Ecological environment remote sensing image classification and feature extraction method based on deep learning

Abstract

The invention provides an ecological environment remote sensing image classification and feature extraction method based on deep learning, and the method comprises the steps: obtaining the remote sensing image data of a target ecological region, and obtaining the region index data of five ecological indexes; ecological sensitivity evaluation and fusion of the five ecological indexes are carried out, and a plurality of ecological sensitivity sub-regions are divided; and according to the ecological sensitivity data of each ecological sensitivity sub-region, selecting an optimal feature extraction algorithm in combination with the importance weight coefficient to perform feature extraction so as to finally obtain a feature data set. The remote sensing image data can be analyzed and processed purposefully and emphatically, and the processing efficiency is improved on the premise that the quality requirement is met.

Description

Technical Field

[0001] The present invention belongs to the technical field of ecological product value realization, and in particular relates to an ecological environment remote sensing image classification and feature extraction method based on deep learning. Background Art

[0002] With the increasing awareness of ecological and environmental protection, the demand for accurate classification and feature extraction of ecological and environmental remote sensing images has become more urgent. In recent years, the rapid development of deep learning technology has brought new breakthroughs in the field of remote sensing image processing. With its powerful automatic feature learning ability, the deep learning model can deeply explore the potential information in remote sensing images, providing strong support for the accurate classification and feature extraction of ecological and environmental remote sensing images.

[0003] However, due to computing power limitations, some existing remote sensing image processing technologies cannot effectively analyze and process remote sensing images. Not only will it be impossible to obtain image data that meets quality requirements, but it will also restrict the accuracy and efficiency of ecological environment monitoring and assessment. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide an ecological environment remote sensing image classification and feature extraction method based on deep learning, aiming to solve the problems raised in the background technology.

[0005] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions: A method for ecological environment remote sensing image classification and feature extraction based on deep learning, the method specifically comprises the following steps: Acquire remote sensing image data of the target ecological area, and perform data processing and index analysis on the remote sensing image data based on AI technology to obtain regional index data including five ecological indicators: elevation, slope, aspect, vegetation type, and land use type; Based on the regional indicator data, ecological sensitivity evaluation and integration of five ecological indicators, namely elevation, slope, aspect, vegetation type and land use type, are carried out to divide the target ecological area into multiple ecologically sensitive sub-areas; Obtaining ecological sensitivity data corresponding to each of the ecologically sensitive sub-areas, wherein the ecological sensitivity data at least includes importance weight coefficients corresponding to elevation, slope, aspect, vegetation type, land use type, and texture features in the ecologically sensitive sub-areas; When it is determined that the importance weight coefficient is greater than the preset importance weight coefficient, the optimal feature extraction algorithm model is selected in the current item to perform feature extraction to obtain the corresponding optimal sub-features, and the optimal sub-features corresponding to each item are integrated to generate a complete feature data set.

[0006] As a further limitation of the technical solution of the embodiment of the present invention, the acquisition of remote sensing image data of the target ecological area, based on AI technology, data processing and index analysis of the remote sensing image data, and acquisition of regional index data including five ecological indicators including elevation, slope, slope aspect, vegetation type and land use type specifically include the following steps: Identify target ecoregions; Conduct remote sensing detection on the target ecological area to obtain remote sensing image data; Based on AI technology, the remote sensing image data is subjected to radiation calibration, atmospheric correction, geometric correction, cropping and splicing to obtain standard remote sensing data; The standard remote sensing data are subjected to an index analysis to obtain regional index data of five ecological indexes including elevation, slope, aspect, vegetation type and land use type.

[0007] As a further limitation of the technical solution of the embodiment of the present invention, the index analysis of the standard remote sensing data to obtain regional index data including five ecological indicators including elevation, slope, aspect, vegetation type and land use type specifically includes the following steps: Using a digital elevation model, a three-dimensional terrain analysis is performed on the standard remote sensing data to obtain elevation data, slope data and aspect data; Performing spectral feature analysis and classification on the standard remote sensing data to obtain vegetation type data; Performing color texture analysis and classification on the standard remote sensing data to obtain land use type data; The elevation data, the slope data, the aspect data, the vegetation type data and the land use type data are integrated to generate regional index data.

[0008] As a further limitation of the technical solution of the embodiment of the present invention, the ecological sensitivity evaluation and integration of five ecological indicators of elevation, slope, aspect, vegetation type and land use type are performed according to the regional indicator data, and the target ecological area is divided into multiple ecologically sensitive sub-areas, which specifically includes the following steps: Conducting ecological sensitivity assessment of elevation based on the regional index data to obtain elevation assessment data; Conducting an ecological sensitivity assessment of the slope based on the regional index data to obtain slope assessment data; Conducting ecological sensitivity assessment of slope aspect according to the regional index data to obtain slope aspect assessment data; Conducting an ecological sensitivity assessment of vegetation types based on the regional indicator data to obtain vegetation type assessment data; Conducting ecological sensitivity assessment of land use types based on the regional indicator data to obtain land use type assessment data; The elevation evaluation data, the slope evaluation data, the slope aspect evaluation data, the vegetation type evaluation data and the land use type evaluation data are integrated to divide the area into multiple ecologically sensitive sub-areas.

[0009] As a further limitation of the technical solution of the embodiment of the present invention, the fusing of the elevation evaluation data, the slope evaluation data, the aspect evaluation data, the vegetation type evaluation data and the land use type evaluation data to divide a plurality of ecologically sensitive sub-areas specifically comprises the following steps: Determine the fusion evaluation weight; According to the fusion evaluation weight, the elevation evaluation data, the slope evaluation data, the slope aspect evaluation data, the vegetation type evaluation data and the land use type evaluation data are fused to obtain fusion evaluation data; Analyzing the fused evaluation data to determine the sensitive level boundary of the target ecological area; According to the sensitivity level boundaries, the target ecological area is divided into multiple ecologically sensitive sub-areas.

[0010] Compared with the prior art, the present invention has the following beneficial effects: The present invention obtains the regional index data of five ecological indicators by acquiring the remote sensing image data of the target ecological area; performs ecological sensitivity evaluation and integration of the five ecological indicators, and divides multiple ecological sensitive sub-areas; selects the optimal feature extraction algorithm according to the ecological sensitivity data of each ecological sensitive sub-area and combines the importance weight coefficient to perform feature extraction to finally obtain a feature data set. The present invention can analyze and process remote sensing image data purposefully and with emphasis, and improves processing efficiency while meeting quality requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0012] Figure 1 A flow chart of a method provided by an embodiment of the present invention is shown.

[0013] Figure 2 A flow chart of data processing and indicator analysis in the method provided by an embodiment of the present invention is shown.

[0014] Figure 3 A flow chart of obtaining regional indicator data in the method provided in an embodiment of the present invention is shown.

[0015] Figure 4 A flow chart of ecological sensitivity assessment and integration in the method provided in an embodiment of the present invention is shown.

[0016] Figure 5 A flow chart of dividing ecologically sensitive sub-areas in the method provided in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0017] 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 and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] In some existing remote sensing image processing technologies, due to computing power limitations, if the remote sensing images cannot be analyzed and processed in a purposeful and focused manner, not only will it be impossible to obtain image data that meets quality requirements, but it will also restrict the accuracy and efficiency of ecological environment monitoring and assessment.

[0019] To solve the above problems, the embodiment of the present invention obtains the regional index data of five ecological indicators by acquiring the remote sensing image data of the target ecological area; performs ecological sensitivity evaluation and fusion of the five ecological indicators, and divides multiple ecological sensitive sub-areas; according to the ecological sensitivity data of each ecological sensitive sub-area, the optimal feature extraction algorithm is selected in combination with the importance weight coefficient to perform feature extraction to finally obtain a feature data set. The present invention can analyze and process remote sensing image data purposefully and with emphasis, and improves processing efficiency while meeting quality requirements.

[0020] Figure 1 A flow chart of a method provided by an embodiment of the present invention is shown.

[0021] Specifically, a method for ecological environment remote sensing image classification and feature extraction based on deep learning, the method specifically includes the following steps: Step S101, obtain remote sensing image data of the target ecological area, and based on AI technology, perform data processing and index analysis on the remote sensing image data to obtain regional index data including five ecological indicators: elevation, slope, slope aspect, vegetation type and land use type.

[0022] In the embodiment of the present invention, by determining the target ecological area, remote sensing detection is performed on the target ecological area to obtain remote sensing image data, and then based on AI technology, radiation calibration, atmospheric correction, geometric correction, cropping and splicing processing are performed on the remote sensing image data to eliminate the influence of factors such as atmosphere and terrain on reflectivity, and ensure the integrity and consistency of the data, generate standard remote sensing data, use digital elevation model and remote sensing image processing software to perform three-dimensional terrain analysis on the standard remote sensing data, and obtain elevation data, slope data and slope aspect data, wherein the elevation data reflects the terrain undulation, slope The data reflects the steepness of the surface, and the slope data indicates the direction of the surface. The spectral feature analysis and classification of standard remote sensing data are performed to extract different types of vegetation such as forests, grasslands, and shrubs to obtain vegetation type data. The color and texture feature analysis and classification of standard remote sensing data are then performed. Combined with machine learning or deep learning algorithms, land use type data including cultivated land, forest land, construction land, and water areas are obtained. Then, the elevation data, slope data, slope aspect data, vegetation type data, and land use type data are integrated to generate regional indicator data for the target ecological area.

[0023] Specifically, Figure 2 A flow chart of data processing and indicator analysis in the method provided by an embodiment of the present invention is shown.

[0024] Among them, in the preferred embodiment provided by the present invention, the remote sensing image data of the target ecological area is obtained, and based on AI technology, data processing and index analysis are performed on the remote sensing image data to obtain regional index data including five ecological indicators including elevation, slope, slope aspect, vegetation type and land use type, which specifically includes the following steps: Step S1011, determining the target ecological area; Step S1012, performing remote sensing detection on the target ecological area to obtain remote sensing image data; Step S1013, based on AI technology, performing radiation calibration, atmospheric correction, geometric correction, cropping and splicing processing on the remote sensing image data to obtain standard remote sensing data; Step S1014, performing index analysis on the standard remote sensing data to obtain regional index data including five ecological indexes: elevation, slope, aspect, vegetation type and land use type.

[0025] Specifically, based on AI technology, the process of performing radiometric calibration on the remote sensing image data specifically includes the following steps: Obtain pixel values ​​(DN values) in remote sensing image data, use the gain and offset parameters of the sensor to perform initial linear calibration on the pixel values, and obtain preliminary radiation values; According to the solar irradiance and the solar zenith angle, the preliminary radiation value is normalized so that the processed radiation value can adapt to different observation conditions and obtain the normalized radiation value; Correcting the normalized radiation value according to the spectral response of the sensor to adjust the radiation value to improve accuracy, thereby obtaining a first corrected radiation value; The pixel values ​​are normalized and input into the deep learning model for prediction to obtain the second corrected radiation value; Combining the first corrected radiation value and the second corrected radiation value, and setting a mixing factor to adjust the importance of the first corrected radiation value and the second corrected radiation value to obtain a combined corrected radiation value; An optimization function is used to determine the best mixing factor to minimize the error between the combined corrected radiation value and the second corrected radiation value to obtain the best mixing factor, and the importance of the first corrected radiation value and the second corrected radiation value is adjusted using the best mixing factor to obtain the final corrected radiation value.

[0026] Specifically, based on AI technology, the process of performing atmospheric correction on the remote sensing image data specifically includes the following steps: Construct an atmospheric radiation transfer model, which divides the atmosphere into multiple layers, each with different aerosol concentrations and water vapor contents, to more accurately simulate the impact of the atmosphere on radiation, and give the model initial parameters of aerosol concentrations and water vapor contents; Using the final corrected radiation value as input, calculate the absorption and scattering effects of each atmospheric layer on radiation to obtain the attenuation effect of each layer, and superimpose the total radiation attenuation to obtain the surface reflectivity; After the optimization of the initial parameters of the model is completed using genetic algorithm, the final surface reflectivity is obtained; Based on the deep neural network (DNN), the final surface reflectivity is used as training data to learn and fit the complex atmospheric radiation transfer model to improve the calculation speed and accuracy and obtain the corrected surface reflectivity.

[0027] Specifically, based on AI technology, the process of geometrically correcting the remote sensing image data includes the following steps: The observation points of remote sensing images are converted into geographic coordinates using projection transformation matrix and normalized; Calculate the difference between the transformed geographic coordinates and the actual geographic coordinates; According to the difference between the converted geographic coordinates and the actual geographic coordinates, the unit matrix is ​​used as the initial estimate, and the projection transformation matrix is ​​optimized by the least square method to minimize the difference and obtain the optimized projection transformation matrix; Use the optimized projection transformation matrix to perform preliminary geometric correction on the remote sensing image to obtain a preliminary geometrically corrected image; The image with preliminary geometric correction is used as input data and input into the convolutional neural network for prediction, so as to obtain the predicted geographic coordinates by performing nonlinear adjustment through the convolutional neural network; The actual geographic coordinates corresponding to the geographic coordinates in the preliminary geometrically corrected image are used as target data to train the convolutional neural network. After the training is completed, the final predicted geographic coordinates are obtained, and then the final predicted geographic coordinates are interpolated back to the image space to finally obtain the geometrically corrected image.

[0028] Specifically, Figure 3 A flow chart of obtaining regional indicator data in the method provided in an embodiment of the present invention is shown.

[0029] Among them, in the preferred embodiment provided by the present invention, the index analysis of the standard remote sensing data to obtain regional index data including five ecological indicators including elevation, slope, aspect, vegetation type and land use type specifically includes the following steps: Step S10141, using a digital elevation model, performing a three-dimensional terrain analysis on the standard remote sensing data to obtain elevation data, slope data, and aspect data; Step S10142, performing spectral feature analysis and classification on the standard remote sensing data to obtain vegetation type data; Step S10143, performing color texture analysis and classification on the standard remote sensing data to obtain land use type data; Step S10144, integrating the elevation data, the slope data, the aspect data, the vegetation type data and the land use type data to generate regional index data.

[0030] Specifically, using a digital elevation model to perform three-dimensional terrain analysis on the standard remote sensing data, obtaining elevation data, slope data and aspect data specifically includes the following steps: Step S10141a, using a digital elevation model to process the corrected remote sensing data, obtain the elevation information of each grid point, and obtain the elevation data by weighted summation of the elevation information of each grid point; Step S10141b, by calculating the elevation data x Direction and y The derivative of the direction gives the slope data; Step S10141c, by calculating the elevation data x Direction and y After taking the ratio of the derivatives of the directions, take the inverse tangent function to get the aspect data.

[0031] Specifically, performing spectral feature analysis and classification on the standard remote sensing data to obtain vegetation type data specifically includes the following steps: Step S10142a, performing normalization processing based on the reflectance value of the near infrared band and the reflectance value of the red band to obtain a normalized vegetation index; In this step, specifically, the reflection value of the red band is subtracted from the reflection value of the near infrared band to obtain a reflection difference, and then the reflection difference is divided by the sum of the reflection value of the near infrared band and the reflection value of the red band to obtain the normalized vegetation index.

[0032] Step S10142b, subtracting the reflection value of the near infrared band from the reflection value of the red band to obtain a reflection difference, obtaining a correction value of the red band and a correction value of the blue band, and calculating an enhanced vegetation index according to the reflection difference, the correction value of the red band and the correction value of the blue band; Specifically, in this step, the gain factor is multiplied by the reflectance difference to obtain a first numerator, and then the first numerator is divided by the first denominator to obtain the enhanced vegetation index. The first denominator is the reflectance value of the near infrared band plus the correction value of the red band minus the correction value of the blue band plus the radiation transmission correction parameter.

[0033] Step S10142c, obtaining a first adjustment parameter, and calculating a land surface adjusted vegetation index according to the first adjustment parameter and the reflection difference; Specifically, in this step, the reflection difference value*(1+first adjustment parameter) is used as the second numerator, the sum of the reflection value of the near infrared band, the reflection value of the red band and the first adjustment parameter is used as the second denominator, and the second numerator is divided by the second denominator to obtain the land surface adjusted vegetation index; Step S10142d, the normalized vegetation index, the enhanced vegetation index and the land surface adjusted vegetation index are combined into a feature vector, and principal component analysis is performed on the feature vector to reduce the dimension of the data and obtain the feature vector after the dimension reduction; Step S10142e: Use a support vector machine to classify the feature vector after dimensionality reduction to determine the vegetation type of each pixel.

[0034] Specifically, performing color texture analysis and classification on the standard remote sensing data to obtain land use type data specifically includes the following steps: Step S10143a, for each pixel, calculate a color histogram according to its reflection values ​​in different bands; Step S10143b, for each pixel point, calculate its color moment in a specific band, the color moment includes the color mean in the specific band, the standard deviation of the color in the specific band, and the skewness of the color in the specific band; Step S10143c: for each pixel, calculate the gray level co-occurrence matrix in a specific band, and extract texture features from the gray level co-occurrence matrix, where the entropy texture features include contrast, energy, homogeneity and entropy; Step S10143d, for each pixel, calculate its local binary pattern; Step S10143e, the color histogram, color moment, texture feature and local binary pattern are combined into a feature vector to obtain a comprehensive feature, and a convolutional neural network (CNN) is used to fuse and classify the comprehensive feature vector to obtain the land use type of each pixel.

[0035] Furthermore, the ecological environment remote sensing image classification and feature extraction method based on deep learning also includes the following steps: Step S102, based on the regional indicator data, ecological sensitivity evaluation and integration of five ecological indicators, namely, elevation, slope, aspect, vegetation type and land use type, are performed to divide the target ecological region into a plurality of ecologically sensitive sub-regions.

[0036] In an embodiment of the present invention, an ecological sensitivity evaluation of elevation is performed according to regional indicator data to obtain elevation evaluation data; an ecological sensitivity evaluation of slope is performed according to regional indicator data to obtain slope evaluation data; an ecological sensitivity evaluation of aspect is performed according to regional indicator data to obtain aspect evaluation data; an ecological sensitivity evaluation of vegetation type is performed according to regional indicator data to obtain vegetation type evaluation data; and an ecological sensitivity evaluation of land use type is performed according to regional indicator data to obtain land use type evaluation data, and then a fusion evaluation weight is determined; and then according to the fusion evaluation weight, the elevation evaluation data, the slope evaluation data, the aspect evaluation data, the vegetation type evaluation data and the land use type evaluation data are fused to obtain fusion evaluation data; by analyzing the fusion evaluation data, the sensitivity level boundaries of different sensitivity levels in the target ecological area are determined; according to the sensitivity level boundaries, the target ecological area is divided into multiple ecologically sensitive sub-areas of different sensitivity levels, including extremely sensitive sub-areas, highly sensitive sub-areas, moderately sensitive sub-areas, slightly sensitive sub-areas and insensitive sub-areas.

[0037] Specifically, Figure 4 A flow chart of ecological sensitivity assessment and integration in the method provided in an embodiment of the present invention is shown.

[0038] Among them, in the preferred embodiment provided by the present invention, the ecological sensitivity evaluation and integration of five ecological indicators of elevation, slope, aspect, vegetation type and land use type are performed according to the regional indicator data, and the target ecological area is divided into multiple ecologically sensitive sub-areas, which specifically includes the following steps: Step S1021, performing an ecological sensitivity assessment of elevation according to the regional index data to obtain elevation assessment data; Step S1022, performing an ecological sensitivity assessment of the slope according to the regional index data, and obtaining slope assessment data; Step S1023, performing an ecological sensitivity assessment of the slope aspect according to the regional index data, and obtaining slope aspect assessment data; Step S1024, performing an ecological sensitivity assessment of vegetation types according to the regional index data, and obtaining vegetation type assessment data; Step S1025, performing an ecological sensitivity assessment of land use types according to the regional index data, and obtaining land use type assessment data; Step S1026, integrating the elevation evaluation data, the slope evaluation data, the slope aspect evaluation data, the vegetation type evaluation data and the land use type evaluation data to divide the area into multiple ecologically sensitive sub-areas.

[0039] Specifically, Figure 5 A flow chart of dividing ecologically sensitive sub-areas in the method provided in an embodiment of the present invention is shown.

[0040] Among them, in the preferred embodiment provided by the present invention, the integration of the elevation evaluation data, the slope evaluation data, the slope aspect evaluation data, the vegetation type evaluation data and the land use type evaluation data to divide a plurality of ecologically sensitive sub-areas specifically comprises the following steps: Step S10261, determining the fusion evaluation weight; Step S10262, according to the fusion evaluation weight, the elevation evaluation data, the slope evaluation data, the aspect evaluation data, the vegetation type evaluation data and the land use type evaluation data are fused to obtain fused evaluation data; Step S10263, analyzing the fused evaluation data to determine the sensitive level boundary of the target ecological area; Step S10264: Divide the target ecological area into multiple ecologically sensitive sub-areas according to the sensitivity level boundaries.

[0041] Furthermore, the ecological environment remote sensing image classification and feature extraction method based on deep learning also includes the following steps: Step S103, obtaining ecological sensitivity data corresponding to each of the ecologically sensitive sub-areas, wherein the ecological sensitivity data at least includes importance weight coefficients corresponding to elevation, slope, aspect, vegetation type, land use type and texture features in the ecologically sensitive sub-areas.

[0042] Step S104, when it is determined that the importance weight coefficient is greater than the preset importance weight coefficient, the optimal feature extraction algorithm model is selected in the current item to perform feature extraction to obtain the corresponding optimal sub-features, and the optimal sub-features corresponding to each item are integrated to generate a complete feature data set.

[0043] Specifically, in step S104, the following sub-steps are included: Step S1041, when it is determined that the importance weight coefficient of the elevation item is greater than the preset importance weight coefficient, the average elevation feature and the elevation range feature in each ecologically sensitive sub-area are extracted using a digital elevation model; Step S1042: when it is determined that the importance weight coefficient of the slope item or the aspect item is greater than the preset importance weight coefficient, the slope and aspect of each ecologically sensitive sub-area are calculated using a spatial analysis algorithm based on a digital elevation model; Step S1043: when it is determined that the importance weight coefficient of the vegetation type item or the land use type item is greater than the preset importance weight coefficient, the vegetation type item or the land use type corresponding to each ecologically sensitive sub-area is calculated and determined using the maximum likelihood classification algorithm; Step S1044: when it is determined that the importance weight coefficient of the texture feature item is greater than the preset importance weight coefficient, the texture feature of each ecologically sensitive sub-region is extracted using the gray level co-occurrence matrix.

[0044] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0045] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0046] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0047] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0048] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for classification and feature extraction of ecological environment remote sensing images based on deep learning, characterized in that: The method specifically comprises the following steps: Acquire remote sensing image data of the target ecological area, and perform data processing and index analysis on the remote sensing image data based on AI technology to obtain regional index data including five ecological indicators: elevation, slope, aspect, vegetation type, and land use type; Based on the regional indicator data, ecological sensitivity evaluation and integration of five ecological indicators, namely elevation, slope, aspect, vegetation type and land use type, are carried out to divide the target ecological area into multiple ecologically sensitive sub-areas; Obtaining ecological sensitivity data corresponding to each of the ecologically sensitive sub-areas, wherein the ecological sensitivity data at least includes importance weight coefficients corresponding to elevation, slope, aspect, vegetation type, land use type, and texture features in the ecologically sensitive sub-areas; When it is determined that the importance weight coefficient is greater than the preset importance weight coefficient, the optimal feature extraction algorithm model is selected in the current item to perform feature extraction to obtain the corresponding optimal sub-features, and the optimal sub-features corresponding to each item are integrated to generate a complete feature data set; The step of acquiring remote sensing image data of the target ecological area, performing data processing and index analysis on the remote sensing image data based on AI technology, and acquiring regional index data including five ecological indexes including elevation, slope, aspect, vegetation type and land use type specifically includes the following steps: Identify target ecoregions; Conduct remote sensing detection on the target ecological area to obtain remote sensing image data; Based on AI technology, the remote sensing image data is subjected to radiation calibration, atmospheric correction, geometric correction, cropping and splicing to obtain standard remote sensing data; The standard remote sensing data are subjected to an index analysis to obtain regional index data of five ecological indexes including elevation, slope, aspect, vegetation type and land use type.

2. The method for classification and feature extraction of ecological environment remote sensing images based on deep learning according to claim 1, characterized in that: Based on AI technology, the process of radiometric calibration of the remote sensing image data specifically includes the following steps: Obtain pixel values ​​in remote sensing image data, use the gain and offset parameters of the sensor to perform initial linear calibration on the pixel values, and obtain preliminary radiation values; According to the solar irradiance and the solar zenith angle, the preliminary radiation value is normalized to obtain the normalized radiation value; Correcting the normalized radiation value according to the spectral response of the sensor to obtain a first corrected radiation value; The pixel values ​​are normalized and input into the deep learning model for prediction to obtain the second corrected radiation value; The first corrected radiation value and the second corrected radiation value are combined, and a mixing factor is set to adjust the importance of the first corrected radiation value and the second corrected radiation value to obtain a combined corrected radiation value; An optimization function is used to determine the best mixing factor to minimize the error between the combined corrected radiation value and the second corrected radiation value to obtain the best mixing factor, and the importance of the first corrected radiation value and the second corrected radiation value is adjusted using the best mixing factor to obtain the final corrected radiation value.

3. The method for classification and feature extraction of ecological environment remote sensing images based on deep learning according to claim 2 is characterized in that: Based on AI technology, the process of performing atmospheric correction on the remote sensing image data specifically includes the following steps: Construct an atmospheric radiation transfer model. The atmospheric radiation transfer model divides the atmosphere into multiple layers, each layer has different aerosol concentrations and water vapor contents, and the initial model parameters of aerosol concentrations and water vapor contents are given; Using the final corrected radiation value as input, calculate the absorption and scattering effects of each atmospheric layer on radiation to obtain the attenuation effect of each layer, and superimpose the total radiation attenuation to obtain the surface reflectivity; After the optimization of the initial parameters of the model is completed using genetic algorithm, the final surface reflectivity is obtained; Based on a deep neural network, the final surface reflectivity is used as training data to learn and fit the complex atmospheric radiation transfer model to obtain the corrected surface reflectivity.

4. The method for classification and feature extraction of ecological environment remote sensing images based on deep learning according to claim 3 is characterized in that: The step of performing index analysis on the standard remote sensing data to obtain regional index data including five ecological indexes, including elevation, slope, aspect, vegetation type and land use type, specifically comprises the following steps: Using a digital elevation model, a three-dimensional terrain analysis is performed on the standard remote sensing data to obtain elevation data, slope data and aspect data; Performing spectral feature analysis and classification on the standard remote sensing data to obtain vegetation type data; Performing color texture analysis and classification on the standard remote sensing data to obtain land use type data; The elevation data, the slope data, the aspect data, the vegetation type data and the land use type data are integrated to generate regional index data.

5. The method for classification and feature extraction of ecological environment remote sensing images based on deep learning according to claim 4, characterized in that: Using a digital elevation model, performing a three-dimensional terrain analysis on the standard remote sensing data to obtain elevation data, slope data, and aspect data specifically includes the following steps: The calibrated remote sensing data is processed using a digital elevation model to obtain the elevation information of each grid point, and the elevation data is obtained by weighted summation of the elevation information of each grid point; By calculating the elevation data in x Direction and y The derivative of the direction gives the slope data; By calculating the elevation data in x Direction and y After taking the ratio of the derivatives of the directions, take the inverse tangent function to get the aspect data.

6. The method for classification and feature extraction of ecological environment remote sensing images based on deep learning according to claim 5, characterized in that: Performing spectral feature analysis and classification on the standard remote sensing data to obtain vegetation type data specifically includes the following steps: Based on the reflectance value of the near infrared band and the reflectance value of the red band, normalization processing is performed to obtain the normalized vegetation index; The reflection value of the near-infrared band is subtracted from the reflection value of the red band to obtain the reflection difference, the correction value of the red band and the correction value of the blue band are obtained, and the enhanced vegetation index is calculated according to the reflection difference, the correction value of the red band and the correction value of the blue band; Obtaining a first adjustment parameter, and calculating a land surface adjusted vegetation index according to the first adjustment parameter and the reflection difference; The normalized vegetation index, enhanced vegetation index and land surface adjusted vegetation index are combined into a feature vector, and principal component analysis is performed on the feature vector to reduce the dimension of the data to obtain the feature vector after dimension reduction; The support vector machine is used to classify the feature vector after dimension reduction to determine the vegetation type of each pixel.

7. The method for classification and feature extraction of ecological environment remote sensing images based on deep learning according to claim 6, characterized in that: Performing color texture analysis and classification on the standard remote sensing data to obtain land use type data specifically includes the following steps: For each pixel, the color histogram is calculated based on the reflectance values ​​of different bands; For each pixel, calculate its color moment in a specific band. The color moment includes the color mean in a specific band, the standard deviation of the color in a specific band, and the skewness of the color in a specific band. For each pixel, the gray level co-occurrence matrix is ​​calculated in a specific band, and texture features are extracted from the gray level co-occurrence matrix. The entropy texture features include contrast, energy, homogeneity and entropy. For each pixel, calculate its local binary pattern; The color histogram, color moment, texture features and local binary patterns are combined into a feature vector to obtain a comprehensive feature. The convolutional neural network performs feature fusion and classification on the comprehensive feature vector to obtain the land use type of each pixel.

8. The method for classification and feature extraction of ecological environment remote sensing images based on deep learning according to claim 7, characterized in that: The method of performing ecological sensitivity evaluation and integration of five ecological indicators, namely, elevation, slope, aspect, vegetation type and land use type, based on the regional indicator data, and dividing the target ecological region into a plurality of ecologically sensitive sub-regions specifically includes the following steps: Conducting ecological sensitivity assessment of elevation based on the regional index data to obtain elevation assessment data; Conducting an ecological sensitivity assessment of the slope based on the regional index data to obtain slope assessment data; Conducting ecological sensitivity assessment of slope aspect according to the regional index data to obtain slope aspect assessment data; Conducting an ecological sensitivity assessment of vegetation types based on the regional indicator data to obtain vegetation type assessment data; Conducting ecological sensitivity assessment of land use types based on the regional indicator data to obtain land use type assessment data; The elevation evaluation data, the slope evaluation data, the slope aspect evaluation data, the vegetation type evaluation data and the land use type evaluation data are integrated to divide the area into multiple ecologically sensitive sub-areas.

9. The method for classification and feature extraction of ecological environment remote sensing images based on deep learning according to claim 8, characterized in that: The step of fusing the elevation evaluation data, the slope evaluation data, the slope aspect evaluation data, the vegetation type evaluation data, and the land use type evaluation data to divide a plurality of ecologically sensitive sub-areas specifically comprises the following steps: Determine the fusion evaluation weight; According to the fusion evaluation weight, the elevation evaluation data, the slope evaluation data, the slope aspect evaluation data, the vegetation type evaluation data and the land use type evaluation data are fused to obtain fusion evaluation data; Analyzing the fused evaluation data to determine the sensitive level boundary of the target ecological area; According to the sensitivity level boundaries, the target ecological area is divided into multiple ecologically sensitive sub-areas.

10. The method for classification and feature extraction of ecological environment remote sensing images based on deep learning according to claim 9, characterized in that: When it is determined that the importance weight coefficient is greater than the preset importance weight coefficient, the optimal feature extraction algorithm model is selected in the current item to perform feature extraction to obtain the corresponding optimal sub-feature, which specifically includes the following steps: When it is determined that the importance weight coefficient of the elevation item is greater than the preset importance weight coefficient, the digital elevation model is used to extract the average elevation characteristics and elevation range characteristics in each ecologically sensitive sub-area; When it is determined that the importance weight coefficient of the slope item or the aspect item is greater than the preset importance weight coefficient, the slope and aspect of each ecologically sensitive sub-area are calculated using a spatial analysis algorithm based on a digital elevation model; When it is determined that the importance weight coefficient of the vegetation type item or the land use type item is greater than the preset importance weight coefficient, the maximum likelihood classification algorithm is used to calculate and determine the vegetation type item or land use type corresponding to each ecologically sensitive sub-area; When it is determined that the importance weight coefficient of the texture feature item is greater than the preset importance weight coefficient, the gray level co-occurrence matrix is ​​used to extract the texture features of each ecologically sensitive sub-area.

Images