland extraction, which includes the following steps: S1. Acquire remote sensing image data of the target area, preprocess the remote sensing images, construct labels and crop slices, and divide the data into training and test sets. S2. Under fixed spatial resolution, construct remote sensing image data with different spectral band combinations, and extract farmland based on a deep learning model to obtain the cross-union ratio (CUNR) accuracy index under the corresponding spectral radiance dimension. IoU ; S3. Under the preset spectral band combination conditions, remote sensing image data with different spatial resolution degradation levels are constructed through multi-scale downsampling, and farmland is extracted based on a deep learning model to obtain the cross-union ratio (CUNR) accuracy index under the corresponding spatial geometric dimensions. IoU ; S4. Based on the results obtained in steps S2 and S3, the number of spectral bands... N band and spatial resolution degradation level L Construct equivalent intersection-union ratios with the variable as the independent variable. Iso-IoU Linear regression model; S5, Utilization Iso-IoU Linear regression models generate isometric maps, enabling a quantitative description of the compensation relationship between spectral radiation information and spatial geometric information.

[0021] In step S1, the remote sensing image includes multispectral remote sensing image, which includes red light band, green light band, blue light band and near-infrared band. The label data was constructed through manual visual interpretation combined with verification.

[0022] In step S2, different spectral band combinations are formed by selecting different subsets of bands; Spectral band combinations include single-band combinations, dual-band combinations, three-band combinations, and multispectral combinations; the spectral band data input to the deep learning model is linearly scaled using a uniform normalization method.

[0023] In step S3, the original high-resolution image is set as the baseline, corresponding to a spatial resolution degradation level of [missing information]. L=0 ; Subsequently, an interpolation algorithm is used to downsample step by step to half resolution, corresponding to the spatial resolution degradation level. L=1 and a quarter resolution, corresponding to the level of spatial resolution degradation. L=2 Meanwhile, the nearest neighbor interpolation algorithm is used to synchronously resample the real label mask corresponding to the remote sensing image.

[0024] In step S4, the equivalent intersection-union ratio is constructed. Iso-IoU The linear regression model satisfies the following relationship: IoU=c0 +c 1 ×N band +c 2 ×L ; in, IoU This indicates the ratio of cultivated land extraction to land consolidation. N band This represents the number of input spectral bands, which is a spectral radiance dimension feature. L It represents the level of degradation of the current resolution relative to the original reference spatial resolution, and is a feature of spatial geometric dimension. c 0 For the fitting constant term; c 1 This represents the spectral gain coefficient, indicating the gain corresponding to adding one more spectral band. IoU Gain; c 2 The spatial sensitivity coefficient represents the loss of accuracy caused by spatial resolution degradation.

[0025] Spatial resolution degradation level L Expressed as a dimensionless relative scaling factor, the calculation formula is: L=log 2 ( s ),in This represents the uniform downsampling scaling factor.

[0026] based on Iso-IoU Linear regression models, at a given target accuracy level IoU target and the current level of spatial resolution degradation L Under these conditions, the number of spectral bands required for compensation to maintain target extraction accuracy. N band The calculation formula is as follows: .

[0027] In step S5, Iso-IoU Contour maps are drawn on the basis of L x-axis N band In a coordinate system with the vertical axis as the ordinate, the slope of the contour lines represents the rate of substitution between spectral radiation information and spatial geometric information, which is used to provide a quantitative basis for selecting sensor configuration under constraints.

[0028] Example:

[0029] Step S1: Obtain remote sensing images of the target area and corresponding pixel-level ground truth labels to construct a benchmark dataset; Step S2: Conduct spectral radiation variation assessment, fix the spatial resolution, and systematically construct different effective band combinations to obtain extraction accuracy; Step S3: Conduct spatial geometric variation assessment, fix the band combination, and obtain the extraction accuracy under different degradation levels through multi-scale downsampling; Step S4: Using the number of spectral bands and the level of spatial resolution degradation as joint independent variables, establish the equivalent crossover ratio (CLOUD). Iso-IoU Joint regression model; Step S5, according to the above Iso-IoU The model derives the equivalent substitution rate and plots contour maps of the space-spectral compensation relationship.

[0030] The extraction accuracy acquisition process in steps S2 and S3 includes: inputting samples under different spatial-spectral configurations into a unified farmland extraction model for training and testing, and calculating the corresponding accuracy. IoU index.

[0031] Two representative heterogeneous satellite datasets were used: the Shanglin County dataset constructed from Gaofen-2 imagery and the iFLYTEK dataset constructed from Jilin-1 imagery. Gaofen-2 imagery has a spatial resolution of 0.8m, while Jilin-1 imagery has a spatial resolution of 0.75–1.1m, providing rich surface details. Both datasets include blue, green, red, and near-infrared bands and have sub-meter spatial resolution, making them suitable for constructing experimental data for precise farmland extraction.

[0032] Furthermore, the data processing involved in steps S1 and S2 above includes: using a sliding window strategy to cut the image and labels into 256×256 pixel square slices. To ensure the objectivity of the modeling parameters, a spatially independent partitioning strategy is implemented. In the two datasets, the samples in the training set are further randomly divided into a training subset and a validation subset in an 8:2 ratio.

[0033] like Figure 2 and Figure 3 As shown, the spectral radiative variation assessment in step S2 above is achieved using the controlled variable method. While maintaining the original spatial resolution, a subset of effective bands from the image is systematically extracted to construct 15 input schemes ranging from single-band to full-band.

[0034] Figure 2 This paper presents the statistical results of extraction accuracy for different spectral band combinations under two datasets and two farmland extraction methods. (a) and (b) show the quantitative evaluation results of the BsiNet model on the iFLYTEK and Shanglin County datasets, respectively; (c) and (d) show the quantitative evaluation results of the REAUnet model under two different experimental data conditions, respectively. Figure 2 It can be seen that as the number of effective spectral bands increases, IoU The overall upward trend of these indicators indicates that abundant spectral radiation information can effectively enhance the distinguishability and regional integrity of cultivated land targets.

[0035] Figure 3 The results further illustrate the visual comparison of farmland extraction under different combinations of spectral bands. (a) and (b) show the visualization results of farmland extraction using the BsiNet model on the iFLYTEK and Shanglin County datasets, respectively; (c) and (d) show the visualization results of farmland extraction using the REAUnet model under two different experimental data conditions, respectively. As the number of input bands decreases, the ability to represent farmland boundaries and the integrity of the farmland area decrease to varying degrees. This result indicates that spectral radiance information enhances the accuracy of farmland extraction.

[0036] like Figure 4 As shown in Tables 1 and 2, in step S3 above, when conducting the spatial geometric variation assessment experiment, a spatial degradation level is introduced to uniformly describe the process of spatial resolution change: L=log 2 ( s ); in, This is the downsampling scaling factor. To ensure semantic consistency during the measurement process, bicubic interpolation is used to downsample the multispectral image slices to simulate the spatial resolution degradation process. For the corresponding ground truth labels, nearest neighbor interpolation is strictly used for resampling. The resampled slices maintain a size of 256×256 pixels and are spatially aligned.

[0037] Accuracy, Precision, Recall, F1-score, and IoU were used as performance indicators for farmland extraction, and their calculation formulas are as follows: ; ； ； ； ； In the formula, This indicates the number of correctly identified farmland pixels. This indicates the number of correctly identified non-farmland pixels. This indicates the number of pixels misclassified as farmland. This indicates the number of missed farmland pixels.

[0038] Table 1 shows the statistical results of extraction accuracy of two different farmland extraction models under different spatial resolution degradation levels under the iFLYTEK dataset conditions; Table 2 shows the statistical results of extraction accuracy of two different farmland extraction methods under different spatial resolution degradation levels under the Shanglin County dataset conditions.

[0039] Table 1

[0040] Table 2

[0041] As shown in Tables 1 and 2, with increasing spatial resolution degradation, accuracy, precision, recall, F1 score, and intersection-over-union ratio all decrease. IoU Indicators such as ) are generally showing a downward trend.

[0042] Figure 4 This paper further demonstrates the visual comparison of farmland extraction results under different levels of spatial degradation. (a) and (b) show the farmland extraction visualization results of the BsiNet model on the iFLYTEK and Shanglin County datasets, respectively; (c) and (d) show the farmland extraction visualization results of the REAUnet model under two different experimental data conditions, respectively. As the level of spatial geometric degradation increases, the ability to represent the spatial geometric details of farmland gradually weakens, and the corresponding extraction accuracy index decreases accordingly, thus providing a basis for subsequent... Iso-IoU The spatial sensitivity coefficient in the model provides a quantitative basis.

[0043] like Figure 5 As shown, this application further uses the number of bands and the level of spatial degradation as joint variables for mathematical modeling, and constructs... Iso-IoU Joint regression model. Figure 5 (a) and (b) in the figure show the level of spatial resolution degradation, respectively. L With extraction accuracy IoU The linear relationship between them, and the number of bands N band With extraction accuracy IoU The results show a linear relationship between the number of bands. N band and the level of spatial resolution degradation L With extraction accuracy IoU There is a strong linear correlation between them.

[0044] Based on the obtained experimental results, the following joint metric model was established using the multiple linear regression method: IoU=c 0 +c 1 ×N band +c 2 ×L ; In this formula, IoU This indicates the accuracy of the crossover ratio (CRO) predicted by the model. N band This represents the number of valid spectral bands in the input, i.e., the radiative dimension feature; L This indicates the level of degradation of the current resolution relative to the original reference spatial resolution, i.e., geometric dimensional features; c 0 For the fitting constant term; c 1 The radiation gain coefficient characterizes the average gain that can be obtained by adding one more spectral band under the same geometric conditions. IoU Increase by percentage points; c 2 This is the spatial sensitivity coefficient, and it is negative, representing the effect of reducing spatial resolution by one level. IoU Loss percentage points.

[0045] like Figure 6 As shown, this application utilizes fitting to complete the process. Iso-IoU The joint model deduced the compensation relationship between spatial geometric information and spectral radiometric information, and generated a spatial-spectral compensation contour map. This was done at the level of spatial degradation. L The horizontal axis represents the number of spectral bands. N band With the vertical axis Figure 6 In the middle, the same contour line ( Iso-IoU (Lines) represent approximately the same final extraction accuracy. (a) and (c) show the contour maps of the BsiNet model on the iFLYTEK and Shanglin County datasets, respectively, while (c) and (d) show the contour maps of the REAUnet model under two different experimental data conditions. This map visually illustrates the compensation substitution rate: for example, when hardware limitations force the spatial resolution to be reduced from... L =1 degenerates to L When the value is 2, the number of additional spectral bands required to maintain the target extraction accuracy can be quantitatively calculated along the contour lines of the target accuracy.

[0046] Furthermore, the mathematical expression for the number of bands required to achieve the target accuracy is as follows: ; The contour maps are used to achieve quantitative equivalent conversion between different spatial-spectral configurations, providing a unified engineering measurement basis for remote sensing sensor configuration optimization, multi-source remote sensing data selection, and remote sensing tasks under resource-constrained conditions.

[0047] By applying the scheme of this application, not only is a unified mathematical compensation relationship between spatial resolution and the number of spectral bands established, but also a quantitative equivalent description of farmland extraction capacity under different spatial-spectral configurations is realized, providing a unified quantitative engineering basis for remote sensing data configuration optimization, remote sensing sensor design, and remote sensing mission deployment under resource-constrained conditions.

[0048] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A geometric and radiation compensation measurement method applied to farmland extraction, characterized in that, The method includes the following steps: S1. Acquire remote sensing image data of the target area, preprocess the remote sensing images, construct labels and crop slices, and divide the data into training and test sets. S2. Under fixed spatial resolution, construct remote sensing image data with different spectral band combinations, and extract farmland based on a deep learning model to obtain the cross-union ratio (CUNR) accuracy index under the corresponding spectral radiance dimension. IoU ; S3. Under the preset spectral band combination conditions, remote sensing image data with different spatial resolution degradation levels are constructed through multi-scale downsampling, and farmland is extracted based on a deep learning model to obtain the cross-union ratio (CUNR) accuracy index under the corresponding spatial geometric dimensions. IoU ; S4. Based on the results obtained in steps S2 and S3, the number of spectral bands... N band and spatial resolution degradation level L Construct equivalent intersection-union ratios with the variable as the independent variable. Iso-IoU Linear regression model; S5, Utilization Iso-IoU Linear regression models generate isometric maps, enabling a quantitative description of the compensation relationship between spectral radiation information and spatial geometric information.

2. The geometric and radiation compensation measurement method for farmland extraction according to claim 1, characterized in that: In step S1, the remote sensing image includes a multispectral remote sensing image, which includes red light band, green light band, blue light band and near-infrared band. The label data is constructed through manual visual interpretation combined with verification.

3. The geometric and radiation compensation measurement method for farmland extraction according to claim 1, characterized in that: In step S2, different spectral band combinations are formed by selecting different subsets of bands. The spectral band combinations include single-band combinations, dual-band combinations, three-band combinations, and multispectral combinations; the spectral band data input to the deep learning model are linearly scaled using a uniform normalization method.

4. The geometric and radiation compensation measurement method for farmland extraction according to claim 1, characterized in that: In step S3, the original high-resolution image is set as the baseline, corresponding to a spatial resolution degradation level of [missing information]. L=0 ; Subsequently, an interpolation algorithm is used to downsample step by step to half resolution, corresponding to the spatial resolution degradation level. L=1 and a quarter resolution, corresponding to the level of spatial resolution degradation. L=2 Meanwhile, the nearest neighbor interpolation algorithm is used to synchronously resample the real label mask corresponding to the remote sensing image.

5. The geometric and radiation compensation measurement method for farmland extraction according to claim 1, characterized in that: In step S4, the equivalent intersection-union ratio is constructed. Iso-IoU The linear regression model satisfies the following relationship: IoU=c 0 +c 1 ×N band +c 2 ×L ； in, IoU This indicates the ratio of cultivated land extraction to land consolidation. N band This represents the number of input spectral bands, which is a spectral radiance dimension feature. L It represents the level of degradation of the current resolution relative to the original reference spatial resolution, and is a feature of spatial geometric dimension. c 0 For fitting constant terms; c 1 This represents the spectral gain coefficient, indicating the gain corresponding to adding one more spectral band. IoU Gain; c 2 The spatial sensitivity coefficient represents the loss of accuracy caused by spatial resolution degradation.

6. The geometric and radiation compensation measurement method for farmland extraction according to claim 5, characterized in that: The level of spatial resolution degradation L Expressed as a dimensionless relative scaling factor, the calculation formula is: L=log 2 ( s ),in This represents the uniform downsampling scaling factor.

7. The geometric and radiation compensation measurement method for farmland extraction according to claim 5, characterized in that: Based on the above Iso-IoU Linear regression models, at a given target accuracy level IoU target and the current level of spatial resolution degradation L Under these conditions, the number of spectral bands required for compensation to maintain target extraction accuracy. N band The calculation formula is as follows: 。 8. The geometric and radiation compensation measurement method for farmland extraction according to claim 1, characterized in that: In step S5 Iso-IoU Contour maps are drawn on the basis of L x-axis N band In a coordinate system with the vertical axis as the ordinate, the slope of the contour lines represents the rate of substitution between spectral radiation information and spatial geometric information, which is used to provide a quantitative basis for selecting sensor configuration under constraints.

Images