Hyperspectral data fusion analysis system

Abstract

The invention discloses a hyperspectral data fusion analysis system, belongs to the technical field of remote sensing data processing and artificial intelligence, and provides a multi-level data fusion technology by integrating data management, preprocessing, fusion, modeling and visual function modules through a modular architecture aiming at the problems of an existing system. Machine learning algorithms such as a support vector machine and a random forest are integrated to realize dynamic parameter optimization and model evaluation, spectrum curve dynamic display, classification map rendering and customized analysis report generation functions are provided through an interactive visualization module, and system security and expansibility are enhanced based on hierarchical architecture and database standard design. And multi-source equipment access, user permission grading and data encryption storage are supported. The hyperspectral data processing efficiency, the analysis precision and the system stability are remarkably improved, and the method can be widely applied to the fields of environment monitoring, agricultural remote sensing, target recognition and the like.

Description

Technical Field

[0001] The present invention relates to remote sensing data processing and artificial intelligence technology, and in particular to a hyperspectral data fusion analysis system. Background Art

[0002] Hyperspectral remote sensing technology, by acquiring spectral information of ground objects in continuous narrow bands, can achieve precise identification and quantitative analysis of target substances, and has important application value in environmental monitoring, agricultural remote sensing, target identification, and other fields. However, existing hyperspectral data analysis systems generally face the following challenges:

[0003] Low data processing efficiency: Hyperspectral data is large in volume and high in dimension. Traditional preprocessing methods (such as denoising and geometric correction) rely on manual intervention, making it difficult to achieve batch automated processing. In addition, there is a lack of unified standardized processes, resulting in poor data compatibility.

[0004] Single fusion method: Existing systems mostly adopt a single-level (such as feature level or decision level) fusion strategy, which makes it difficult to take into account the complementarity of spectral and spatial features, and lacks support for multi-algorithm collaboration (such as principal component analysis and gray-level co-occurrence matrix), limiting the flexibility and accuracy of the analysis.

[0005] Insufficient modeling and visualization: Classification algorithms (such as support vector machines and random forests) have low integration, model parameter optimization relies on empirical adjustment, and there is a lack of real-time feedback of dynamic evaluation indicators (such as F1 value and recall rate); at the same time, analysis results are mostly presented in static charts, which make it difficult to intuitively display spatial distribution characteristics.

[0006] Weak system scalability and security: Existing platforms mostly adopt a loose architecture design with high coupling between modules, making it difficult to adapt to the needs of multi-source data access; database naming and field design lack unified standards, resulting in data redundancy and security risks.

[0007] In order to solve the above problems, the applicant proposes a hyperspectral data fusion analysis system. Summary of the Invention

[0008] The purpose of the present invention is to provide a hyperspectral data fusion analysis system to solve the problems in the prior art.

[0009] To achieve the above objectives, the present invention provides the following technical solutions: a hyperspectral data fusion analysis system, characterized by comprising the following modules:

[0010] Data management module, which supports uploading hyperspectral data in various formats, metadata-based data query and data deletion functions;

[0011] The data preprocessing module includes an image preprocessing unit, a geometric correction unit, and a spectral smoothing unit. The image preprocessing unit is used to denoise and crop the hyperspectral data. The geometric correction unit is used to eliminate geometric distortion of the image. The spectral smoothing unit uses a Savitzky-Golay filter to smooth the spectral curve.

[0012] The data fusion module includes a feature-level fusion unit and a decision-level fusion unit. The feature-level fusion unit is used to extract and fuse the features of the visible spectrum and the near-infrared spectrum. The decision-level fusion unit is used to classify or identify the visible spectrum and the near-infrared spectrum respectively and then fuse the decision results.

[0013] An analysis and modeling module includes a classification analysis unit, an identification analysis unit, and a quantitative analysis unit. The classification analysis unit uses a support vector machine or a random forest algorithm to classify the fused data. The quantitative analysis unit is used to establish a quantitative relationship between the spectrum and the target parameter.

[0014] The result display module is used to display spectral curves in the form of charts, visualize classification results in the form of maps, and generate reports containing data processing processes and analysis results.

[0015] Optionally, the data preprocessing module further includes:

[0016] Radiation correction unit, used to perform radiation correction on uploaded hyperspectral data;

[0017] The spectral normalization unit is used to unify spectral data from different sources into a standardized range.

[0018] Optionally, the data fusion module also supports the following functions:

[0019] Provides spectral feature extraction algorithms for principal component analysis and linear discriminant analysis;

[0020] Provides gray-level co-occurrence matrix and local binary pattern image feature extraction algorithms;

[0021] Supports visual analysis of feature layer fusion and decision layer fusion.

[0022] Optionally, the analysis and modeling module further includes:

[0023] The model evaluation and optimization unit is used to evaluate model performance through accuracy, recall, and F1 value indicators, and supports model parameter optimization.

[0024] Optionally, the system further comprises:

[0025] User management module, used by administrators to classify and register user permissions;

[0026] System login module, enter the main operation interface after user name and password authentication;

[0027] The data export and import module supports standardized conversion of batch data and cross-platform transmission.

[0028] Optionally, the database design of the system complies with the following specifications:

[0029] Table names and field names must use lowercase letters and underscores. Database keywords are prohibited.

[0030] The database table structure includes user table, spectral data table, preprocessing parameter table and analysis result table.

[0031] Optionally, the system runs on a Windows operating system, is connected to external devices via an RS232 interface, and has a built-in multi-protocol communication function to be compatible with different acquisition instruments.

[0032] A hyperspectral data fusion analysis method includes the following steps:

[0033] Obtain visible spectrum and near-infrared spectrum data through hyperspectral acquisition equipment;

[0034] Upload the data to the system and perform radiation correction, geometric correction and spectral smoothing preprocessing;

[0035] Select feature-level fusion or decision-level fusion method to fuse the preprocessed data;

[0036] Use classification algorithms to classify and analyze the fused data and establish a quantitative relationship between the spectrum and the target parameters;

[0037] Generate visualization results including spectral curves, classification maps, and analysis reports.

[0038] Optionally, the spectrum smoothing process uses a Savitzky-Golay filter, and the classification algorithm includes a support vector machine or a random forest algorithm.

[0039] Optionally, the method further comprises:

[0040] Dynamically evaluate model performance and improve classification accuracy through parameter optimization;

[0041] Export analysis results in a standard format to support cross-platform sharing and application.

[0042] Beneficial effect: efficient data processing and automation capabilities

[0043] It supports batch data import, conversion, and export, and combines the automated pre-processing process of radiation correction, geometric correction, and spectral smoothing to significantly reduce manual intervention and improve the processing efficiency of hyperspectral data.

[0044] Database connection pool management and secondary data caching technology are used to optimize data access speed and ensure efficient storage and call of massive data.

[0045] Multi-level integration and precise analysis

[0046] By integrating the dual strategies of feature level (such as principal component analysis and gray-level co-occurrence matrix) and decision level (classification result optimization), and combining the complementarity of visible spectrum and near-infrared spectrum, the comprehensiveness of data representation and analysis accuracy are improved.

[0047] Integrate machine learning algorithms such as support vector machine (SVM) and random forest, optimize model parameters in real time through dynamic evaluation indicators (accuracy, F1 value), and enhance the reliability of classification and recognition results.

[0048] Interactive visualization and interpretability of results

[0049] It provides dynamic display of spectral curves, classification result map rendering and interactive report generation functions, intuitively presenting data characteristics and spatial distribution, lowering professional barriers and improving user decision-making efficiency.

[0050] It supports visual analysis of multiple image features (such as LBP, GLCM) and spectral features, helping users quickly understand the inherent laws of the data.

[0051] System scalability and security assurance

[0052] It adopts a layered architecture design (presentation layer, business logic layer, data access layer), with low coupling and high cohesion between modules, which facilitates functional expansion and multi-source data access.

[0053] Follow strict database naming conventions (lowercase letters + underscore combination), design multi-table structures such as user tables and spectral data tables, and combine user permission hierarchical management to effectively avoid data redundancy and unauthorized access.

[0054] It runs on Windows system, has built-in multi-protocol communication function and RS232 interface, and is compatible with various acquisition devices, ensuring system stability and cross-platform applicability.

[0055] Wide application scenarios and economic efficiency

[0056] Through the quantitative analysis module, a mathematical model of spectrum and target parameters (such as soil composition and vegetation coverage) is established to provide low-cost, high-precision solutions for environmental monitoring, precision agriculture, target identification and other fields.

[0057] Reduce the dependence of traditional hyperspectral analysis on professional software, lower the threshold for technology application and operation and maintenance costs, and promote technology accessibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is the ER diagram of the system according to the embodiment of the present invention;

[0059] Figure 2 It is a system class diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The following describes preferred embodiments of the present invention with reference to the accompanying drawings to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0061] System architecture design

[0062] This system adopts a layered architecture design and is divided into the following four layers:

[0063] Presentation layer: responsible for user interface display and interaction, using Web 2.0 technologies (such as Ajax and jQuery) to optimize user experience, supporting dynamic rendering of spectral curves, classification result map display and interactive report generation.

[0064] Business logic layer: Implements core business functions, including data preprocessing, fusion algorithm calls, model training and evaluation, etc., adopts LOB (Line-of-Business) and MDD (Model-Driven Development) design concepts to improve functional reusability.

[0065] Data access layer: manages database interactions, supports database connection pools, transaction management, and secondary cache technology to ensure efficient and secure data access.

[0066] Service access layer: Integrates external service interfaces (such as device communication protocols and third-party algorithm libraries) to support multi-source data access and cross-platform compatibility.

[0067] Technology selection and platform support

[0068] Operating system: runs on Windows system, compatible with Windows 10 and above, ensuring system stability and security.

[0069] Development language: Java / Python is used for the backend, HTML5 / CSS3 / JavaScript is used for the front end, and MySQL / PostgreSQL is used for the database.

[0070] Hardware support: Supports connection to hyperspectral acquisition equipment via RS232 interface, built-in multi-protocol communication function (such as TCP / IP, HTTP), and is compatible with mainstream spectrometers and sensors.

[0071] Overview of system functional modules

[0072] The system's core functional modules include:

[0073] Data management module: supports data upload, query, deletion and batch import / export.

[0074] Data preprocessing module: covers radiation correction, geometric correction, spectral smoothing and other functions.

[0075] Data fusion module: realizes feature-level and decision-level fusion, integrating PCA, LBP and other algorithms.

[0076] Analysis and modeling module: built-in classification algorithms such as SVM and random forest, supporting model evaluation and optimization.

[0077] Result display module: provides visual output of spectral curves, classification maps and analysis reports.

[0078] User management module: implements user authority classification and login authentication.

[0079] Detailed design of functional modules

[0080] Data Management Module

[0081] Data upload: supports uploading hyperspectral data in ENVI, HDF5, CSV and other formats, with a maximum single file size of 10GB.

[0082] Data query: fuzzy query and screening based on metadata (such as acquisition time, geographic location, spectral range).

[0083] Data operations: supports batch deletion, export to standardized formats (such as GeoTIFF), and provides data version management functions.

[0084] Use distributed file storage systems (such as HDFS) to manage massive data, and combine with Elasticsearch to achieve fast metadata retrieval.

[0085] Use the Apache POI library to process Excel format data and the GDAL library to parse geospatial data.

[0086] Data preprocessing module

[0087] Radiation correction: Eliminate the effects of atmospheric scattering based on radiation transfer models (such as MODTRAN).

[0088] Geometric correction: Use polynomial correction method or RPC (Rational Polynomial Coefficients) model to eliminate image distortion.

[0089] Spectral smoothing: Apply Savitzky-Golay filter to smooth the spectral curve, with adjustable window size.

[0090] Radiation correction is implemented by calling the ENVI / IDL algorithm library; geometric correction relies on OpenCV's image transformation function.

[0091] Spectral smoothing is implemented using Python's SciPy library and supports user-defined parameter configuration.

[0092] Data fusion module

[0093] Feature-level fusion: Extract PCA features and LBP texture features of visible light and near-infrared spectra, and generate a comprehensive feature vector through weighted fusion.

[0094] Decision-level fusion: DS evidence theory fusion is performed on multi-source classification results (such as SVM classification probability) to improve target recognition accuracy.

[0095] Visual analysis: Provides interactive charts such as feature distribution heat maps and decision confidence curves.

[0096] Feature extraction is implemented using the PCA module of Scikit-learn and LBP of OpenCV; the fusion algorithm is developed based on custom Python logic.

[0097] The visualization uses ECharts and D3.js libraries, supporting dynamic data binding and multi-dimensional interaction.

[0098] Analysis and Modeling Module

[0099] Classification analysis: supports SVM (kernel function can be linear / RBF), random forest (number of trees can be adjusted), and other algorithms.

[0100] Quantitative analysis: Establish a PLS (partial least squares regression) model to predict parameters such as soil moisture content and vegetation index.

[0101] Model evaluation: Provides indicators such as confusion matrix, ROC curve, F1 value, and supports cross-validation and hyperparameter grid search optimization.

[0102] Technical implementation:

[0103] The classification and regression models are implemented based on Scikit-learn and TensorFlow; model evaluation uses MLflow to track experimental records.

[0104] Users can adjust parameters through the web interface, and the system will provide real-time feedback on training progress and evaluation results.

[0105] Results display module

[0106] Spectral curve display: Dynamically draw spectral reflectance curves, supporting band selection and local zoom.

[0107] Classification map rendering: overlay the classification results onto the geographic base map and use color coding to distinguish the types of land features.

[0108] Report generation: Automatically generate PDF / HTML format reports including data processing flow, model parameters and result summary.

[0109] Map rendering uses Leaflet.js and the GeoJSON data format; report generation relies on the Jinja2 template engine and the WeasyPrint library.

[0110] User management module

[0111] Permission classification: divided into three levels: administrator, ordinary user, and guest. Administrators can assign data access and operation permissions.

[0112] Login authentication: Adopts OAuth 2.0 protocol and supports two-factor authentication (SMS / email verification code).

[0113] User data is stored in a MySQL database, and permission control is implemented through the Spring Security framework.

[0114] Data processing flow design

[0115] Data collection and upload

[0116] The visible and near-infrared band data are acquired through a hyperspectral imager (such as the HySpex VNIR-1800).

[0117] The data is uploaded to the system storage server via the RS232 interface or FTP protocol, and the metadata is automatically extracted and entered into the database.

[0118] Data preprocessing process

[0119] Radiation correction: eliminate atmospheric scattering and sensor noise, and output apparent reflectance data.

[0120] Geometric correction: Correct image geometric distortion based on ground control points (GCP) or DEM data.

[0121] Spectral smoothing: Apply Savitzky-Golay filter to remove high-frequency noise and enhance the separability of spectral features.

[0122] Data fusion and analysis

[0123] Feature-level fusion: Extract spectral PCA features and image LBP features, and perform weighted fusion to generate a comprehensive feature set.

[0124] Decision-level fusion: Confidence-weighted multi-classifier results are outputted as the final classification label.

[0125] Modeling and Optimization: Training SVM / Random Forest models and optimizing hyperparameters through grid search.

[0126] Result visualization and output

[0127] The classification results are rendered as geographic maps, supporting layer overlay and transparency adjustment.

[0128] Analysis reports are automatically generated, and users can customize templates and output formats.

[0129] Database design

[0130] Database ER diagram

[0131] The system database contains the following core tables:

[0132] User table (user): stores username, password hash, permission level, etc.

[0133] Spectral data table (spectral_data): records data ID, acquisition time, geographic location, file path, etc.

[0134] Preprocessing parameter table (preprocess_params): saves radiation correction coefficients, geometric correction model parameters, etc.

[0135] Analysis result table (analysis_result): stores classification labels, confidence levels, quantitative prediction values, etc.

[0136] Table structure design

[0137] Field naming conventions: Use a combination of lowercase letters and underscores (such as user_name), and do not use SQL keywords.

[0138] Index optimization: Create B+ tree indexes for high-frequency query fields (such as spectral_data.collection_time).

[0139] Data security policy

[0140] Encrypted storage: Sensitive data (such as user passwords) is encrypted using AES-256.

[0141] Backup mechanism: Daily full backup + incremental backup, stored in an off-site disaster recovery server.

[0142] System security and scalability

[0143] Security Design

[0144] Permission control: Based on the RBAC (role-based access control) model, the scope of user operations is limited.

[0145] Log audit: records user login, data operation and other behaviors, and supports abnormal behavior alarms.

[0146] Network security: Data is transmitted via the HTTPS protocol, and firewall rules restrict unauthorized IP access.

[0147] Scalability design

[0148] Modular architecture: Each functional module communicates through RESTful API and supports independent upgrade and replacement.

[0149] Plug-in mechanism: allows third parties to develop data preprocessing plug-ins or fusion algorithms and integrate them through standardized interfaces.

[0150] Login interface

[0151] Enter your username and password. Verification code and password reset functions are supported.

[0152] The interface style is simple and conforms to the Material Design design specifications.

[0153] Main operation interface

[0154] The left navigation bar contains the entrance to the functional modules, and the right side is the data visualization area.

[0155] Provides drag-and-drop data upload, parameter configuration panel and real-time progress prompts.

[0156] Analysis report interface

[0157] Report preview, editing, and export are supported, and charts are interactive (e.g. click to view details).

[0158] System testing and deployment

[0159] Unit testing: Write test cases for core algorithms (such as Savitzky-Golay filtering and SVM classification).

[0160] Integration testing: simulates multi-user concurrent operations to verify system stability and performance.

[0161] User acceptance testing: Invite domain experts to try out the product, collect feedback, and optimize features.

[0162] Hardware configuration: Intel Xeon server (64GB RAM + 1TB SSD) is recommended, GPU acceleration is optional.

[0163] Software dependencies: Pre-installed Java Runtime, Python 3.8, MySQL 8.0, and necessary library files.

[0164] This technical solution, through modular architecture design, multi-level data fusion, and intelligent analysis modeling, has built an efficient and accurate hyperspectral data fusion and analysis system. Future work will explore the application of deep learning algorithms (such as convolutional neural networks) in feature extraction and expand support for multimodal data (such as LiDAR), promoting the implementation of hyperspectral technology in a wider range of industrial scenarios.

[0165] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, from all points of view, the embodiments should be regarded as illustrative and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes that fall within the meaning and range of equivalents of the claims are included in the present invention. Any reference signs in the claims should not be construed as limiting the claim to which they relate.

[0166] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A hyperspectral data fusion analysis system, characterized in that: Includes the following modules: Data management module, which supports uploading hyperspectral data in various formats, metadata-based data query and data deletion functions; The data preprocessing module includes an image preprocessing unit, a geometric correction unit, and a spectral smoothing unit. The image preprocessing unit is used to denoise and crop the hyperspectral data. The geometric correction unit is used to eliminate geometric distortion of the image. The spectral smoothing unit uses a Savitzky-Golay filter to smooth the spectral curve. The data fusion module includes a feature-level fusion unit and a decision-level fusion unit. The feature-level fusion unit is used to extract and fuse the features of the visible spectrum and the near-infrared spectrum. The decision-level fusion unit is used to classify or identify the visible spectrum and the near-infrared spectrum respectively and then fuse the decision results. An analysis and modeling module includes a classification analysis unit, an identification analysis unit, and a quantitative analysis unit. The classification analysis unit uses a support vector machine or a random forest algorithm to classify the fused data. The quantitative analysis unit is used to establish a quantitative relationship between the spectrum and the target parameter. The result display module is used to display spectral curves in the form of charts, visualize classification results in the form of maps, and generate reports containing data processing processes and analysis results.

2. The system according to claim 1, wherein: The data preprocessing module further includes: Radiation correction unit, used to perform radiation correction on uploaded hyperspectral data; The spectral normalization unit is used to unify spectral data from different sources into a standardized range.

3. The system according to claim 1, wherein: The data fusion module also supports the following functions: Provides spectral feature extraction algorithms for principal component analysis and linear discriminant analysis; Provides gray-level co-occurrence matrix and local binary pattern image feature extraction algorithms; Supports visual analysis of feature layer fusion and decision layer fusion.

4. The system according to claim 1, wherein: The analysis and modeling module also includes: The model evaluation and optimization unit is used to evaluate model performance through accuracy, recall, and F1 value indicators, and supports model parameter optimization.

5. The system according to claim 1, wherein: The system further comprises: User management module, used by administrators to classify and register user permissions; System login module, enter the main operation interface after user name and password authentication; The data export and import module supports standardized conversion of batch data and cross-platform transmission.

6. The system according to claim 1, wherein: The database design of the system follows the following specifications: Table names and field names must use lowercase letters and underscores. Database keywords are prohibited. The database table structure includes user table, spectral data table, preprocessing parameter table and analysis result table.

7. The system according to claim 1, wherein: The system runs on the Windows operating system, connects to external devices via an RS232 interface, and has a built-in multi-protocol communication function to be compatible with different acquisition instruments.

8. A hyperspectral data fusion analysis method, characterized in that: The following steps are involved: Obtain visible spectrum and near-infrared spectrum data through hyperspectral acquisition equipment; Upload the data to the system and perform radiation correction, geometric correction and spectral smoothing preprocessing; Select feature-level fusion or decision-level fusion method to fuse the preprocessed data; Use classification algorithms to classify and analyze the fused data and establish a quantitative relationship between the spectrum and the target parameters; Generate visualization results including spectral curves, classification maps, and analysis reports.

9. The method according to claim 8, characterized in that The spectrum smoothing process uses a Savitzky-Golay filter, and the classification algorithm includes a support vector machine or a random forest algorithm.

10. The method according to claim 8, characterized in that The method further comprises: Dynamically evaluate model performance and improve classification accuracy through parameter optimization; Export analysis results in a standard format to support cross-platform sharing and application.

Images