Task-based hyperspectral quality assessment method driven by reliability of mineral identification

Abstract

The application provides a mineral identification reliability driven task type hyperspectral quality evaluation method, relates to the technical field of hyperspectral remote sensing, and comprises the following steps: initializing P target evaluation mineral evaluation task containers; performing spatial index segmentation on a hyperspectral data cube to obtain P spatial data subsets; performing mineral exclusive quality analysis process isolation to output P mineral reliability feature sets; calculating P mineral identification reliability indexes; mapping and extracting P pixel level local signal-to-noise ratios and P pixel level absorption depth stabilities; performing reliability partitioning to output P reliability grading mapping results; and outputting a global mineral identification reliability heat map. The application solves the technical problem in the prior art that the hyperspectral data quality evaluation method focuses on using global image quality indexes for evaluation and cannot reflect key diagnostic bands relied on by specific mineral identification, thereby leading to unreliable subsequent mineral identification results.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral remote sensing technology, specifically to a mission-oriented hyperspectral quality assessment method driven by mineral identification reliability. Background Technology

[0002] Mineral identification relies primarily on the absorption and reflection characteristics of individual pixels in hyperspectral images within specific spectral bands. These characteristic bands typically exhibit unique responses to different minerals. For instance, iron oxides have absorption characteristics in the 850-1000 nm band, clay minerals have an absorption valley around 2200 nm, and carbonate minerals have absorption characteristics around 2300-2350 nm. Therefore, the accuracy of mineral identification is highly dependent on the quality of hyperspectral data in these characteristic bands.

[0003] In existing technologies, hyperspectral data quality assessment focuses on using global image quality metrics, such as signal-to-noise ratio (SNR) and radiometric accuracy. These metrics often quantify the quality of the entire image, but they cannot reflect the key diagnostic bands on which specific mineral identification depends. For example, although two images may be similar in global SNR, they may differ in quality in the key bands of a specific mineral. This global quality assessment method cannot effectively reveal the actual reliability of mineral identification. Users cannot quickly and quantitatively determine the applicability of the data to the target mineral identification task before data processing, resulting in unreliable subsequent mineral identification results. Furthermore, it is necessary to rely on experience-based judgment or verification after a complete processing flow, which increases time and computational costs. Summary of the Invention

[0004] To address the aforementioned deficiencies or improvement needs of existing technologies, this application provides a task-oriented hyperspectral quality assessment method driven by the reliability of mineral identification. This method addresses the technical problem that existing hyperspectral data quality assessment methods focus on using global image quality indicators for evaluation, failing to reflect the key diagnostic bands upon which specific mineral identification depends, thus leading to unreliable subsequent mineral identification results.

[0005] This application provides a task-oriented hyperspectral quality assessment method driven by mineral identification reliability, the method comprising:

[0006] Initialize P evaluation task containers for P target minerals based on the mineral types to be detected; perform spatial indexing on the hyperspectral data cube to obtain P spatial data subsets; load the P spatial data subsets in parallel into the P evaluation task containers and perform mineral-specific quality analysis processing in isolation to output P mineral reliability feature sets; calculate P mineral identification reliability indices based on the P mineral reliability feature sets; using the P mineral identification reliability indices as a grading benchmark, extract P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stability from the cache data areas of the P evaluation task containers; perform reliability partitioning based on the P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stability to output P reliability grading mapping results; output a global mineral identification reliability heatmap by spatially stitching the P reliability grading mapping results.

[0007] In one implementation, initializing P evaluation task containers for P target evaluation minerals based on the mineral types to be detected includes:

[0008] Based on the P target evaluation minerals constituting the mineral type to be detected, P benchmark task containers are retrieved from the task container library; P exclusive spectral parameter sets for the P target evaluation minerals are retrieved from the preset spectral library; the P exclusive spectral parameter sets are mapped and loaded into the P benchmark task containers for dynamic parameter adjustment and correction, thus completing the initialization of the P evaluation task containers.

[0009] In one implementation, spatial indexing is performed on the hyperspectral data cube to obtain P spatial data subsets, including:

[0010] Based on the number of computing nodes P, the hyperspectral data cube is spatially divided into P rectangular sub-regions; spatial neighborhood image data is retrieved from the hyperspectral data cube to perform buffer compensation for the P rectangular sub-regions, resulting in P dynamic buffers; the P dynamic buffers are then mapped back to the P rectangular sub-regions to obtain P spatial buffer compensation subsets; after performing two-dimensional georegistration and binding of spatial coordinate system and projection parameters on the P dynamic buffer compensation subsets, data BIL format reconstruction is performed to generate the P spatial data subsets.

[0011] In one implementation, the P dedicated spectral parameter sets are mapped and loaded into the P reference task containers for dynamic parameter tuning and calibration, completing the initialization of the P evaluation task containers, including:

[0012] The first dedicated spectral parameter set is analyzed to obtain the first diagnostic band parameters, the first feature quantization parameters, and the first spatial association settings. A first mineral feature window is dynamically generated based on the first diagnostic band parameters. A first signal-to-noise ratio calculation template is loaded according to the first feature quantization parameters in an adaptive configuration. The first endmember extraction sensitive parameters are retrieved in a targeted manner according to the first spatial association settings. The first mineral feature window, the first signal-to-noise ratio calculation template, and the first endmember extraction sensitive parameters are injected into the first benchmark task container to perform hardware acceleration parameter compilation, thereby completing the initialization of the first evaluation task container.

[0013] In one implementation, the P subsets of spatial data are loaded in parallel into the P evaluation task containers for isolated execution of a mineral-specific quality analysis process, outputting P sets of mineral reliability features, including:

[0014] Within the first mineral feature window, the first signal-to-noise ratio (SNR) calculation template is used to perform local SNR calculations on a pixel-by-pixel basis on the first spatial data subset, outputting a first pixel-level SNR set; the mean SNR of the first pixel-level SNR set is calculated; after performing continuum removal on the first spatial data subset, absorption depth stability analysis is performed based on the reflectance data of the first spatial data subset within the first mineral feature window to obtain a first absorption depth standard deviation; within the first spatial data subset, the first endmember extraction sensitive parameters are used to perform endmember spectral separability calculations to generate a first spectral confusion index; the first window identifier of the first mineral feature window is called, and by integrating the first window identifier, the mean SNR of the first pixel-level SNR, the first absorption depth standard deviation, and the first spectral confusion index, a first mineral reliability feature set is output.

[0015] In one implementation, calculating P mineral identification reliability indices based on the P mineral reliability feature sets includes:

[0016] Based on the first window identifier, the piecewise linear normalization threshold is retrieved, and adaptive piecewise linear normalization is performed on the first pixel-level signal-to-noise ratio mean, outputting the first normalized signal-to-noise ratio mean; the first absorption depth standard deviation is linearly normalized after inverse transformation, outputting the first normalized absorption depth stability; based on the first window identifier, the first weight vector is retrieved, and the first weight vector is used to perform weighted integration of the first normalized signal-to-noise ratio mean, the first normalized absorption depth stability, and the first spectral confusion index, outputting the first benchmark reliability index; the regional geological prior probability and sensor band attenuation factor are fused, and the confidence level of the first benchmark reliability index is calibrated, outputting the first mineral identification reliability index.

[0017] In one implementation, using the P mineral identification reliability indices as a grading benchmark, P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stability are extracted from the cache data areas of the P evaluation task containers, including:

[0018] Based on the P mineral identification reliability indices, multi-level accuracy grading is performed to obtain P grading accuracy strategies; the memory mapping addresses of the P evaluation task containers are parsed to locate P cell-level local signal-to-noise ratio buffers and P cell-level absorption depth stability buffers; using the P grading accuracy strategies, the P cell-level local signal-to-noise ratio and P cell-level absorption depth stability are extracted in parallel mapping within the P cell-level local signal-to-noise ratio buffers and P cell-level absorption depth stability buffers.

[0019] In one implementation, reliability partitioning is performed based on the P pixel-level local signal-to-noise ratios and the P pixel-level absorption depth stability, outputting P reliability classification mapping results, including:

[0020] Based on the P window identifiers of the P evaluation task containers, P mineral-specific normalized parameters are retrieved; the P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stability are normalized into P normalized signal-to-noise ratio sets and P normalized absorption depth stability sets using the P mineral-specific normalized parameters; digital elevation model data is loaded to perform terrain slope correction on the P normalized signal-to-noise ratio sets, resulting in P terrain-corrected signal-to-noise ratio sets; regional geological map data is called to perform mineral boundary buffer degradation processing on the P normalized absorption depth stability sets, resulting in P mineral degradation... The system first obtains a set of absorption depths at each of the P levels; then, based on the P window identifiers, it retrieves P mineral association weight ratios and constructs a P-pixel-level comprehensive reliability weight model using the P terrain-corrected signal-to-noise ratio sets, P mineral downgrade absorption depth sets, and P mineral association weight ratios; finally, it performs reliability level classification on the P-pixel-level comprehensive reliability weight model according to a preset mineral-specific classification threshold table to obtain P reliability level data; and then, after performing morphological closing operations on the P reliability level data to eliminate isolated pixels, it performs partition boundary smoothing processing to obtain the P reliability classification mapping results.

[0021] In one implementation, spatial boundary markers are monitored during the data extraction process, and a dynamic buffer compensation mechanism is triggered when a spatial subset boundary cell is detected.

[0022] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0023] Based on the type of mineral to be detected, an evaluation task container for the corresponding target mineral is initialized. Each evaluation task container is specifically designed for a particular mineral type, ensuring a highly customized and targeted evaluation process. The hyperspectral data cube is divided into multiple spatial data subsets through spatial indexing. This operation ensures the parallelism and efficiency of the processing in the spatial dimension, while optimizing the local processing capabilities of the data. The P partitioned spatial data subsets are then loaded in parallel into each evaluation task container. Within each container, a mineral-specific quality analysis process is executed, outputting its respective mineral reliability feature set. By processing the quality analysis of each mineral in parallel, the overall evaluation process time is significantly shortened. In particular, it improves efficiency in processing large-scale hyperspectral data. By using the mineral identification reliability index as a grading benchmark, pixel-level local signal-to-noise ratio and absorption depth stability are extracted from the cached data area of ​​the task container. This process ensures that the quality assessment of minerals in different regions can be accurately mapped to local data, thereby effectively capturing the details of mineral features in space. Based on the extracted local signal-to-noise ratio and absorption depth stability data, reliability partitioning is performed, and reliability grading mapping results are generated. Finally, multiple grading mapping results are spatially stitched to generate a global mineral identification reliability heatmap, which intuitively displays the reliability of mineral identification in different regions, thereby avoiding misjudgments and improving mineral identification efficiency. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A schematic diagram of the task-oriented hyperspectral quality assessment method driven by mineral identification reliability provided in this application is shown.

[0026] Figure 2 This paper illustrates the initialization process of the evaluation task container in the mineral identification reliability-driven task-oriented hyperspectral quality assessment method provided in this application. Detailed Implementation

[0027] This application provides a task-oriented hyperspectral quality assessment method driven by the reliability of mineral identification, which addresses the technical problem that existing hyperspectral data quality assessment methods focus on using global image quality indicators for evaluation, failing to reflect the key diagnostic bands on which specific mineral identification depends, leading to unreliable subsequent mineral identification results.

[0028] 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. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0029] In the description of this invention, unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprising" will be understood to include the stated elements or components without excluding other elements or other components.

[0030] This invention provides a task-oriented hyperspectral quality assessment method driven by mineral identification reliability. (See also...) Figure 1 The method includes:

[0031] Y100: Initialize P evaluation task containers for P target evaluation minerals based on the mineral types to be detected.

[0032] Based on the type of mineral to be detected, P target evaluation minerals are identified, such as iron oxides, clay minerals, and carbonate minerals. These minerals have different spectral characteristics and different signal quality requirements in hyperspectral images. An evaluation task container is initialized for each target evaluation mineral. These evaluation task containers are independent processing modules that can dynamically load the required feature analysis tools and algorithms according to the different spectral characteristics of the mineral. They are specifically designed for quality evaluation of mineral identification tasks. For example, for carbonate minerals, their specific band parameters are loaded, such as absorption characteristics around 2300-2350 nm; for clay minerals, their specific band parameters are loaded, such as absorption characteristics around 2200 nm; and for iron oxides, their specific bands are loaded, such as absorption characteristics around 500-600 nm and 850-1000 nm.

[0033] Y200: Perform spatial index partitioning on the hyperspectral data cube to obtain P spatial data subsets.

[0034] The hyperspectral data cube to be evaluated is three-dimensional, containing spatial information for multiple bands. Based on the number of computational nodes, i.e., the number of P evaluation task containers, the hyperspectral data cube is divided into corresponding spatial data subsets. These spatial data subsets are rectangular regions, each corresponding to a specific task container. Each container only processes the mineral type and band it is responsible for, reducing the computational burden. Each spatial subset may have incomplete data at its boundaries. For example, during spatial indexing, the boundary regions of subsets may lack complete image metadata. Therefore, spatial neighborhood image metadata is used to compensate for these boundary regions, making the data boundaries of each spatial subset more complete. Before data processing, the spatial data subsets undergo dual registration using geographic coordinates and projection parameters to ensure the spatial accuracy of the data.

[0035] Y300: Load the P spatial data subsets in parallel into the P evaluation task containers and execute the mineral-specific quality analysis process in isolation to output P mineral reliability feature sets.

[0036] Through multi-threaded parallel computation, P subsets of spatial data are loaded into P task containers, each executing its own mineral-specific quality analysis process independently without interference. Within each task container, mineral characteristic bands are first extracted and analyzed. For example, for iron oxides, the task container specifically analyzes the signal-to-noise ratio (SNR) and absorption depth stability in the 500-600nm and 850-1000nm bands. The sharpness of the features is evaluated by calculating the local SNR within each specific band range, ensuring signal quality in these characteristic bands during mineral identification. The stability of mineral features within the target band window is assessed, and the spatial consistency of spectral features is evaluated by calculating the standard deviation of absorption depth for all pixels. Each task container outputs a mineral reliability feature set based on its specific mineral's SNR, absorption depth stability, and spectral confusion metrics.

[0037] Y400: Calculate the reliability index of the P mineral identification based on the P mineral reliability feature sets.

[0038] Based on the quality feature set of each mineral type, a weighted fusion is performed to generate the corresponding Mineral Identification Reliability Index (MIRI). Specifically, for each mineral type, local signal-to-noise ratio (SNR) and absorption depth stability are two core indicators. By weighted fusion of these indicators, the reliability index of each mineral within its specific spectral range is calculated. In addition to SNR and depth stability, the spectral confusion index is also considered, which reflects the separability between different minerals. If the spectral characteristics of minerals are very similar, the confusion index is high, and the reliability of mineral identification will decrease. All mineral feature scores and spectral confusion indices are weighted and fused to obtain the final Mineral Identification Reliability Index (MIRI). The higher the index, the stronger the reliability of mineral identification.

[0039] Y500: Using the P mineral identification reliability indices as a grading benchmark, extract P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stability from the cache data areas of the P evaluation task containers.

[0040] Each task container stores mineral-related signal-to-noise ratio (SNR) and absorption depth stability data in its dedicated cache data area. This cache data area refers to the data temporarily stored by the evaluation task container during processing, containing detailed analysis results for each pixel. For each pixel, its corresponding local SNR is extracted. The local SNR reflects the clarity of the signal within a specified band. A strong signal with low noise results in a high SNR, indicating reliable mineral identification for that pixel. Similar to SNR, for each pixel, its corresponding absorption depth stability value is also extracted. This value reflects the depth stability of the mineral feature. Higher stability means better spatial consistency of the mineral feature, more accurately reflecting its composition. Based on the mineral identification reliability index, the system determines which pixel's detailed data to extract from the cache. High-reliability mineral regions will have higher local SNR and absorption depth stability; therefore, the extracted data will be used for subsequent grading processing.

[0041] Y600: Perform reliability partitioning based on the P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stability, and output P reliability classification mapping results.

[0042] Based on the extracted pixel-level local signal-to-noise ratio and pixel-level absorption depth stability, reliability partitioning is performed. This partitioning process divides pixels into different reliability levels according to preset thresholds, such as high reliability, medium reliability, and low reliability zones. By jointly scoring the local signal-to-noise ratio and absorption depth stability, a corresponding reliability level is assigned to each pixel. Based on the partitioning, a reliability classification mapping result is generated for each task container. Each mapping result contains the reliability level of each pixel extracted from the data and is assigned according to its spatial location.

[0043] Y700: By spatially stitching together the P reliability classification mapping results, output a global mineral identification reliability heatmap.

[0044] The reliability grading mapping results of each assessment task container are stitched together according to spatial location to restore complete geographic coverage. During stitching, it is crucial to ensure that the boundaries and zoning information of each region are correctly aligned to avoid data loss or misalignment. Based on the stitched global reliability mapping results, a global mineral identification reliability heatmap is generated. This heatmap displays the reliability distribution of mineral identification across the entire exploration area, using different colors to represent different reliability levels; for example, red represents high reliability, green represents medium reliability, and blue represents low reliability. The global mineral identification reliability heatmap visually shows which areas have high mineral identification quality and which areas may have significant uncertainties, facilitating subsequent decision-making by geological exploration personnel.

[0045] Furthermore, such as Figure 2 As shown, P evaluation task containers for P target evaluation minerals are initialized according to the types of minerals to be detected, including:

[0046] Y110: Based on the P target evaluation minerals constituting the mineral type to be detected, retrieve P baseline task containers from the task container library.

[0047] Y120: Retrieve the P specific spectral parameter sets for the P target evaluation minerals from the preset spectral library.

[0048] Y130: Map and load the P dedicated spectral parameter sets into the P reference task containers for dynamic parameter tuning and correction, and complete the initialization of the P evaluation task containers.

[0049] In the region to be explored, there are multiple mineral types, each with different spectral characteristics and identification criteria. By classifying these mineral characteristics, the number of mineral types to be evaluated, i.e., P target evaluation minerals, can be determined. The task container library is a predefined repository containing evaluation task templates for different mineral types. These task containers are used for spectral data analysis, processing, correction, and result output for specific mineral types. Based on the P target evaluation minerals to be explored, the corresponding task container is retrieved from the task container library. For example, different mineral types have different analysis paths and algorithm models.

[0050] A spectral library is a database containing various mineral spectral datasets. These datasets represent the absorption and reflection characteristics of different minerals across various wavelengths. The purpose of a spectral library is to provide standardized spectral characteristic data for different mineral types, which can be used for subsequent mineral identification, quality analysis, and other tasks. For each target mineral being evaluated, a specific set of spectral parameters is retrieved from the spectral library. These parameters include information such as the mineral's characteristic wavelengths, absorption peaks, and spectral shape, for further spectral analysis. The spectral characteristics of different minerals vary greatly; therefore, each mineral type has its unique set of spectral parameters.

[0051] Mapping loading is the process of associating a specific set of spectral parameters with a baseline task container. The spectral parameter set of each mineral is loaded into the corresponding task container so that the task container can correctly perform the mineral analysis task. The mapping process ensures that the spectral feature data used by each baseline task container is consistent with its corresponding mineral type, thereby ensuring the accuracy of data matching and processing results in the evaluation process.

[0052] The characteristics of mineral spectral data are influenced not only by mineral type but also by acquisition conditions and environmental factors. Therefore, after loading the spectral parameter set, the task container adjusts the parameters to adapt to the actual spectral data. This calibration process improves the model's accuracy and ensures that the task container can execute the correct algorithm based on the actual data. For example, if the spectral data contains noise, bias, or errors, the task container will make necessary adjustments to ensure the accuracy of subsequent processing. Through these initialization steps, the task container ensures that it processes data based on the correct mineral spectral characteristics in subsequent analyses.

[0053] Furthermore, spatial indexing is performed on the hyperspectral data cube to obtain P spatial data subsets, including:

[0054] Y210: Based on the number of computing nodes P, the hyperspectral data cube is divided into P rectangular sub-regions in the spatial dimension.

[0055] Y220: Retrieve spatial neighborhood image data from the hyperspectral data cube to perform buffer compensation for the P rectangular sub-regions, resulting in P dynamic buffers.

[0056] Y230: Reverse map the P dynamic buffers to the P rectangular sub-regions to obtain P spatial buffer compensation subsets.

[0057] Y240: After performing two-dimensional georegistration and binding of spatial coordinate system and projection parameters on the P dynamic buffer compensation subsets, data BIL format reconstruction is performed to generate the P spatial data subsets.

[0058] A hyperspectral data cube refers to three-dimensional data acquired through remote sensing or imaging systems. It contains the spatial coordinates of each pixel and reflection or absorption information across multiple spectral bands. The data has three dimensions: spatial, spectral, and temporal. The number of computing nodes, P, refers to the amount of computational resources used when processing hyperspectral data in parallel. To efficiently process the data, the entire hyperspectral data cube is spatially divided into P rectangular sub-regions. The size of the rectangular sub-regions can be adjusted according to the number of computing nodes, P. Typically, the entire dataset is evenly divided so that each node processes an equal amount of data during parallel processing.

[0059] The spatial boundary of each rectangular sub-region may overlap with other sub-regions. Especially when processing data, the edge areas of the sub-regions may have incomplete, missing or incorrect data. Spatial neighborhood image data refers to the image data of other images adjacent to the edge of the current rectangular sub-region in the hyperspectral data cube. By extracting these neighborhood data, compensation processing can be performed on the edge areas.

[0060] Buffer compensation refers to adding data from neighboring pixels to the edges of each rectangular sub-region to ensure that data loss or errors do not occur during data processing due to edge effects. This spatially expands the range of each rectangular sub-region, increasing data continuity. The buffer area of ​​each rectangular sub-region is dynamically adjusted according to actual needs to ensure the data integrity of each sub-region and avoid the impact of edge effects during segmentation on the analysis results. After buffer compensation, each rectangular sub-region is expanded and improved, forming a dynamic buffer for subsequent mapping and processing.

[0061] The data in the buffer is reverse-mapped from its dynamic buffer back to the original rectangular sub-region. This is to accurately reorganize additional neighborhood image data into the corresponding rectangular sub-regions to ensure data integrity. After the reverse mapping, spatial buffer-compensated subsets are obtained, which contain the data from the original rectangular sub-regions and have already undergone edge compensation processing.

[0062] Two-dimensional georegistration refers to aligning the spatial coordinate system of each spatial buffer compensation subset with a unified reference coordinate system to ensure spatial consistency of the data. Projection parameters also need to be adjusted synchronously to ensure that each data subset has an accurate positional relationship in geographic space. BIL format is a remote sensing data storage format that interleaves spectral data row by row. To optimize storage and subsequent processing, each spatial data subset is reassembled using BIL format. The reassembly process sorts the hyperspectral data by row and band for efficient subsequent reading and analysis. The resulting P spatial data subsets are hyperspectral data subsets that have undergone spatial coordinate system and projection parameter calibration and format reassembly. These subsets can be easily processed in parallel, ensuring data accuracy and consistency.

[0063] Furthermore, mapping and loading the P dedicated spectral parameter sets into the P reference task containers for dynamic parameter tuning and calibration completes the initialization of the P evaluation task containers, including:

[0064] Y131: Analyze the first dedicated spectral parameter set to obtain the first diagnostic band parameters, the first feature quantization parameters, and the first spatial correlation settings.

[0065] Y132: Dynamically generate the first mineral feature window based on the first diagnostic band parameters.

[0066] Y133: The first signal-to-noise ratio calculation template is loaded based on the first feature quantization parameter adaptive configuration.

[0067] Y134: Based on the first spatial association setting, retrieve the first terminal to extract sensitive parameters.

[0068] Y135: Inject the first mineral feature window, the first signal-to-noise ratio calculation template, and the first endmember extraction sensitive parameters into the first benchmark task container to perform hardware acceleration parameter compilation and complete the initialization of the first evaluation task container.

[0069] The first dedicated spectral parameter set consists of spectral feature data related to the first target mineral being evaluated. Diagnostic bands refer to bands with significant spectral characteristics in a specific mineral or substance. For example, in mineral detection, the reflectance or absorption characteristics of certain bands can be used to accurately diagnose the type and properties of a mineral. The first diagnostic band parameters are important band parameters related to the target mineral extracted from this dedicated spectral parameter set, used for mineral identification. Feature quantification refers to the quantitative analysis of mineral spectral data to extract features for identification and classification. These parameters can be ratios between bands, absorption depths, reflectance differences, etc. The first feature quantification parameters include these quantitative features for subsequent calculations and analysis. The first spatial association setting refers to spatial information related to mineral features, such as the spectral characteristics of adjacent pixels, the distribution patterns of spatial neighborhoods, etc. The distribution of different mineral types exhibits spatial clustering or interrelationships.

[0070] In hyperspectral data processing, a mineral feature window refers to a specific spectral range or band window. This window is specifically used to extract the spectral features of the target mineral. For example, certain band windows can effectively distinguish the spectral features of different minerals. Dynamically generated features windows automatically calculate and determine a band window suitable for the current mineral type based on the first diagnostic band parameters. This dynamically generated feature window can better adapt to the spectral characteristics of different minerals and improve the accuracy of identification.

[0071] Signal-to-noise ratio (SNR) measures the ratio of signal strength to noise level. In remote sensing and spectral analysis, SNR is a key indicator for evaluating data quality; a high SNR signifies clear signals, low noise, and high data quality. The first SNR calculation template is an algorithmic framework used to calculate the signal-to-noise ratio based on data characteristics, thereby evaluating the quality of spectral data. Adaptive configuration means automatically adjusting and loading the appropriate SNR calculation template for the current analysis task based on the first feature quantization parameters. This method selects the optimal algorithm and parameter configuration according to different data characteristics, ensuring the accuracy and effectiveness of SNR calculation.

[0072] Endmembers are pure spectral signatures in a spectral dataset that represent a substance or mineral. In mineral analysis, endmember extraction uses algorithms to identify and extract the most representative spectral features for quantitative mineral analysis. Sensitive parameters refer to key parameters that significantly impact the mineral identification process. During endmember extraction, some parameters can greatly affect the results, such as the selection of spectral features and the weighting of bands. Based on the first spatial association settings, and according to specific spatial patterns and association rules, the most suitable first endmembers are retrieved from the database to extract sensitive parameters. This means that the most appropriate extraction strategy is automatically selected based on the spatial characteristics of the mineral.

[0073] The generated first mineral feature window, first signal-to-noise ratio calculation template, and first endmember extraction sensitive parameters are input as key parameters into the first benchmark task container. Hardware acceleration refers to using dedicated hardware to accelerate the data processing process. In mineral analysis, hardware acceleration can significantly improve data processing speed. Parameter compilation refers to compiling the various parameters extracted earlier according to task requirements to enable efficient execution on hardware. Once all parameters are injected and compiled, the first evaluation task container will be initialized. This container will contain all necessary parameters and be able to execute mineral analysis tasks in a hardware-accelerated environment.

[0074] Furthermore, the P spatial data subsets are loaded in parallel into the P evaluation task containers for isolated mineral-specific quality analysis processing, outputting P mineral reliability feature sets, including:

[0075] Y310: Within the first mineral feature window, the first signal-to-noise ratio calculation template is used to perform local signal-to-noise ratio calculation on the first spatial data subset element by element, and the first pixel-level signal-to-noise ratio set is output.

[0076] Y320: Calculate the mean signal-to-noise ratio (SNR) of the first pixel-level SNR set.

[0077] Y330: After performing continuum removal on the first spatial data subset, perform absorption depth stability analysis based on the reflectance data of the first spatial data subset within the first mineral feature window to obtain the first absorption depth standard deviation.

[0078] Y340: Within the first spatial data subset, the first endmember is used to extract sensitive parameters to perform endmember spectral separability calculation, generating a first spectral confusion index.

[0079] Y350: Calls the first window identifier of the first mineral feature window, and outputs the first mineral reliability feature set by integrating the first window identifier, the first pixel-level signal-to-noise ratio mean, the first absorption depth standard deviation and the first spectral confusion index.

[0080] Using a pre-configured first signal-to-noise ratio (SNR) calculation template, the SNR of each pixel is calculated. This template calculates the SNR of each pixel by considering different spectral characteristics, helping to identify areas with better data quality. Pixel-by-pixel calculation means that the SNR is calculated independently for each pixel. This process is used to determine the quality differences between different pixels and to mark data areas with low SNR, aiding subsequent analysis. After calculation, the SNR data of all pixels are aggregated into a first-pixel-level SNR set. This set contains the SNR value of each pixel and is used for subsequent quality control and mineral feature analysis.

[0081] The mean SNR of the first pixel-level SNR is calculated by averaging all SNR values ​​in the first pixel-level SNR set. This mean SNR reflects the average data quality of the entire spatial data subset. A high mean SNR indicates that most pixels have a high SNR and good data quality. A low mean SNR indicates that there are many low-quality pixels in the data.

[0082] Continuum removal refers to the removal of outliers or noise that affect data stability. This can be achieved through filtering, smoothing, or data cleaning techniques to ensure more accurate subsequent analysis. In mineral spectral analysis, reflectance refers to the intensity of light reflected by a substance. Variations in reflectance data within a specific wavelength band reflect the spectral characteristics of a mineral and are an important basis for mineral identification and analysis. Absorption depth refers to the degree to which a mineral absorbs light at a specific wavelength. A larger absorption depth indicates specific chemical composition or structural characteristics of the mineral. By analyzing the stability of absorption depth, it can be determined whether the mineral has uniform spectral characteristics. Standard deviation is an indicator of the fluctuation in data distribution, reflecting the degree of data dispersion. The first absorption depth standard deviation represents the degree of variation in absorption depth data within the mineral characteristic window. A smaller standard deviation indicates stable variation in absorption depth and relatively consistent mineral characteristics. A larger standard deviation indicates significant differences in mineral characteristics.

[0083] Endmember spectral separability refers to whether the spectral features of different minerals or substances can be clearly distinguished in spectral space. The purpose of endmember extraction is to extract the most representative spectral information from complex hyperspectral data, and separability measures whether these endmembers can be effectively distinguished. The spectral confusion index is an indicator that measures the degree of confusion between the spectra of different minerals. It is used to describe difficult regions in mineral identification. If the spectral features of two minerals are very similar, the spectral confusion index will be high, indicating that the two minerals are difficult to distinguish in the current spectral window. Based on the sensitivity parameters of the first endmember extraction, the endmember spectra are analyzed, and their separability is calculated. This step generates the first spectral confusion index by evaluating the overlap of spectral distributions, representing the degree of spectral confusion between different minerals in the current feature window.

[0084] A window identifier is a unique identifier for a specific mineral feature window, used to distinguish and call different feature windows. Each window represents a specific set of spectral ranges and mineral analysis features. Combining four key indicators—the first window identifier, the first pixel-level signal-to-noise ratio mean, the first absorption depth standard deviation, and the first spectral confusion index—generates a comprehensive first mineral reliability feature set.

[0085] Furthermore, calculating the P mineral identification reliability indices based on the P mineral reliability feature sets includes:

[0086] Y410: Based on the first window identifier, retrieve the piecewise linear normalization threshold, perform adaptive piecewise linear normalization on the first pixel-level signal-to-noise ratio mean, and output the first normalized signal-to-noise ratio mean.

[0087] Y420: Perform a reciprocal transformation on the first absorption depth standard deviation and then linearly normalize it to output the first normalized absorption depth stability.

[0088] Y430: Retrieves the first weight vector based on the first window identifier, and uses the first weight vector to perform a weighted integration of the first normalized signal-to-noise ratio mean, the first normalized absorption depth stability, and the first spectral confusion index, and outputs the first benchmark reliability index.

[0089] Y440: Integrates regional geological prior probability and sensor band attenuation factor, performs confidence calibration of the first benchmark reliability index, and outputs the first mineral identification reliability index.

[0090] Piecewise linear normalization is a method of mapping data to a specific range. It processes different ranges of data by defining multiple linear intervals. In this step, the piecewise linear normalization threshold for the first pixel-level signal-to-noise ratio mean is retrieved through the first window identifier, and adaptive piecewise linear normalization is performed on it. This process maps the signal-to-noise ratio value to a fixed range, such as between 0 and 1, based on the distribution of the signal-to-noise ratio value, thereby eliminating the differences between different regions and making subsequent processing more consistent. After normalization is completed, the first normalized signal-to-noise ratio mean is output. It is the standardized signal-to-noise ratio data, which is convenient for further weighted processing.

[0091] The reciprocal transformation is a method of reversing data processing, used to convert high values ​​to low values ​​and vice versa. In this step, the first standard deviation of absorption depth is reciprocally transformed. This means that the larger standard deviation of absorption depth is converted to a smaller value so that it can be compared with other parameters on the same scale. The reciprocally transformed standard deviation of absorption depth data is then linearly normalized, that is, the data is mapped to a specified range, such as 0 to 1, to ensure that its dimensions and range are consistent with those of other parameters. After completing the reciprocal transformation and normalization, the first normalized absorption depth stability is output, which represents the stability of the mineral's absorption depth after normalization.

[0092] The weight vector is used to specify the importance of each feature. Each feature is assigned a different weight according to its contribution to mineral identification. The corresponding first weight vector is retrieved based on the first window identifier. The three feature values—the first normalized signal-to-noise ratio mean, the first normalized absorption depth stability, and the first spectral confusion index—are multiplied by their respective weights and then summed. This weighted integration method can synthesize the contributions of each feature to obtain a comprehensive index, namely the first benchmark reliability index. It is a preliminary assessment of mineral reliability, reflecting the overall reliability of mineral identification under the current features.

[0093] Regional geological prior probability, based on geological exploration or known mineral distribution and combined with geological principles, provides additional background information for mineral identification, helping to make more reasonable adjustments to the reliability of minerals in certain areas. The sensor band attenuation factor is a parameter related to the sensor hardware, representing the attenuation of the sensor's response capability in a specific band. This factor can be adjusted according to the sensor's response capability in different bands to ensure more accurate signal reliability assessment. Confidence calibration aims to adjust the first benchmark reliability index using external information, thereby improving the credibility of the results. After calibration, the results no longer rely solely on hyperspectral data but also incorporate additional geological and sensor information to improve the reliability of the final output. After calibration, the output first mineral identification reliability index provides a more accurate and reliable mineral identification result.

[0094] Furthermore, using the P mineral identification reliability indices as a grading benchmark, P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stability are extracted from the cache data areas of the P evaluation task containers, including:

[0095] Y510: Perform multi-level accuracy grading based on the P mineral identification reliability indices to obtain P grading accuracy strategies.

[0096] Y520: parse the memory mapping addresses of the P evaluation task containers, and locate the P cell-level local signal-to-noise ratio buffers and the P cell-level absorption depth stability buffers.

[0097] Y530: Using the aforementioned P-level precision strategy, the P-level local signal-to-noise ratio and P-level absorption depth stability are extracted by parallel mapping in the P-level local signal-to-noise ratio buffer and the P-level absorption depth stability buffer.

[0098] Multi-level accuracy grading involves classifying minerals into different accuracy levels based on their identification reliability index. By evaluating the reliability index of different minerals, the required processing accuracy is determined. For example, high-reliability minerals require lower computational accuracy, while low-reliability minerals require higher precision to ensure the accuracy of the identification results. Based on the reliability index of each mineral, a grading accuracy strategy is generated, specifying the computational accuracy used under different mineral categories.

[0099] Each evaluation task container allocates specific regions in memory to store data during execution. These regions are associated with the external storage system through memory mapping, facilitating fast data access. Resolving memory-mapped addresses refers to finding the memory regions associated with each task container through address mapping. Each task container allocates caches for different mineral features. For example, a cell-level local signal-to-noise ratio cache stores the signal-to-noise ratio calculation results for each cell, and an absorption depth stability cache stores the results of absorption depth stability analysis. These caches are bound to the task container through memory mapping, ensuring that data can be quickly retrieved and processed.

[0100] Based on the generated P hierarchical precision strategies, data access is performed through parallel mapping extraction in P pixel-level local signal-to-noise ratio (SNR) buffers and P pixel-level absorption depth stability buffers. This means that multiple data blocks can be processed simultaneously, improving computational efficiency. Parallel mapping extraction is performed for each pixel-level SNR and absorption depth stability, i.e., data is quickly extracted from the corresponding buffer. Each evaluation task container extracts the corresponding high-precision or low-precision data from the buffer according to its corresponding precision strategy. Parallel computing significantly improves processing efficiency, especially when dealing with large amounts of data. Each evaluation task container can execute data extraction and processing tasks in parallel on multiple computing nodes, avoiding the computational bottleneck of a single node. The data extraction process is controlled by the hierarchical precision strategy, ensuring that corresponding operations are performed under different precision requirements. This method automatically adjusts the precision based on the mineral's reliability index, thereby improving computational efficiency while ensuring the accuracy of the identification results.

[0101] Furthermore, based on the P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stability, reliability partitioning is performed, and P reliability classification mapping results are output, including:

[0102] Y610: Retrieve P mineral-specific normalized parameters based on the P window identifiers of the P evaluation task containers.

[0103] Y620: The P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stability are normalized into P normalized signal-to-noise ratio sets and P normalized absorption depth stability sets using the P mineral-specific normalization parameters.

[0104] Y630: Load digital elevation model data and perform terrain slope correction on the P normalized signal-to-noise ratio sets to obtain P terrain-corrected signal-to-noise ratio sets.

[0105] Y640: Call the regional geological map data to perform mineral boundary buffering downgrading on the P normalized absorption depth stability sets to obtain P mineral downgraded absorption depth sets.

[0106] Y650: Based on the P window identifiers, retrieve the P mineral association weight ratios, and construct a P-pixel-level comprehensive reliability weight model based on the P terrain correction signal-to-noise ratio sets, the P mineral degradation absorption depth sets, and the P mineral association weight ratios.

[0107] Y660: Based on the preset mineral-specific grading threshold table, the reliability level of the P pixel-level comprehensive reliability weight model is divided into P reliability level data.

[0108] Y670: After performing morphological closing operations to eliminate isolated pixels on the P reliability level data, partition boundary smoothing processing is performed to obtain the P reliability level mapping results.

[0109] Each evaluation task container has a window identifier to identify the mineral feature window corresponding to that container. Mineral-specific normalization parameters are a set of parameters associated with a specific mineral. These parameters help to standardize or normalize the data according to the characteristics of the mineral. These parameters may include specific signal-to-noise ratio standards, absorption depth conversion coefficients, etc.

[0110] The normalized signal-to-noise ratio (SNR) set is obtained by applying mineral-specific normalization parameters to standardize or normalize the local SNR of each pixel. Specifically, this involves performing adaptive linear normalization on the SNR data, transforming the SNR values ​​to a standard range, such as 0 to 1, thereby eliminating measurement biases between different minerals. Similarly, the absorption depth stability data undergoes normalization, which includes linear normalization after a reciprocal transformation. The reciprocal transformation involves taking the reciprocal of the standard deviation of the absorption depth; for example, a lower standard deviation indicates a more stable absorption depth. Then, linear normalization is performed on the reciprocal value to standardize it to a uniform range. Ultimately, through these two normalization steps, P normalized SNR sets and P normalized absorption depth stability sets are obtained. These two datasets provide a unified standard data foundation for subsequent reliability analysis.

[0111] Digital elevation models (DEMs) are terrain height data obtained through remote sensing or ground measurement. They provide height information for each pixel on the ground. DEM data has a significant impact on hyperspectral data analysis because terrain changes can lead to geometric distortion or illumination changes in image data, which in turn affect mineral identification results.

[0112] By loading digital elevation model data, the normalized signal-to-noise ratio (SNR) data is corrected according to the terrain slope. Terrain slope correction adjusts the SNR value based on the slope of each pixel, reducing errors caused by terrain factors. In areas with steeper slopes, the correction is applied more forcefully to ensure accuracy. Ultimately, after terrain slope correction, P terrain-corrected SNR sets are generated. These corrected data more accurately reflect the spectral characteristics of the minerals, reducing interference from terrain factors.

[0113] Regional geological maps are maps that describe the geological features, strata distribution, mineral resources, and geological structures of a specific area. They provide information on mineral distribution within a geological context and are a crucial data source for mineral exploration and analysis. Mineral boundary buffering refers to regional adjustments based on the boundaries of strata or veins to better accommodate the effects of geological changes. In some cases, mineral boundaries can introduce noise or errors during mineral identification, especially when hyperspectral data is affected by mineral variations. Buffering and degradation processing reduces boundary noise interference by lowering the absorption depth stability value of areas close to mineral boundaries. The goal of this step is to smooth the signal in areas of mineral variation and reduce uncertainty in mineral identification. After this processing, the resulting mineral degradation absorption depth set is more stable, reflects the true characteristics of the minerals, and reduces interference from mineral boundaries.

[0114] The identification of each mineral does not rely on a single feature, but rather on a combination of multiple features. The mineral association weight ratio is a weighted parameter determined based on the characteristics of different minerals. Each mineral's features have different importance under different conditions, and these weight ratios are used to balance the contribution of each feature to the mineral identification result. Based on the signal-to-noise ratio, absorption depth stability, and association weight ratio unique to each mineral, a pixel-level comprehensive reliability weight model is constructed. This model comprehensively evaluates the mineral identification reliability of each pixel by integrating the signal-to-noise ratio after terrain correction, the absorption depth stability after downgrading, and the association weights between minerals.

[0115] The mineral-specific grading threshold table is a standard table built based on prior geological knowledge and historical data. It includes reliability grading standards for mineral identification results. Different thresholds are set according to the characteristics of each mineral to determine its reliability. For example, a high signal-to-noise ratio and good absorption depth stability correspond to a higher reliability level. Reliability level classification is based on a comprehensive reliability weight model and grading thresholds for each pixel to determine its reliability level. For example, the results are divided into high reliability, medium reliability, and low reliability levels. Each pixel's score is compared with a preset threshold to ultimately classify it into the corresponding level. After this classification, each pixel obtains corresponding reliability level data, which provides the foundation for subsequent spatial partitioning and visualization.

[0116] Morphological operations primarily manipulate the shape of the image to eliminate noise, fill holes, and smooth boundaries. Closing operations process the image through erosion (removing edge details) and dilation (filling holes). Here, the goal of closing operations is to eliminate isolated pixels—isolated, discontinuous, or misidentified pixels that may have been generated by noise during mineral identification. After removing isolated pixels, partition boundary smoothing is used to eliminate unnatural boundaries, especially edge effects caused by previous data processing. Smoothing makes the boundaries of mineral distribution more natural and consistent, avoiding unreasonable partition lines or overly refined edges. After morphological closing operations and boundary smoothing, the final reliability grading mapping result presents the mineral identification reliability level of each pixel. This mapping result can be used for further analysis, visualization, and mineral exploration decision support.

[0117] Furthermore, spatial boundary markers are monitored during the data extraction process, and a dynamic buffer compensation mechanism is triggered when a spatial subset boundary cell is detected.

[0118] Spatial boundaries refer to the boundary regions between multiple spatial subsets. During data extraction, it's crucial to identify which pixels lie at these boundaries. These boundary pixels typically exhibit significant characteristic differences, possibly due to substantial changes in mineral characteristics or spectral properties of neighboring regions. Spatial subset boundary pixels are located at the edges of spatial subsets, and their spectral information may be influenced by neighboring data blocks. These pixels are geographically located across different regions, potentially spanning different geological or mineral features, leading to relatively unstable signal-to-noise ratios and absorption depth stability. The core of the dynamic buffer compensation mechanism is to generate a buffer at the boundaries of spatial subsets to compensate for these boundary pixels. This buffer smooths the signal of the boundary pixels, reducing the impact of boundary effects on overall data processing. For example, it can use data from adjacent regions or pre-calculated spectral models to fill in or adjust the signals of these boundary pixels, aiming to reduce the anomalous influence of spatial subset boundary pixels and improve the reliability of boundary pixel data.

[0119] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A mineral identification reliability-driven task-oriented hyperspectral quality assessment method, characterized in that, The method comprises: The method comprises: According to the mineral type to be detected, initialize P target evaluation mineral P evaluation task containers; Perform spatial index segmentation on the hyperspectral data cube to obtain P spatial data subsets; Load the P spatial data subsets into the P evaluation task containers in parallel to execute mineral-specific quality analysis processes in isolation, output P mineral reliability feature sets; According to the P mineral reliability feature sets, calculate P mineral recognition reliability indexes; Take the P mineral recognition reliability indexes as grading benchmarks, map and extract P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stabilities from the cache data area of the P evaluation task containers; According to the P pixel-level local signal-to-noise ratios and the P pixel-level absorption depth stabilities, perform reliability partitioning, and output P reliability grading mapping results; Through spatial splicing of the P reliability grading mapping results, output a global mineral recognition reliability heat map; Wherein, loading the P spatial data subsets into the P evaluation task containers in parallel to execute mineral-specific quality analysis processes in isolation, and output P mineral reliability feature sets, comprise: In the first mineral feature window, a first signal-to-noise ratio calculation template is used to perform local signal-to-noise ratio calculation on the first spatial data subset element by element, and output a first pixel-level signal-to-noise ratio set; Calculate the first pixel-level signal-to-noise ratio mean of the first pixel-level signal-to-noise ratio set; After performing continuum removal on the first spatial data subset, perform absorption depth stability analysis according to the reflectance data of the first spatial data subset in the first mineral feature window to obtain the first absorption depth standard deviation; In the first spatial data subset, a first endmember extraction sensitive parameter is used to perform endmember spectral separability calculation, and a first spectral confusion index is generated; Call the first window identifier of the first mineral feature window, and output the first mineral reliability feature set by integrating the first window identifier, the first pixel-level signal-to-noise ratio mean, the first absorption depth standard deviation and the first spectral confusion index; Wherein, according to the P mineral reliability feature sets, calculate P mineral recognition reliability indexes, comprising: According to the first window identifier, call the piecewise linear normalization threshold, perform adaptive piecewise linear normalization on the first pixel-level signal-to-noise ratio mean, and output the first normalized signal-to-noise ratio mean; After performing inverse conversion and linear normalization on the first absorption depth standard deviation, output the first normalized absorption depth stability; According to the first window identifier, call the first weight vector, and use the first weight vector to perform weighted integration of the first normalized signal-to-noise ratio mean, the first normalized absorption depth stability and the first spectral confusion index, and output the first benchmark reliability index; 2. The mineral identification reliability-driven task-based hyperspectral quality assessment method of claim 1, wherein, Fuse regional geological prior probability and sensor band attenuation factor to perform confidence calibration of the first benchmark reliability index, and output the first mineral recognition reliability index. According to the mineral type to be detected, initialize P target evaluation mineral P evaluation task containers, comprising: According to the P target evaluation minerals constituting the mineral type to be detected, call P benchmark task containers from the task container library; retrieve P sets of exclusive spectral parameters of P target evaluation minerals from a preset spectral library; map load the P sets of exclusive spectral parameters to the P reference task containers for dynamic parameter correction, and complete initialization of the P evaluation task containers.

3. The mineral identification reliability-driven task-based hyperspectral quality assessment method of claim 1, wherein, perform spatial index segmentation on the hyperspectral data cube to obtain P spatial data subsets, including: segment the hyperspectral data cube in the spatial dimension into P rectangular sub-regions according to the number P of computing nodes; retrieve spatial neighborhood pixel data from the hyperspectral data cube for buffer compensation of the P rectangular sub-regions to obtain P dynamic buffer regions; reverse map the P dynamic buffer regions to the P rectangular sub-regions to obtain P spatial buffer compensation subsets; after performing double-dimensional geographical registration binding of the spatial coordinate system and the projection parameters on the P dynamic buffer compensation subsets, implement data BIL format reorganization to generate the P spatial data subsets.

4. The mineral identification reliability-driven task-based hyperspectral quality assessment method of claim 2, wherein, map load the P sets of exclusive spectral parameters to the P reference task containers for dynamic parameter correction, and complete initialization of the P evaluation task containers, including: analyze the first set of exclusive spectral parameters to obtain first diagnostic waveband parameters, first feature quantization parameters, and first spatial correlation settings; dynamically generate a first mineral feature window based on the first diagnostic waveband parameters; adaptively configure a first signal-to-noise ratio calculation template according to the first feature quantization parameters; directly retrieve first endmember extraction sensitive parameters according to the first spatial correlation settings; inject the first mineral feature window, the first signal-to-noise ratio calculation template, and the first endmember extraction sensitive parameters into a first reference task container to perform hardware acceleration parameter compilation, and complete initialization of a first evaluation task container.

5. The mineral identification reliability-driven task-based hyperspectral quality assessment method of claim 1, wherein, use the P mineral recognition reliability indexes as grading benchmarks to map extract P pixel-level local signal-to-noise ratios and P pixel-level absorption depth stabilities from the cache data area of the P evaluation task containers, including: perform multi-level precision grading according to the P mineral recognition reliability indexes to obtain P grading precision strategies; analyze the memory mapping addresses of the P evaluation task containers to locate P pixel-level local signal-to-noise ratio cache areas and P pixel-level absorption depth stability cache areas; use the P grading precision strategies to perform parallel mapping extraction of the P pixel-level local signal-to-noise ratios and the P pixel-level absorption depth stabilities in the P pixel-level local signal-to-noise ratio cache areas and the P pixel-level absorption depth stability cache areas.

6. The mineral identification reliability-driven task-based hyperspectral quality assessment method of claim 5, wherein, perform reliability partitioning based on the P pixel-level local signal-to-noise ratios and the P pixel-level absorption depth stabilities to output P reliability grading mapping results, including: retrieve P mineral exclusive normalization parameters based on P window identifiers of the P evaluation task containers; use the P mineral exclusive normalization parameters to normalize the P pixel-level local signal-to-noise ratios and the P pixel-level absorption depth stabilities into P normalized signal-to-noise ratio sets and P normalized absorption depth stability sets; load digital elevation model data to perform terrain slope correction on the P normalized signal-to-noise ratio sets to obtain P terrain-corrected signal-to-noise ratio sets; The regional geological map data is called to perform mineral boundary buffer degradation processing on the P sets of normalized absorption depth stability, to obtain P sets of mineral degraded absorption depth; P mineral correlation weight ratios are called according to the P window identifiers, and P pixel-level comprehensive reliability weight models are constructed based on the P sets of terrain correction signal-to-noise ratios, the P sets of mineral degraded absorption depth and the P mineral correlation weight ratios; The P pixel-level comprehensive reliability weight models are subjected to reliability grade division according to a preset mineral exclusive grading threshold table to obtain P reliability grade data; After performing a morphological closing operation on the P reliability grade data to eliminate isolated pixels, partition boundary smoothing processing is performed to obtain the P reliability grading mapping results.

7. The mineral identification reliability-driven task-based hyperspectral quality assessment method of claim 5, wherein, During the data extraction process, the spatial boundary identifier is monitored, and when a spatial subset boundary pixel is detected, a dynamic buffer compensation mechanism is triggered.

Images