Remote sensing intelligent interpretation sample data set construction method coupled with geoscience knowledge

Abstract

The invention discloses a remote sensing intelligent interpretation sample data set construction method coupled with geoscience knowledge, and the method comprises the steps: constructing an interpretation sample partition index system through employing an analytic hierarchy process based on introduced factors; performing overlay analysis on the one-way partition spatial data of each influence factor index, and determining a minimum partition unit of the comprehensive sample; performing iterative clustering on each index of the minimum partition unit of the comprehensive sample based on an SOFM model, adjusting and optimizing an obtained stable clustering partition result, and obtaining a final comprehensive sample partition; based on a sample collection proportion, uniformly selecting sample collection areas in each comprehensive sample partition; and generating a remote sensing intelligent interpretation sample data set coupled with geoscience knowledge based on the remote sensing image data prepared in the sample acquisition area and all sample elements in the sample acquisition area. The method has the advantages that uniform spatial distribution of interpretation samples in different types of regions is realized, the accurate interpretation requirement of remote sensing image data is improved, and large-scale application of remote sensing intelligent interpretation and automatic information extraction is met.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image deep learning interpretation sample technology, and in particular to a method for constructing a remote sensing intelligent interpretation sample data set coupled with geological knowledge. Background Art

[0002] In recent years, deep learning technology has developed rapidly. With its ability to learn and apply massive data, it has shown significant advantages in remote sensing image information extraction. Intelligent interpretation and information extraction of satellite remote sensing based on artificial intelligence deep learning technology has also become a new direction for remote sensing development. A large number of image interpretation samples are the basis for intelligent image interpretation based on deep learning. The extraction of some large-scale ground object information relies on multi-source remote sensing information data samples to meet the needs of intelligent training and extraction. Therefore, rich and complete interpretation samples can effectively improve the accuracy and adaptability of deep learning algorithm model construction.

[0003] However, the current interpretation sample data are usually based on partial images of local areas, or use images from a single satellite or sensor to collect information about objects and make samples. The image characteristics or spatial distribution of the same object in different regions or environments are different, and are greatly affected by factors such as topography, regional differences, and seasonal differences. Therefore, the number and types of deep learning samples used are limited, and the samples are unevenly distributed. The trained models are prone to problems such as poor generalization ability, lack of universality, and low accuracy of intelligent interpretation in complex environments. The support for the automatic interpretation capability of multi-source remote sensing images is still insufficient. Summary of the invention

[0004] The purpose of the present invention is to provide a method for constructing a remote sensing intelligent interpretation sample data set coupled with geological knowledge, so as to solve the above-mentioned problems existing in the prior art.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows: A method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge includes the following steps: S1. Constructing a partition indicator system: Based on the introduced natural and social factors, the hierarchical analysis method is used to construct the interpretation sample partition index system; the one-way partition spatial data of each influencing factor index is superimposed and analyzed to determine the minimum partition unit of the comprehensive sample; S2. Determine the sample partitioning: Based on the self-organizing feature mapping model, the various indicators of the smallest partition unit of the comprehensive sample are iteratively clustered, and the obtained stable clustering partition results are adjusted and optimized to obtain the final comprehensive sample partition with good integrity; S3. Generate sample data set: Based on the sample collection ratio in each comprehensive sample partition, sample collection areas are evenly selected in each comprehensive sample partition; based on the remote sensing image data prepared in the sample collection area and all sample elements in the sample collection area, a remote sensing intelligent interpretation sample data set coupled with geological knowledge is generated.

[0006] Preferably, step S1 specifically includes the following contents: S11. Based on the characteristics of the sample elements to be collected, considering the zoning of natural geographical conditions and the intensity of human activities, natural elements and social factors are introduced; S12. Based on the introduced factors, the hierarchical analysis method is used to construct the interpretation sample partition index system; S13. Based on the determined influencing factor indicators, prepare the single partition spatial data of each indicator, perform superposition analysis on the single partition spatial data of each indicator, and obtain the minimum partition unit of the comprehensive sample.

[0007] Preferably, natural factors include topography, landform, land cover, climate, altitude, hydrology, and soil; social factors include planting structure, human activities, and road density.

[0008] Preferably, the interpretation sample partition index system includes a target layer, a criterion layer and an index layer. The criterion layer includes natural factors and social factors. The index layer determines specific influencing factor indicators in each criterion layer factor. The overall idea of ​​the interpretation sample partition index system is expressed as follows:

[0009] in, It is the fusion result of factors at different criteria levels. Its influencing factors are divided into natural factors and social factors ; Natural factors Including specific Natural factor indicators, For the natural factor indicators; social factors Including specific social factor indicators, For the social factor indicators.

[0010] Preferably, each individual indicator in the minimum partition unit of the comprehensive sample has a unique individual partition type.

[0011] Preferably, step S2 specifically includes the following contents: S21. According to the minimum partition unit of the comprehensive sample and the indicator system categories corresponding to each unit, a two-dimensional matrix is ​​constructed as input layer data, the number of sample partition categories is set as cluster categories, and the self-organizing feature mapping model is iteratively run to obtain stable cluster partition results; S22. Based on the stable cluster partitioning results, combined with other boundary partitions or the integrity of natural factors, adjust and optimize the cluster partitioning results to obtain the final comprehensive sample partition with good integrity.

[0012] Preferably, the self-organizing feature mapping model consists of two parts: an input layer and a competition layer. The dimension of the input layer vector is the same as the dimension of each node in the competition layer. The neurons in the competition layer are arranged in a two-dimensional form into a node matrix. Each neuron in the competition layer is randomly assigned a weight for unsupervised training. The best matching neuron is selected as the winning neuron through a competition mechanism, and then the weights of the winning neuron and the neurons in its neighborhood are adjusted. This process is iterated, and input data with similar features but small differences are brought close together, and those with large differences are separated, thereby achieving automatic clustering of the input data.

[0013] Preferably, the adjustment and optimization in step S22 includes removing fragmented categories and merging categories.

[0014] Preferably, step S3 specifically includes the following contents: S31. In the comprehensive sample partition, determine the sample collection ratio in each sample partition according to the preset conditions, and evenly select sample collection areas in the comprehensive sample partition according to the ratio; S32, preparing remote sensing image data in the sample collection area, and collecting all sample elements in the sample collection area based on the remote sensing image data; S33. Integrate all collected sample elements to obtain a remote sensing intelligent interpretation sample data set coupled with geological knowledge.

[0015] Preferably, the preset conditions in step S31 include sample partition area and sample distribution density.

[0016] The beneficial effects of the present invention are as follows: the present invention targets the sample demand situation of large-scale automatic interpretation capabilities, constructs a large-scale interpretation sample data set covering various different geographic spatial environments, realizes uniform spatial distribution of interpretation samples in different types of regions, can improve the requirements for accurate interpretation of remote sensing image data, meet the large-scale application of remote sensing intelligent interpretation and automatic information extraction, and assist in the research and development of remote sensing image intelligent interpretation, natural resource monitoring and supervision, and industry remote sensing applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a technical roadmap of the construction method in the embodiment of the present invention; Figure 2 It is a network topology diagram of the SOFM model in an embodiment of the present invention; Figure 3 4 is a distribution diagram of sample collection areas selected in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation methods described herein are only used to explain the present invention and are not used to limit the present invention.

[0019] like Figure 1 As shown, in this embodiment, in view of the vast territory and diverse land cover types in my country, a method for constructing a sample data set for regional remote sensing intelligent information extraction is designed by coupling natural factors such as topography, landform, land cover, climate, altitude, hydrology, soil, and social factors such as planting structure and human activities. The method mainly includes the following three parts: 1. Constructing a Zoning Index System Based on the introduced natural and social factors, the hierarchical analysis method is used to construct the interpretation sample partition index system; the one-way partition spatial data of each influencing factor index is superimposed and analyzed to determine the minimum partition unit of the comprehensive sample. Specifically, it includes the following contents: 1.1. Based on the characteristics of the sample elements to be collected, taking into account the zoning of natural geographical conditions and the intensity of human activities, factors such as topography, landform, precipitation, vegetation, etc. are introduced.

[0020] 1.2. Based on the introduced factors, the hierarchical analysis method is used to construct the interpretation sample partition index system. The interpretation sample partition index system includes the target layer, the criterion layer and the indicator layer. The criterion layer includes natural factors and social factors. The indicator layer determines the specific influencing factor indicators in each criterion layer factor. The overall idea of ​​the interpretation sample partition index system can be expressed as:

[0021] in, It is the fusion result of factors at different criteria levels. Its influencing factors are divided into natural factors and social factors ; Natural factors Including specific natural factor indicators, such as climate indicators, topographic indicators, etc. For the natural factor indicators; social factors Including specific social factor indicators, such as human activity factors, road density factors, etc. For the social factor indicators.

[0022] Taking the collection of cultivated land samples in Province A as an example, considering the natural geographical influencing factors of cultivated land element distribution, a single natural factor is adopted in the criterion layer, and climate, precipitation, landform and vegetation are introduced as influencing indicators. The hierarchical analysis method is used to construct the interpretation sample zoning indicator system. The specific indicator system is shown in Table 1.

[0023] Table 1 Cultivated land sample zoning indicator system

[0024] 1.3. Based on the determined influencing factor indicators, prepare the individual partition spatial data of climate zones, precipitation, landform types, and vegetation types, and superimpose and analyze the individual partition spatial data to obtain the minimum unit of the comprehensive sample partition, that is, each individual indicator in the minimum unit of the sample partition has a unique individual partition type.

[0025] 2. Determine the sample area division Based on the self-organizing feature mapping model, the various indicators of the smallest partition unit of the comprehensive sample are iteratively clustered, and the obtained stable clustering partition results are adjusted and optimized to obtain the final comprehensive sample partition with good integrity. Specifically, it includes the following contents: 2.1. A two-dimensional matrix was constructed as the input layer data based on the minimum partition unit of the comprehensive sample and the climate, precipitation, landform, and climate indicator categories corresponding to each unit. The number of partition categories was set to 5, and the self-organizing feature mapping model (SOFM) was iterated 1000 times to finally obtain the stable clustering partition results.

[0026] The Self-Organizing Feature Map (SOFM) model is an unsupervised, competitive learning data clustering algorithm that can map high-dimensional data sets to low-dimensional space and obtain similar relationships between data. Figure 2 As shown in the figure, the SOFM network consists of two parts: the input layer and the competition layer. The dimension of the input layer vector is the same as the dimension of each node in the competition layer. The neurons in the competition layer are arranged in a two-dimensional form into a node matrix. Each neuron in the competition layer is randomly assigned a weight for unsupervised training. The best matching neuron is selected as the winning neuron through the competition mechanism, and then the weights of the winning neuron and the neurons in its neighborhood are adjusted. This process is iterated, and input data with similar features are closer, and data with large differences are separated, thereby realizing automatic clustering of input data.

[0027] 2.2. Based on the stable cluster partitioning results, combined with other boundary partitions or the integrity of natural factors, adjust and optimize the cluster partitioning results to obtain the final comprehensive sample partition with good integrity.

[0028] According to the clustering results of the model, the clustering results were optimized in combination with the overall terrain distribution trend and cultivated land distribution of Province A, and the clustering types of some minimum partition units were adjusted to form the final five comprehensive sample partitions of Ⅰ, Ⅱ, Ⅲ, Ⅳ, and Ⅴ.

[0029] 3. Generate sample data set: Based on the sample collection ratio in each comprehensive sample partition, sample collection areas are evenly selected in each comprehensive sample partition; based on the remote sensing image data prepared in the sample collection area and all sample elements in the sample collection area, a remote sensing intelligent interpretation sample data set coupled with geological knowledge is generated.

[0030] 3.1. In the comprehensive sample partitioning, the sample collection ratio in each sample partition is determined according to the preset conditions (such as sample partition area, sample partition density and other requirements), and the sample collection area is evenly selected in the comprehensive sample partition according to the ratio.

[0031] Among the five cultivated land sample zones in Province A, zones I and III have better natural conditions and more cultivated land; zone II is mostly mountainous and cultivated land is distributed in low-altitude areas in mountainous areas; zone IV is relatively large in area, but has poor natural conditions and cultivated land is mostly distributed in a small area around residential areas; zone V is mainly plateau mountainous and has less cultivated land. Taking all factors into consideration, sample collection areas are evenly selected within the sample zones in a ratio of 1:1:2:3:1. Figure 3 It is a sample collection area of ​​50,000 plots selected according to the collection ratio of each cultivated land sample district in Province A.

[0032] 3.2. Prepare remote sensing image data in the sample collection area, and collect all sample elements in the sample collection area based on the remote sensing image data.

[0033] 3.3. Integrate all collected sample elements to obtain a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge.

[0034] By adopting the above technical solution disclosed in the present invention, the following beneficial effects are obtained: The present invention provides a method for constructing a remote sensing interpretation sample data set coupled with geological knowledge. According to the sample demand situation of large-scale automatic interpretation capability, a large-scale interpretation sample data set covering various different geographic spatial environments is constructed to achieve uniform spatial distribution of interpretation samples in different types of areas. It can improve the requirements for accurate interpretation of remote sensing image data, meet the large-scale application of remote sensing intelligent interpretation and automatic information extraction, and facilitate the research and development of remote sensing image intelligent interpretation, natural resource monitoring and supervision, and industry remote sensing applications.

[0035] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be considered as the scope of protection of the present invention.

Claims

1. A method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge, characterized by: The following steps are included: S1. Constructing a partition indicator system: Based on the introduced natural and social factors, the hierarchical analysis method is used to construct the interpretation sample partition index system; the one-way partition spatial data of each influencing factor index is superimposed and analyzed to determine the minimum partition unit of the comprehensive sample; S2. Determine the sample partitioning: Based on the self-organizing feature mapping model, the various indicators of the smallest partition unit of the comprehensive sample are iteratively clustered, and the obtained stable clustering partition results are adjusted and optimized to obtain the final comprehensive sample partition with good integrity; S3. Generate sample data set: Based on the sample collection ratio in each comprehensive sample partition, sample collection areas are evenly selected in each comprehensive sample partition; Based on the remote sensing image data prepared in the sample collection area and all sample elements in the sample collection area, a remote sensing intelligent interpretation sample data set coupled with geological knowledge is generated.

2. The method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge according to claim 1 is characterized by: Step S1 specifically includes the following contents: S11. Based on the characteristics of the sample elements to be collected, considering the zoning of natural geographical conditions and the intensity of human activities, natural elements and social factors are introduced; S12. Based on the introduced factors, the hierarchical analysis method is used to construct the interpretation sample partition index system; S13. Based on the determined influencing factor indicators, prepare the single partition spatial data of each indicator, perform superposition analysis on the single partition spatial data of each indicator, and obtain the minimum partition unit of the comprehensive sample.

3. The method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge according to claim 2 is characterized by: Natural factors include topography, landform, surface cover, climate, altitude, hydrology, and soil; social factors include planting structure, human activities, and road density.

4. The method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge according to claim 2 is characterized by: The interpretation sample partition index system includes a target layer, a criterion layer and an index layer. The criterion layer includes natural factors and social factors. The index layer determines specific influencing factor indicators in each criterion layer factor. The overall idea of ​​the interpretation sample partition index system is expressed as follows: in, It is the fusion result of factors at different criteria levels. Its influencing factors are divided into natural factors and social factors ; Natural factors Including specific Natural factor indicators, For the natural factor indicators; social factors Including specific social factor indicators, For the social factor indicators.

5. The method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge according to claim 2, characterized in that: Each individual indicator in the minimum partition unit of the comprehensive sample has a unique individual partition type.

6. The method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge according to any one of claims 1 to 5, characterized in that: Step S2 specifically includes the following contents: S21. According to the minimum partition unit of the comprehensive sample and the indicator system categories corresponding to each unit, a two-dimensional matrix is ​​constructed as input layer data, the number of sample partition categories is set as cluster categories, and the self-organizing feature mapping model is iteratively run to obtain stable cluster partition results; S22. Based on the stable cluster partitioning results, combined with other boundary partitions or the integrity of natural factors, adjust and optimize the cluster partitioning results to obtain the final comprehensive sample partition with good integrity.

7. The method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge according to claim 6 is characterized by: The self-organizing feature mapping model consists of two parts: an input layer and a competition layer. The dimension of the input layer vector is the same as the dimension of each node in the competition layer. The neurons in the competition layer are arranged in a two-dimensional form into a node matrix. Each neuron in the competition layer is randomly assigned a weight for unsupervised training. The best matching neuron is selected as the winning neuron through a competition mechanism, and then the weights of the winning neuron and the neurons in its neighborhood are adjusted. This process is iterated, and input data with similar features and small differences are close to each other, and those with large differences are separated, thereby realizing automatic clustering of the input data.

8. The method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge according to claim 6, characterized in that: The adjustment and optimization in step S22 includes removing fragmented categories and merging categories.

9. The method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge according to claim 7 or 8, characterized in that: Step S3 specifically includes the following contents: S31. In the comprehensive sample partition, determine the sample collection ratio in each sample partition according to the preset conditions, and evenly select sample collection areas in the comprehensive sample partition according to the ratio; S32, preparing remote sensing image data in the sample collection area, and collecting all sample elements in the sample collection area based on the remote sensing image data; S33. Integrate all collected sample elements to obtain a remote sensing intelligent interpretation sample data set coupled with geological knowledge.

10. The method for constructing a remote sensing intelligent interpretation sample data set coupled with geoscientific knowledge according to claim 9, characterized in that: The preset conditions in step S31 include sample partition area and sample distribution density.

Images