Remote sensing image classification method and system, computer device and storage medium

Abstract

This invention relates to the field of remote sensing technology, specifically to remote sensing image classification methods, systems, computer equipment, and storage media. The remote sensing image classification method includes the following steps: based on raw remote sensing data, using satellite cloud image analysis and surface temperature mapping techniques to perform comprehensive analysis of microclimate characteristics; using K-means clustering analysis algorithm to classify the data; and generating a comprehensive climate characteristic dataset. The beneficial effects of this invention are that by integrating K-means clustering, adaptive histogram equalization, convolutional neural networks, long short-term memory networks, and fully convolutional network algorithms, it can effectively identify geographic patterns and monitor abnormal climate events. Simultaneously, it optimizes surface classification, combines Gaussian filtering, MODIS cloud detection algorithms, and data assimilation techniques to deeply preprocess raw data, improve data quality, enrich information content, and effectively predict and respond to climate change using long short-term memory networks and time series analysis techniques.

Description

Technical Field

[0001] This invention relates to the field of remote sensing technology, and in particular to remote sensing image classification methods, systems, computer equipment, and storage media. Background Technology

[0002] Remote sensing technology mainly involves acquiring information about the Earth's surface from a distance. It uses electromagnetic waves of different wavelengths to scan the Earth's surface, thereby capturing images and data. It is widely used in many fields such as map making, environmental monitoring, agriculture, and urban planning. Remote sensing images can provide important information about surface features, vegetation cover, land use, and environmental changes.

[0003] Remote sensing image classification is a technique used to locate and process remote sensing data. Its purpose is to quickly and accurately retrieve specific images from a large amount of remote sensing image data, supporting various scientific research, commercial, and government decision-making activities. For example, in environmental monitoring, remote sensing image classification can quickly locate images of polluted areas or find images of specific crop growth conditions in agriculture. Image processing techniques are used to preprocess the original remote sensing images, such as denoising and enhancing contrast to improve image quality. Pattern recognition technology is used to identify specific patterns or objects in the images, thereby automatically identifying and classifying large amounts of image data to achieve retrieval.

[0004] Traditional remote sensing image classification techniques have several shortcomings, including: climate feature analysis is often coarse, making it difficult to accurately capture subtle climate changes and detailed distributions of surface temperature, thus limiting the depth and accuracy of their application in climate research and environmental monitoring; they are not precise enough in identifying complex geographical patterns, making it difficult to provide sufficiently in-depth insights for environmental monitoring and geographical research, especially in understanding and predicting the impact of climate change on the geographical environment; and the classification of land surface types in remote sensing images lacks precision and detail in traditional methods. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies by proposing a remote sensing image classification method, system, computer equipment, and storage medium.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a remote sensing image classification method, comprising the following steps:

[0007] S1: Based on the original remote sensing data, satellite cloud image analysis and surface temperature mapping technology are used to conduct a comprehensive analysis of microclimate characteristics. K-means clustering analysis algorithm is used to classify the data and generate a comprehensive climate characteristic dataset.

[0008] S2: Based on the comprehensive climate feature dataset, image quality optimization is performed using a contrast-limited adaptive histogram equalization method and convolutional neural network technology to generate image quality optimization results;

[0009] S3: Based on the image quality optimization results, the fractal dimension calculation method is used to identify and analyze the geographic patterns of remote sensing images, and generate complex pattern recognition data;

[0010] S4: Based on the complex pattern recognition data, use long short-term memory networks and time series analysis techniques to detect abnormal patterns in the time series and generate an abnormal climate event analysis report;

[0011] S5: Based on the abnormal climate event analysis report, a fully convolutional network is used to perform semantic segmentation of the image to classify the land surface and generate a land surface type classification map.

[0012] The comprehensive climate feature dataset includes temperature, humidity, and cloud cover distribution data. The image quality optimization results specifically refer to remote sensing images with improved illumination, contrast, and sharpness. The complex pattern recognition data includes geographical features and climate change data. The abnormal climate event analysis report specifically refers to unconventional climate and potential environmental risks. The land surface type classification map includes remote sensing images of cities, farmland, forests, and multiple land surface covers.

[0013] As a further aspect of the present invention, based on raw remote sensing data, a comprehensive analysis of microclimate characteristics is performed using satellite cloud image analysis and surface temperature mapping techniques. The data is then classified using a K-means clustering algorithm to generate a comprehensive climate characteristic dataset. The specific steps are as follows:

[0014] S101: Based on the original remote sensing data, Gaussian filtering is used to denoise and enhance the image, generating preprocessed remote sensing data;

[0015] S102: Based on the preprocessed remote sensing data, cloud information data is extracted and generated using the MODIS cloud detection algorithm;

[0016] S103: Based on the cloud information data, the MODTRAN algorithm is used to obtain the surface temperature distribution and generate a surface temperature distribution map;

[0017] S104: Based on the surface temperature distribution map, cloud map and temperature data are integrated using a data assimilation method to generate fused climate data;

[0018] S105: Based on the fused climate data, the K-means clustering algorithm is used to divide the climate regions and generate climate region classification data;

[0019] S106: Based on the climate region classification data, a multi-dimensional data fusion method is used to integrate the data, and the climate characteristics are extracted by principal component analysis to generate a comprehensive climate characteristic dataset.

[0020] As a further aspect of the present invention, based on the comprehensive climate feature dataset, the steps of performing image quality optimization using a contrast-limited adaptive histogram equalization method and convolutional neural network technology to generate image quality optimization results are as follows:

[0021] S201: Based on the comprehensive climate feature dataset, image contrast is optimized using contrast-limited adaptive histogram equalization technology to generate contrast-optimized data.

[0022] S202: Based on the contrast optimization data, feature enhancement is performed using a VGG-16 convolutional neural network model to generate feature-enhanced data;

[0023] S203: Based on the feature enhancement data, median filtering technology is used to perform image denoising processing to generate denoised image data;

[0024] S204: Based on the noise-reduced image data, USM sharpening technology is used to improve image clarity and generate sharpened image data;

[0025] S205: Based on the sharpened image data, a grayscale world algorithm is used to perform color balancing and generate color correction data;

[0026] S206: Based on the color correction data, the structural similarity index algorithm is used to evaluate the image quality and generate image quality optimization results.

[0027] As a further aspect of the present invention, based on the image quality optimization results, the fractal dimension calculation method is used to identify and analyze the geographic patterns of remote sensing images, and the steps to generate complex pattern recognition data are as follows:

[0028] S301: Based on the image quality optimization results, the multi-scale features of the image are extracted using two-dimensional discrete wavelet transform technology to generate multi-scale feature data;

[0029] S302: Based on the multi-scale feature data, a random forest algorithm is used for feature selection and geographic pattern recognition to generate geographic pattern data;

[0030] S303: Based on the aforementioned geographic pattern data, the fractal dimension is calculated using the box counting method to generate fractal dimension analysis data;

[0031] S304: Based on the fractal dimension analysis data, the Euclidean distance algorithm is used to perform similarity pattern recognition and generate pattern similarity data;

[0032] S305: Based on the pattern similarity data, use the support vector machine method to classify geographic patterns and generate geographic pattern classification data;

[0033] S306: Based on the aforementioned geographic pattern classification data, artificial neural networks and variance analysis methods are used to perform pattern recognition and deep analysis to generate complex pattern recognition data.

[0034] As a further aspect of the present invention, based on the complex pattern recognition data, the steps of using long short-term memory networks and time series analysis techniques to detect abnormal patterns in time series data and generate an abnormal climate event analysis report are as follows:

[0035] S401: Based on the complex pattern recognition data, the Z-score normalization method is used to process the data to generate time series data;

[0036] S402: Based on the time series data, an autoregressive moving average model is used to decompose the series and generate time series decomposed data;

[0037] S403: Based on the time series decomposed data, a long short-term memory network is used to generate a time series prediction model;

[0038] S404: Based on the time series prediction model, the statistical threshold method is used to detect abnormal patterns and generate preliminary abnormal pattern detection results;

[0039] S405: Based on the preliminary abnormal pattern detection results, the DBSCAN clustering algorithm is used to filter and verify the abnormal patterns to generate abnormal pattern data;

[0040] S406: Based on the aforementioned anomaly pattern data, use multiple regression analysis and the ARIMA model to predict future climate anomalies and generate an anomaly climate event analysis report.

[0041] As a further aspect of the present invention, based on the abnormal climate event analysis report, the steps of using a fully convolutional network to perform semantic segmentation of the image for land surface classification and generating a land surface type classification map are as follows:

[0042] S501: Based on the aforementioned abnormal climate event analysis report, a text processing algorithm is used to extract and convert data to generate an image dataset with adaptive semantic segmentation.

[0043] S502: Based on the image dataset of the adaptive semantic segmentation, gamma correction and edge enhancement techniques are used to optimize the image and generate optimized image data;

[0044] S503: Based on the optimized image data, a fully convolutional network model is used for learning and training to generate an FCN training model;

[0045] S504: Based on the FCN training model, pixel-by-pixel classification technology is used to perform semantic segmentation of the image and generate preliminary semantic segmentation results;

[0046] S505: Based on the preliminary semantic segmentation results, morphological operations are used to perform image processing to generate semantic segmentation data;

[0047] S506: Based on the semantic segmentation data, GIS technology is used to classify and spatially analyze land surface types, and a land surface type classification map is generated.

[0048] The remote sensing image classification system is used to perform remote sensing image classification methods. The system includes a data preprocessing module, a climate feature analysis module, an image quality optimization module, a geographic pattern recognition module, an abnormal climate analysis module, a land surface classification module, a meteorological data integration module, and an environmental impact assessment module.

[0049] The data preprocessing module uses Gaussian filtering noise reduction algorithm and histogram equalization technology to enhance the image based on the original remote sensing data, and generates preprocessed remote sensing data.

[0050] The climate feature analysis module analyzes microclimate features based on preprocessed remote sensing data, using the MODIS cloud detection algorithm, MODTRAN algorithm and data assimilation method to generate a comprehensive climate feature dataset.

[0051] The image quality optimization module is based on a comprehensive climate feature dataset and uses CLAHE technology and VGG-16 network to optimize image quality and generate image quality optimization results.

[0052] The geographic pattern recognition module uses two-dimensional discrete wavelet transform and random forest algorithm to perform geographic pattern recognition based on image quality optimization results, generating complex pattern recognition data.

[0053] The abnormal climate analysis module analyzes abnormal climate events based on complex pattern recognition data, using long short-term memory networks and ARIMA models, and generates an abnormal climate event analysis report.

[0054] The land surface classification module is based on the abnormal climate event analysis report. It uses a fully convolutional network model and GIS technology to perform semantic segmentation and land surface classification on the image, and generates a land surface type classification map.

[0055] The meteorological data integration module integrates meteorological data based on the land surface type classification map, using data fusion algorithms and spatial interpolation techniques to generate a meteorological information dataset.

[0056] The environmental impact assessment module is based on meteorological information datasets and uses environmental models and impact assessment algorithms to analyze the impact of climate change on the environment and generate an environmental impact assessment report.

[0057] As a further aspect of the present invention, the Gaussian filtering denoising algorithm specifically uses a Gaussian kernel to reduce image noise; the MODIS cloud detection algorithm specifically uses a remote sensing analysis method for identifying cloud layers; the MODTRAN algorithm specifically uses a surface temperature inversion based on an atmospheric model; the random forest algorithm specifically uses multiple decision trees for feature evaluation and classification; the ARIMA model specifically refers to an autoregressive integral moving average method used for time series analysis and prediction; the GIS technology specifically refers to a spatial analysis and map-making method; the environmental model specifically refers to a computational model used to simulate the impact of climate change on ecosystems and human activities; and the impact assessment algorithm includes assessing the potential impacts of climate change on biodiversity, water resources, and agricultural output.

[0058] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the remote sensing image classification system as described above.

[0059] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the remote sensing image classification method as described above.

[0060] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0061] This invention significantly improves the accuracy and efficiency of extracting key information from remote sensing data through K-means clustering, contrast-limited adaptive histogram equalization, convolutional neural networks, long short-term memory networks, and fully convolutional network algorithms. This not only makes the identification of geographic patterns and the monitoring of anomalous climate events more accurate but also optimizes the effectiveness of land surface classification. By combining various data processing techniques such as Gaussian filtering, MODIS cloud detection algorithms, and data assimilation methods, in-depth preprocessing and comprehensive analysis of raw data are performed, which not only improves data quality but also enriches the information content of the data, providing a more comprehensive foundation for subsequent analysis. Through contrast optimization, feature enhancement, noise reduction, sharpening techniques, and color correction, images are made clearer and easier to identify and analyze. Long short-term memory networks and time series analysis techniques are used to more effectively detect and predict anomalous patterns, enhancing the ability to respond promptly to environmental and climate change. Detailed land surface type classification through fully convolutional networks not only improves the accuracy of classification but also enhances the understanding and management capabilities of land surface resources. Attached Figure Description

[0062] Figure 1This is a schematic diagram of the workflow of the present invention;

[0063] Figure 2 This is a detailed flowchart of S1 of the present invention;

[0064] Figure 3 This is a detailed flowchart of the S2 process of the present invention;

[0065] Figure 4 This is a detailed flowchart of the S3 process of the present invention;

[0066] Figure 5 This is a detailed flowchart of the S4 process of the present invention;

[0067] Figure 6 This is a detailed flowchart of S5 of the present invention;

[0068] Figure 7 This is a system flowchart of the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0070] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0071] Example 1

[0072] Please see Figure 1 This invention provides a technical solution: a remote sensing image classification method, comprising the following steps:

[0073] S1: Based on the original remote sensing data, satellite cloud image analysis and surface temperature mapping technology are used to conduct a comprehensive analysis of microclimate characteristics. K-means clustering analysis algorithm is used to classify the data and generate a comprehensive climate characteristic dataset.

[0074] S2: Based on a comprehensive climate feature dataset, an adaptive histogram equalization method with contrast limitation and convolutional neural network technology are used to optimize image quality and generate image quality optimization results.

[0075] S3: Based on the image quality optimization results, the fractal dimension calculation method is used to identify and analyze the geographic patterns of remote sensing images, and generate complex pattern recognition data;

[0076] S4: Based on complex pattern recognition data, long short-term memory networks and time series analysis techniques are used to detect abnormal patterns in time series data and generate an analysis report of abnormal climate events.

[0077] S5: Based on the analysis report of abnormal climate events, a fully convolutional network is used to perform semantic segmentation of the image for land surface classification, generating a land surface type classification map.

[0078] The comprehensive climate feature dataset includes temperature, humidity, and cloud cover distribution data; image quality optimization results specifically include remote sensing images with improved illumination, contrast, and sharpness; complex pattern recognition data includes geographic features and climate change data; anomalous climate event analysis reports specifically include unconventional climates and potential environmental risks; land surface type classification maps include urban areas, farmland, forests, and remote sensing images with multiple land surface covers.

[0079] By combining satellite cloud image analysis, surface temperature mapping technology, and K-means clustering analysis algorithm, the ability to identify microclimate features was significantly improved. The application of contrast-limited adaptive histogram equalization and convolutional neural network technology enhanced image quality, including improvements in illumination, contrast, and sharpness, making image analysis more accurate and efficient. Fractal dimension calculation methods effectively identified and analyzed complex geographic patterns in remote sensing images, providing an important tool for understanding geographic features and monitoring environmental changes. Furthermore, combining long short-term memory networks and time series analysis technology accurately detected abnormal climate patterns, aiding in timely warnings and management of unconventional climate events and environmental risks. Image semantic segmentation using fully convolutional networks achieved refined land surface type classification, such as urban areas, farmland, and forests, which is of great significance for land use planning, agricultural monitoring, and ecological protection. These effects collectively enhance the application value of remote sensing data, making the extraction of key information from remote sensing data more efficient and accurate.

[0080] Please see Figure 2 Based on raw remote sensing data, satellite cloud image analysis and surface temperature mapping techniques are used to conduct a comprehensive analysis of microclimate characteristics. K-means clustering is then used for data classification to generate a comprehensive climate characteristic dataset. The specific steps are as follows:

[0081] S101: Based on the original remote sensing data, Gaussian filtering is used to denoise and enhance the image, generating preprocessed remote sensing data;

[0082] S102: Based on preprocessed remote sensing data, cloud information data is extracted and generated using the MODIS cloud detection algorithm;

[0083] S103: Based on cloud information data, the MODTRAN algorithm is used to obtain the surface temperature distribution and generate a surface temperature distribution map;

[0084] S104: Based on the surface temperature distribution map, cloud map and temperature data are integrated using data assimilation methods to generate fused climate data;

[0085] S105: Based on fused climate data, K-means clustering algorithm is used to divide climate regions and generate climate region classification data;

[0086] S106: Based on climate region classification data, a multi-dimensional data fusion method is used to integrate the data, and the climate characteristics are extracted through principal component analysis to generate a comprehensive climate characteristic dataset.

[0087] In step S101, a Gaussian filtering algorithm is used to denoise the raw remote sensing data to enhance image quality and generate preprocessed remote sensing data, reducing noise interference encountered in subsequent processing. In step S102, based on the preprocessed remote sensing data, the MODIS cloud detection algorithm is applied to extract cloud information and generate cloud information data. This step focuses on identifying and analyzing cloud characteristics to provide key data for subsequent climate analysis. The MODTRAN algorithm is used to process the cloud information data to obtain the surface temperature distribution and generate a surface temperature distribution map to show the spatial distribution of surface temperature. The cloud map and temperature data are combined through data assimilation methods. The process generates fused climate data. By integrating data from different sources, the accuracy and reliability of climate feature analysis are improved. In step S105, the K-means clustering algorithm is used to process the fused climate data to divide climate regions and generate climate region classification data. This allows a large area to be divided into different climate regions based on climate characteristics, which helps to conduct more targeted analysis and research. In step S106, the climate region classification data is further integrated by combining multi-dimensional data fusion methods. Climate features are extracted by principal component analysis to generate a comprehensive climate feature dataset, which provides a foundation for subsequent remote sensing image classification and analysis.

[0088] Please see Figure 3 Based on a comprehensive climate feature dataset, an adaptive histogram equalization method with contrast constraints and convolutional neural network technology are used to optimize image quality. The specific steps for generating the image quality optimization results are as follows:

[0089] S201: Based on a comprehensive climate feature dataset, an adaptive histogram equalization technique with contrast limitation is used to optimize image contrast and generate contrast-optimized data.

[0090] S202: Based on contrast-optimized data, a VGG-16 convolutional neural network model is used for feature enhancement to generate feature-enhanced data;

[0091] S203: Based on feature-enhanced data, median filtering technology is used to perform image denoising processing to generate denoised image data;

[0092] S204: Based on denoised image data, USM sharpening technology is used to improve image clarity and generate sharpened image data;

[0093] S205: Based on sharpened image data, a grayscale world algorithm is used to perform color balance and generate color correction data;

[0094] S206: Based on color correction data, the structural similarity index algorithm is used to evaluate image quality and generate image quality optimization results.

[0095] In step S201, contrast optimization is performed on the image using contrast-limited adaptive histogram equalization. By adjusting the image's histogram, the image contrast is improved, making details in the image more clearly visible, thus generating contrast-optimized data. In step S202, based on the contrast-optimized data, a VGG-16 convolutional neural network model is applied for feature enhancement. The model extracts and enhances features in the image using deep learning technology, further improving the image quality and analytical value, generating feature-enhanced data. Step S203 involves using median filtering to denoise the feature-enhanced data. Median filtering reduces image noise by replacing the value of each pixel in the image with the median value of its neighboring pixels, generating... In step S204, the denoised image data is processed using Unsharp Mask (USM) sharpening technology to improve image clarity. This technology enhances the edge contrast of the image, making the details of the image more distinct, and generates sharpened image data. Step S205 includes using the Gray World algorithm to perform color balancing on the sharpened image data. This algorithm assumes that the average scene color is gray and achieves color correction by adjusting the color distribution of the image, generating color-corrected data. In step S206, the structural similarity index algorithm is used to evaluate the image quality of the color-corrected data. This algorithm evaluates the image processing effect by comparing the similarity between the original image and the processed image, and finally generates image quality optimization results.

[0096] Please see Figure 4 Based on the image quality optimization results, the fractal dimension calculation method is used to identify and analyze geographic patterns in remote sensing images, and the specific steps for generating complex pattern recognition data are as follows:

[0097] S301: Based on the image quality optimization results, the two-dimensional discrete wavelet transform technique is used to extract multi-scale features of the image and generate multi-scale feature data.

[0098] S302: Based on multi-scale feature data, the random forest algorithm is used for feature selection and geographic pattern recognition to generate geographic pattern data;

[0099] S303: Based on geographic pattern data, box counting is used to calculate fractal dimension and generate fractal dimension analysis data.

[0100] S304: Based on fractal dimension analysis data, the Euclidean distance algorithm is used to perform similarity pattern recognition and generate pattern similarity data;

[0101] S305: Based on pattern similarity data, the support vector machine method is used to classify geographic patterns and generate geographic pattern classification data;

[0102] S306: Based on geographic pattern classification data, artificial neural networks and analysis of variance are used for pattern recognition and deep analysis to generate complex pattern recognition data.

[0103] In step S301, the image quality optimization result is processed using two-dimensional discrete wavelet transform technology to extract multi-scale features of the image. The key to this step is capturing important features of the image from different scales, thereby generating multi-scale feature data. In step S302, the random forest algorithm is applied to filter and process this multi-scale feature data to identify geographic patterns in the image. The random forest algorithm improves the accuracy of identification by considering the outputs of multiple decision trees, generating geographic pattern data. Step S303 involves calculating the fractal dimension of the aforementioned geographic pattern data using box counting. This calculation method can quantify the complexity and detail of patterns in the image, thereby generating fractal dimension analysis data. In step S304, Euclidean... Distance algorithms process fractal dimension analysis data to identify similar patterns. This step helps identify and classify similar geographic features by calculating the Euclidean distance between different patterns, generating pattern similarity data. Step S305 includes using support vector machines to further process the pattern similarity data to achieve accurate classification of geographic patterns. Support vector machines can process high-dimensional data and provide accurate classification results, generating geographic pattern classification data. In step S306, artificial neural networks and analysis of variance are used to perform in-depth analysis of geographic pattern classification data. This step combines the powerful learning ability of artificial neural networks with the statistical methods of analysis of variance to deeply understand and identify complex geographic patterns in images, ultimately generating complex pattern recognition data.

[0104] Please see Figure 5Based on complex pattern recognition data, and employing long short-term memory networks and time series analysis techniques, the specific steps for detecting anomalous patterns in time series data and generating an analysis report of anomalous climate events are as follows:

[0105] S401: Based on complex pattern recognition data, Z-score normalization is used to process the data and generate time series data;

[0106] S402: Based on time series data, an autoregressive moving average model is used to decompose the series and generate time series decomposed data;

[0107] S403: Based on time series decomposition data, a long short-term memory network is used to generate a time series prediction model;

[0108] S404: Based on the time series prediction model, the statistical threshold method is used to detect abnormal patterns and generate preliminary abnormal pattern detection results;

[0109] S405: Based on the preliminary abnormal pattern detection results, the DBSCAN clustering algorithm is used to filter and verify the abnormal patterns, and generate abnormal pattern data;

[0110] S406: Based on anomalous pattern data, use multiple regression analysis and ARIMA model to predict future climate anomalies and generate an anomalous climate event analysis report.

[0111] In step S401, the Z-score standardization method is used to standardize the data, transforming it into time series data more suitable for analysis. By reducing bias and outliers, the time series analysis becomes more accurate and reliable. In step S402, an autoregressive moving average model is used to decompose these time series data. By decomposing the time series data, the model reveals the basic trends and periodicity in the data, generating time series decomposed data, which provides a foundation for subsequent prediction and analysis. Step S403 involves using a Long Short-Term Memory (LSTM) network to process the time series decomposed data to generate a time series prediction model. The LSM network can effectively capture long-term dependencies in the time series. In step S404, based on the time series prediction model, a statistical thresholding method is used to detect abnormal patterns. This step identifies data points that deviate from the normal pattern by setting a statistical anomaly threshold, generating preliminary anomaly pattern detection results. Step S405 includes using the DBSCAN clustering algorithm to screen and verify these preliminary anomaly pattern detection results. DBSCAN is a density-based clustering algorithm that can identify closely connected clusters of points in the data, thereby effectively distinguishing and confirming anomaly patterns and generating anomaly pattern data. In step S406, the anomaly pattern data is processed using multiple regression analysis and the ARIMA model to predict future climate anomalies and generate an anomaly climate event analysis report. This method, which combines multiple regression analysis and the ARIMA model, not only provides a deep understanding of current anomaly patterns but also provides a powerful tool for predicting future climate anomalies.

[0112] Please see Figure 6 Based on the analysis report of abnormal climate events, the steps for generating a land surface type classification map by using a fully convolutional network for semantic segmentation of images are as follows:

[0113] S501: Based on the analysis report of abnormal climate events, a text processing algorithm is used to extract and convert data to generate an image dataset with adaptive semantic segmentation.

[0114] S502: An image dataset based on adaptive semantic segmentation, which uses gamma correction and edge enhancement techniques to optimize images and generate optimized image data;

[0115] S503: Based on the optimized image data, a fully convolutional network model is used for learning and training to generate an FCN training model;

[0116] S504: Based on the FCN training model, pixel-by-pixel classification technology is used to perform semantic segmentation of the image and generate preliminary semantic segmentation results;

[0117] S505: Based on the preliminary semantic segmentation results, morphological operations are used to perform image processing to generate semantic segmentation data;

[0118] S506: Based on semantic segmentation data, use GIS technology to classify and spatially distribute land surface types, and generate a land surface type classification map.

[0119] In step S501, based on the abnormal climate event analysis report, text processing algorithms are used to extract and convert the data in the report to generate an adaptive semantic segmentation image dataset, converting text information into a data format suitable for image processing. In step S502, based on the adaptive semantic segmentation image dataset, gamma correction and edge enhancement techniques are used to optimize the image. By adjusting the brightness and contrast of the image and enhancing the image edges, the image quality is improved, providing clearer image data for subsequent segmentation processing. Step S503 involves using a fully convolutional network model to learn and train on the optimized image data, generating an FCN training model. The fully convolutional network is capable of recognition and classification at the pixel level. In step S504, based on the FCN training model, the image is semantically segmented using pixel-by-pixel classification technology. This accurately assigns each pixel in the image to a specific category, generating a preliminary semantic segmentation result. Step S505 involves further processing the preliminary semantic segmentation result using morphological operations. Morphological operations are an image processing technique that modifies the shape of an image by applying structuring elements, which helps improve the quality and accuracy of the segmentation result. In step S506, GIS technology is used to classify and spatially distribute the semantic segmentation data to land surface types. GIS technology can effectively analyze and visualize the spatial distribution of land surface types, thereby generating a land surface type classification map.

[0120] Please see Figure 7 The remote sensing image classification system is used to perform remote sensing image classification methods. The system includes a data preprocessing module, a climate feature analysis module, an image quality optimization module, a geographic pattern recognition module, an abnormal climate analysis module, a land surface classification module, a meteorological data integration module, and an environmental impact assessment module.

[0121] The data preprocessing module uses Gaussian filtering noise reduction algorithm and histogram equalization technology to enhance the image based on the original remote sensing data, and generates preprocessed remote sensing data.

[0122] The climate feature analysis module analyzes microclimate features based on preprocessed remote sensing data, using the MODIS cloud detection algorithm, MODTRAN algorithm and data assimilation method to generate a comprehensive climate feature dataset.

[0123] The image quality optimization module is based on a comprehensive climate feature dataset and uses CLAHE technology and VGG-16 network to optimize image quality and generate image quality optimization results.

[0124] The geographic pattern recognition module uses two-dimensional discrete wavelet transform and random forest algorithm to perform geographic pattern recognition based on image quality optimization results, generating complex pattern recognition data.

[0125] The abnormal climate analysis module uses complex pattern recognition data, long short-term memory networks and ARIMA models to analyze abnormal climate events and generate abnormal climate event analysis reports.

[0126] The land surface classification module is based on the analysis report of abnormal climate events. It uses a fully convolutional network model and GIS technology to perform semantic segmentation and land surface classification on images, and generates a land surface type classification map.

[0127] The meteorological data integration module integrates meteorological data based on the land surface type classification map, using data fusion algorithms and spatial interpolation techniques to generate a meteorological information dataset.

[0128] The environmental impact assessment module uses meteorological information datasets, environmental models, and impact assessment algorithms to analyze the environmental impacts of climate change and generate environmental impact assessment reports.

[0129] Gaussian filtering denoising algorithms specifically reduce image noise using Gaussian kernels; MODIS cloud detection algorithms are remote sensing analysis methods used to identify cloud layers; MODTRAN algorithms are surface temperature inversion based on atmospheric models; random forest algorithms utilize multiple decision trees for feature evaluation and classification; ARIMA models refer to the autoregressive integral moving average method used for time series analysis and prediction; GIS technology is based on spatial analysis and map-making methods; environmental models refer to computational models used to simulate the impacts of climate change on ecosystems and human activities; and impact assessment algorithms include assessing the potential impacts of climate change on biodiversity, water resources, and agricultural yields.

[0130] The Gaussian filtering denoising algorithm and histogram equalization technique in the data preprocessing module significantly improved the quality of the raw remote sensing data. By reducing noise and improving image contrast, this module laid a solid foundation for subsequent analysis. Secondly, the climate feature analysis module, combining the MODIS cloud detection algorithm, MODTRAN algorithm, and data assimilation methods, effectively analyzed microclimate characteristics and generated a comprehensive climate feature dataset. This provides important data support for understanding and predicting climate change. The image quality optimization module further improved image quality, particularly through the combination of CLAHE technology and the VGG-16 network, optimizing image sharpness and feature representation. The geographic pattern recognition module's two-dimensional discrete wavelet transform and random forest algorithm effectively identified geographic patterns and generated complex pattern recognition data, providing in-depth insights into understanding surface features and environmental changes. The combination of Long Short-Term Memory (LSTM) networks and ARIMA models in the climate analysis module enables the system to accurately analyze and predict anomalous climate events and generate anomalous climate event analysis reports. This provides information for climate change monitoring and response. The application of fully convolutional network models and GIS technology in the land surface classification module demonstrates high efficiency in image semantic segmentation and land surface classification, generating land surface type classification maps. This is crucial for land use planning and ecological protection. The data fusion algorithm and spatial interpolation technology in the meteorological data integration module combine land surface classification results with meteorological data to generate meteorological information datasets, providing a more comprehensive perspective for climate change research and environmental monitoring. The environmental impact assessment module's environmental models and impact assessment algorithms comprehensively analyze the environmental impacts of climate change and generate environmental impact assessment reports, which are of great significance for formulating policies for environmental protection and climate change adaptation.

[0131] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the remote sensing image classification system described above.

[0132] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the remote sensing image classification method described above.

[0133] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method of classifying a remote sensing image, characterized in that, Includes the following steps: Based on raw remote sensing data, satellite cloud image analysis and surface temperature mapping techniques are used to conduct a comprehensive analysis of microclimate characteristics. K-means clustering analysis algorithm is used to classify the data and generate a comprehensive climate characteristic dataset. Based on the comprehensive climate feature dataset, an image quality optimization method with contrast limitation and convolutional neural network technology are used to optimize the image quality and generate image quality optimization results. Based on the image quality optimization results, the fractal dimension calculation method is used to identify and analyze the geographic patterns of remote sensing images, and generate complex pattern recognition data. Based on the complex pattern recognition data, long short-term memory networks and time series analysis techniques are used to detect abnormal patterns in time series data and generate an analysis report of abnormal climate events. Based on the aforementioned abnormal climate event analysis report, a fully convolutional network is used to perform semantic segmentation of the image for land surface classification, generating a land surface type classification map; The comprehensive climate feature dataset includes temperature, humidity, and cloud cover distribution data. The image quality optimization results specifically refer to remote sensing images with improved illumination, contrast, and sharpness. The complex pattern recognition data includes geographical features and climate change data. The abnormal climate event analysis report specifically refers to unconventional climate and potential environmental risks. The land surface type classification map includes remote sensing images of cities, farmland, forests, and multiple land surface covers.

2. The remote sensing image classification method according to claim 1, characterized in that, Based on raw remote sensing data, satellite cloud image analysis and surface temperature mapping techniques are used to conduct a comprehensive analysis of microclimate characteristics. K-means clustering is then used for data classification to generate a comprehensive climate characteristic dataset. The specific steps are as follows: Based on the original remote sensing data, Gaussian filtering is used to denoise the image and enhance it to generate preprocessed remote sensing data. Based on the preprocessed remote sensing data, cloud information data is extracted and generated using the MODIS cloud detection algorithm. Based on the cloud information data, the MODTRAN algorithm is used to obtain the surface temperature distribution and generate a surface temperature distribution map. Based on the surface temperature distribution map, cloud maps and temperature data are integrated using a data assimilation method to generate fused climate data. Based on the fused climate data, the K-means clustering algorithm is used to divide climate regions and generate climate region classification data. Based on the climate region classification data, a multi-dimensional data fusion method was used to integrate the data, and the climate characteristics were extracted by principal component analysis to generate a comprehensive climate characteristic dataset.

3. The remote sensing image classification method according to claim 2, characterized in that, Based on the comprehensive climate feature dataset, the image quality is optimized using a contrast-limited adaptive histogram equalization method and convolutional neural network technology. The specific steps for generating the image quality optimization results are as follows: Based on the comprehensive climate feature dataset, the image contrast is optimized using a contrast-limited adaptive histogram equalization technique to generate contrast-optimized data. Based on the contrast optimization data, a VGG-16 convolutional neural network model is used for feature enhancement to generate feature-enhanced data. Based on the feature enhancement data, median filtering technology is used to perform image denoising to generate denoised image data. Based on the denoised image data, USM sharpening technology is used to improve image clarity and generate sharpened image data; Based on the sharpened image data, a grayscale world algorithm is used to perform color balancing and generate color correction data. Based on the color correction data, the structural similarity index algorithm is used to evaluate image quality and generate image quality optimization results.

4. The remote sensing image classification method according to claim 3, characterized in that, Based on the image quality optimization results, the fractal dimension calculation method is used to identify and analyze geographic patterns in remote sensing images, and the specific steps for generating complex pattern recognition data are as follows: Based on the image quality optimization results, the two-dimensional discrete wavelet transform technique is used to extract multi-scale features of the image and generate multi-scale feature data. Based on the multi-scale feature data, the random forest algorithm is used for feature selection and geographic pattern recognition to generate geographic pattern data. Based on the aforementioned geographic pattern data, the fractal dimension is calculated using the box counting method to generate fractal dimension analysis data. Based on the fractal dimension analysis data, the Euclidean distance algorithm is used to perform similarity pattern recognition and generate pattern similarity data. Based on the pattern similarity data, a support vector machine method is used to classify geographic patterns and generate geographic pattern classification data. Based on the aforementioned geographic pattern classification data, artificial neural networks and analysis of variance methods are used for pattern recognition and deep analysis to generate complex pattern recognition data.

5. The remote sensing image classification method according to claim 4, characterized in that, Based on the complex pattern recognition data, the steps for detecting anomalous patterns in time series data and generating an anomalous climate event analysis report using long short-term memory networks and time series analysis techniques are as follows: Based on the complex pattern recognition data, Z-score normalization is used to process the data and generate time series data. Based on the time series data, an autoregressive moving average model is used to decompose the series and generate time series decomposed data. Based on the time series decomposed data, a time series prediction model is generated using a long short-term memory network. Based on the time series prediction model, the statistical threshold method is used to detect abnormal patterns and generate preliminary abnormal pattern detection results. Based on the preliminary abnormal pattern detection results, the DBSCAN clustering algorithm is used to filter and verify the abnormal patterns, generating abnormal pattern data. Based on the aforementioned anomaly pattern data, multiple regression analysis and the ARIMA model are used to predict future climate anomalies and generate an anomaly event analysis report.

6. The remote sensing image classification method according to claim 5, characterized in that, Based on the aforementioned abnormal climate event analysis report, the specific steps for generating a land surface type classification map by using a fully convolutional network for semantic segmentation of the image are as follows: Based on the aforementioned abnormal climate event analysis report, a text processing algorithm is used to extract and convert data to generate an image dataset with adaptive semantic segmentation. Based on the image dataset of the adaptive semantic segmentation, gamma correction and edge enhancement techniques are used to optimize the images and generate optimized image data. Based on the optimized image data, a fully convolutional network model is used for learning and training to generate an FCN training model. Based on the FCN training model, pixel-by-pixel classification technology is used to perform semantic segmentation of the image and generate preliminary semantic segmentation results. Based on the preliminary semantic segmentation results, morphological operations are used for image processing to generate semantic segmentation data. Based on the semantic segmentation data, GIS technology is used to classify and spatially analyze land surface types, generating a land surface type classification map.

7. A remote sensing image classification system, characterized in that, The remote sensing image classification system is used to execute the remote sensing image classification method according to any one of claims 1 to 6. The system includes a data preprocessing module, a climate feature analysis module, an image quality optimization module, a geographic pattern recognition module, an abnormal climate analysis module, a land surface classification module, a meteorological data integration module, and an environmental impact assessment module. The data preprocessing module uses Gaussian filtering noise reduction algorithm and histogram equalization technology to enhance the image based on the original remote sensing data, and generates preprocessed remote sensing data. The climate feature analysis module analyzes microclimate features based on preprocessed remote sensing data, using the MODIS cloud detection algorithm, MODTRAN algorithm and data assimilation method to generate a comprehensive climate feature dataset. The image quality optimization module is based on a comprehensive climate feature dataset and uses CLAHE technology and VGG-16 network to optimize image quality and generate image quality optimization results. The geographic pattern recognition module uses two-dimensional discrete wavelet transform and random forest algorithm to perform geographic pattern recognition based on image quality optimization results, generating complex pattern recognition data. The abnormal climate analysis module analyzes abnormal climate events based on complex pattern recognition data, using long short-term memory networks and ARIMA models, and generates an abnormal climate event analysis report. The land surface classification module is based on the abnormal climate event analysis report. It uses a fully convolutional network model and GIS technology to perform semantic segmentation and land surface classification on the image, and generates a land surface type classification map. The meteorological data integration module integrates meteorological data based on the land surface type classification map, using data fusion algorithms and spatial interpolation techniques to generate a meteorological information dataset. The environmental impact assessment module is based on meteorological information datasets and uses environmental models and impact assessment algorithms to analyze the impact of climate change on the environment and generate an environmental impact assessment report.

8. The remote sensing image classification system according to claim 7, characterized in that, The Gaussian filtering denoising algorithm specifically uses Gaussian kernels to reduce image noise; the MODIS cloud detection algorithm is a remote sensing analysis method for identifying cloud layers; the MODTRAN algorithm is a surface temperature inversion based on an atmospheric model; the random forest algorithm uses multiple decision trees for feature evaluation and classification; the ARIMA model refers to an autoregressive integral moving average method used for time series analysis and prediction; the GIS technology is based on spatial analysis and map-making methods; the environmental model refers to a computational model used to simulate the impacts of climate change on ecosystems and human activities; and the impact assessment algorithm includes assessing the potential impacts of climate change on biodiversity, water resources, and agricultural output.

9. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program, and when the processor executes the computer program, it implements the remote sensing image classification system according to any one of claims 7 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the remote sensing image classification method according to any one of claims 1 to 6.

Images