Blockchain-based remote sensing agricultural big data management system

Abstract

The application relates to the technical field of remote sensing agricultural data management, and discloses a remote sensing agricultural big data management system based on a blockchain. The system comprises image data, which is preprocessed through radiation, atmospheric and geometric correction to generate standardized remote sensing images; a feature extraction module performs multi-scale segmentation on the images, identifies crop feature regions, extracts texture, spectral and shape feature values, and fuses to form a multi-dimensional feature vector; a blockchain storage module generates feature fingerprints through hash operation on the vector, combines a time stamp and a previous block hash value to construct a new block, and forms an unalterable agricultural data chain after verification by a consensus mechanism; a path analysis module traverses the block sequence, extracts storage paths and screens a high-correlation feature path set; and an intelligent classification module extracts feature labels and generates an agricultural data classification structure through a multi-dimensional matching algorithm. The system realizes standardized processing, safe storage and intelligent application of remote sensing agricultural data, and adapts to the development needs of agricultural modernization.

Description

Technical Field

[0001] This invention relates to the field of remote sensing agricultural data management technology, specifically a blockchain-based remote sensing agricultural big data management system. Background Technology

[0002] In the process of agricultural modernization, data has become a core element driving improved production efficiency. Remote sensing technology, with its wide coverage and real-time dynamic monitoring capabilities, is widely used in scenarios such as crop growth monitoring, agricultural resource surveys, and disaster early warning. With the diversification of remote sensing equipment, satellite remote sensing, UAV remote sensing, and ground sensors, among other multi-source devices, together constitute an agricultural data acquisition network, continuously generating massive amounts of raw image data. This data contains various key information throughout the crop growth cycle; however, its sources are scattered, its formats vary, and the raw data commonly suffers from problems such as radiation distortion, atmospheric scattering interference, and geometric deformation, directly affecting the usability of the data.

[0003] Traditional agricultural big data management often employs a centralized storage architecture, with data frequently scattered across independent systems belonging to different entities such as agricultural producers, monitoring agencies, and research institutions, creating isolated data silos. Inconsistent data standards across these systems severely hinder data flow and sharing, preventing collaborative applications across entities and stages. Furthermore, under a centralized storage model, data security relies on a single central node, making it vulnerable to malicious attacks, human tampering, or data loss due to hardware failures, thus lacking effective guarantees for data authenticity and integrity.

[0004] In the data processing stage, traditional methods for feature extraction from remote sensing images are often limited to a single scale and a single dimension, making it difficult to comprehensively capture the complex characteristics of crops. Some schemes only extract spectral or shape features, ignoring key information such as texture, resulting in one-sided feature data that cannot accurately reflect the true growth state of crops. In the data classification process, traditional methods often rely on single feature matching, lacking the exploration of relationships between features, resulting in classification results with insufficient logic, making it difficult to meet the refined data needs of different scenarios such as agricultural production, resource allocation, and market regulation.

[0005] Blockchain technology, with its characteristics of distributed ledger, cryptographic verification, and immutability, has demonstrated unique advantages in the field of data management and has been successfully applied in industries such as finance and logistics. However, in the field of remote sensing agricultural big data management, existing technologies have mostly focused on simple on-chain data storage, failing to address the specific needs of remote sensing data preprocessing and feature extraction. Summary of the Invention

[0006] The purpose of this invention is to provide a blockchain-based remote sensing agricultural big data management system to solve the problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides a blockchain-based remote sensing agricultural big data management system, the system comprising:

[0008] The remote sensing data acquisition module is used to receive raw image data collected by multi-source agricultural remote sensing equipment, perform standardized preprocessing operations on the raw image data, including radiometric correction, atmospheric correction and geometric correction, and generate standardized remote sensing images.

[0009] The feature extraction module is used to perform multi-scale segmentation processing on the standardized remote sensing image, identify crop feature regions in the image, extract texture feature values, spectral feature values ​​and shape feature values ​​of each feature region, and fuse them to generate a multi-dimensional feature vector;

[0010] The blockchain storage module is used to perform hash operations on the multi-dimensional feature vector to generate feature fingerprints, combine the timestamp and the hash value of the previous block to construct a new block, and add it to the blockchain after verification through the consensus mechanism to form an immutable agricultural data chain.

[0011] The path analysis module is used to traverse the block sequence in the blockchain, extract the storage path of feature vectors, analyze the intersection nodes between different paths, calculate the path similarity, and filter the set of highly correlated feature paths.

[0012] The intelligent classification module is used to extract corresponding crop feature labels based on the terminal blocks in the feature path set, use a multi-dimensional matching algorithm to determine the correlation between feature labels, and generate an agricultural data classification structure.

[0013] Preferably, the remote sensing data acquisition module includes:

[0014] The data receiving unit is used to connect satellite remote sensing equipment, UAV remote sensing equipment and ground sensor network to receive multispectral image data, hyperspectral image data and thermal infrared image data;

[0015] The correction processing unit is used to perform radiometric calibration processing on the received image data, eliminate sensor errors, perform atmospheric parameter inversion correction, eliminate atmospheric scattering effects, implement geometric fine correction, and eliminate image distortion caused by terrain undulations.

[0016] The quality assessment unit is used to calculate the signal-to-noise ratio and spatial resolution of the corrected image, set the quality threshold, and screen standardized remote sensing images that meet the quality requirements.

[0017] Preferably, the feature extraction module includes:

[0018] The image segmentation unit is used to segment standardized remote sensing images into regions using the watershed algorithm, identify the boundaries of crop planting areas, and extract independent crop units.

[0019] The feature calculation unit is used to calculate the gray-level co-occurrence matrix texture features of each cropping unit, extract the spectral features of the normalized vegetation index and the enhanced vegetation index, and calculate the shape features of the region's perimeter, area, perimeter-to-area ratio, compactness, aspect ratio, and shape index.

[0020] The vector generation unit is used to normalize the texture feature values, spectral feature values ​​and shape feature values, and then perform weighted fusion according to preset weights to generate a feature vector of fixed dimensions.

[0021] Preferably, the blockchain storage module includes:

[0022] The hash operation unit is used to perform SHA-256 hash calculation on the feature vector to generate a unique feature fingerprint;

[0023] The block building unit is used to combine the feature fingerprint with the collection timestamp and the hash value of the previous block, and add a random number to generate a new block;

[0024] The consensus verification unit uses an improved Byzantine fault-tolerant algorithm to verify new blocks and adds them to the end of the blockchain after reaching consensus through a node voting mechanism.

[0025] Preferably, the path analysis module includes:

[0026] The path extraction unit is used to traverse from the starting block of the blockchain, record the storage location sequence of each feature vector, and generate the feature storage path;

[0027] The similarity calculation unit is used to calculate the similarity between different paths using a dynamic time warping algorithm and to set a similarity threshold.

[0028] The path filtering unit is used to filter critical paths with similarity exceeding a threshold based on the similarity calculation results, and to establish a set of critical paths.

[0029] Preferably, the intelligent classification module includes:

[0030] The tag extraction unit is used to extract stored crop variety tags, growth cycle tags, and pest and disease tags from the terminal blocks of the critical path.

[0031] The association analysis unit is used to construct association rules between tags, calculate tag co-occurrence frequency, and establish a tag association network.

[0032] The classification generation unit is used to identify tag communities based on the tag association network and the community detection algorithm to generate a multi-level agricultural data classification system.

[0033] Preferably, the data receiving unit includes:

[0034] Multi-source interface components are used to adapt to the data transmission protocols of different remote sensing devices, including satellite data reception protocols, UAV image transmission protocols, and IoT sensor communication protocols;

[0035] The data caching component is used to establish a multi-level data caching mechanism, which stores data in a hierarchical manner according to data priority to ensure the stability of data reception.

[0036] The time synchronization component is used to synchronize multi-source remote sensing data using the network time protocol to ensure the consistency of time series data.

[0037] Preferably, the image segmentation unit includes:

[0038] The preprocessing component is used to perform Gaussian filtering on standardized remote sensing images to reduce noise and enhance image edge features;

[0039] The segmentation algorithm component employs a multi-resolution segmentation algorithm, combining spectral and spatial features to merge regions.

[0040] The boundary optimization component is used to perform morphological processing on the segmentation boundaries, eliminating jagged edges and generating smooth crop region boundaries.

[0041] Preferably, the consensus verification unit includes:

[0042] The node management component is used to maintain a list of validator nodes in the blockchain network and record node reputation values.

[0043] The voting mechanism component is used to organize validator nodes to vote on new blocks in multiple rounds and to tally the voting results.

[0044] The block confirmation component is used to add a digital signature to a new block after consensus is reached, thus confirming the block's validity.

[0045] Preferably, the classification generation unit includes:

[0046] A community identification component is used to identify densely connected subgraphs in a label association network using the Leiden algorithm.

[0047] Hierarchical building components are used to construct a classification hierarchy based on the connection strength between subgraphs;

[0048] The index generation component is used to generate a unique identifier for each category node and establish a category index mapping table.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] The remote sensing data acquisition module performs standardized preprocessing on raw image data acquired by multi-source agricultural remote sensing equipment. Radiometric correction, atmospheric correction, and geometric correction operations eliminate the effects of equipment imaging differences, atmospheric interference, and geometric distortion, respectively, transforming raw data acquired from different sources and under different conditions into standardized remote sensing images with a unified format. This processing method ensures consistency and comparability of the scattered and disorganized raw data, providing a standardized and high-quality data source for subsequent feature extraction and classification analysis, and avoiding analytical biases caused by inconsistent data formats and varying quality.

[0051] The feature extraction module employs a multi-scale segmentation process, enabling accurate identification of crop feature regions in standardized remote sensing images. It comprehensively extracts texture, spectral, and shape features from each region, fusing them to form a multi-dimensional feature vector. This multi-dimensional, multi-scale feature extraction mode overcomes the limitations of traditional single-dimensional, single-scale extraction, fully preserving key information about crop growth status and more comprehensively and realistically reflecting crop characteristics. This provides rich foundational data for subsequent data classification and correlation analysis.

[0052] The blockchain storage module generates unique feature fingerprints for multi-dimensional feature vectors through hash operations. These fingerprints are then combined with timestamps and the hash values ​​of previous blocks to construct new blocks. After verification through a consensus mechanism, these blocks are added to the blockchain, forming an agricultural data chain. The distributed storage architecture eliminates reliance on a single central node; each participating node maintains a complete copy of the data. Even if some nodes fail or are attacked, the integrity and availability of the data remain unaffected. The association between feature fingerprints, timestamps, and block hash values ​​ensures that the generation, flow, and modification of every piece of data can be traced throughout the entire process. Any attempt to tamper with data will be detected by disrupting the consistency of the hash values, technically guaranteeing the authenticity and integrity of the data. Simultaneously, encryption is applied throughout the entire data storage process, effectively preventing security risks such as data leakage and unauthorized access, and protecting the legitimate rights and interests of stakeholders related to agricultural data.

[0053] The path analysis module extracts the storage paths of feature vectors by traversing the block sequence in the blockchain, analyzes the intersection nodes of different paths and calculates path similarity, and filters out a set of highly correlated feature paths. This process delves into the potential relationships between feature data scattered throughout the blockchain, connecting originally isolated feature data into a logically related network, revealing the intrinsic connections between different characteristics of crops, and extending the value of data beyond single pieces of information to the deep patterns contained in feature associations.

[0054] The intelligent classification module extracts corresponding crop feature tags from terminal blocks in the feature path set, and determines the correlation between feature tags through a multi-dimensional matching algorithm to generate an agricultural data classification structure. This classification method abandons the limitations of traditional single-feature matching, and is based on multi-dimensional feature association, making the classification results more consistent with the actual characteristics and growth patterns of crops, and the classification structure is more logical and practical. Users can quickly locate the required data based on the classification structure without having to sift through massive amounts of data one by one, improving the convenience of data retrieval and use. Attached Figure Description

[0055] Figure 1 This is a schematic diagram illustrating the working principle of the blockchain-based remote sensing agricultural big data management system described in this invention.

[0056] Figure 2 A flowchart illustrating the operation of the remote sensing data acquisition module;

[0057] Figure 3 A flowchart illustrating the operation of the feature extraction module;

[0058] Figure 4 A visualization of the blockchain data structure and proof-of-work mechanism;

[0059] Figure 5 This is a visualization diagram for feature storage path similarity analysis and clustering. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Please see Figure 1 This invention provides a blockchain-based remote sensing agricultural big data management system, the system comprising:

[0062] By integrating remote sensing technology, image processing, blockchain storage, and intelligent analysis, the system achieves reliable management and intelligent classification of agricultural data. The system includes a remote sensing data acquisition module, a feature extraction module, a blockchain storage module, a path analysis module, and an intelligent classification module. The remote sensing data acquisition module receives raw image data collected by multi-source agricultural remote sensing equipment and performs standardized preprocessing operations on the raw image data, including radiometric correction, atmospheric correction, and geometric correction, to generate standardized remote sensing images. The feature extraction module performs multi-scale segmentation processing on the standardized remote sensing images, identifies crop feature regions in the images, extracts texture feature values, spectral feature values, and shape feature values ​​for each feature region, and fuses them to generate multi-dimensional feature vectors. The blockchain storage module performs hash operations on the multi-dimensional feature vectors to generate feature fingerprints, combines them with timestamps and previous block hash values ​​to construct new blocks, verifies them through a consensus mechanism, and adds them to the blockchain, forming an immutable agricultural data chain. The path analysis module traverses the block sequence in the blockchain, extracts the storage paths of the feature vectors, analyzes the intersection nodes between different paths, calculates path similarity, and filters a set of highly correlated feature paths. The intelligent classification module extracts corresponding crop feature labels from the terminal blocks in the feature path set, uses a multi-dimensional matching algorithm to determine the relationship between feature labels, and generates an agricultural data classification structure.

[0063] Example 1: See Figure 2In practical implementation, the data receiving unit of the remote sensing data acquisition module is responsible for establishing physical and logical connections with various agricultural remote sensing devices. The multi-source interface component of the data receiving unit is configured with multiple communication protocol parsers. The satellite data receiving protocol parser processes downlink data streams from Earth observation satellites, the UAV image transmission protocol parser decodes real-time video streams and still image data from UAV platforms, and the IoT sensor communication protocol parser collects continuous monitoring data uploaded from ground sensor networks. The multi-source interface component parses and encapsulates the received raw byte streams, converting them into a standardized data format defined internally by the system, unifying the data structure and metadata description. The data caching component manages a multi-level storage cache. High-priority data received in real-time, such as disaster early warning image data, is stored in a memory-based high-speed cache layer to ensure fast response, while low-frequency access historical data, such as periodic census data, is stored in a large-capacity hard disk-based storage layer. The data caching component dynamically adjusts the data distribution according to preset data classification and storage strategies to maintain a balance between the efficiency and capacity of the caching system. The time synchronization component runs a network time protocol client. It initiates a synchronization request to a reference time server deployed within the local area network to obtain a high-precision time reference signal. The component adds a unified timestamp to each incoming remote sensing data record, correcting time deviations caused by device clock drift or network transmission delays, thus providing a consistent timing basis for subsequent time-series data generation. In specific implementation, the correction processing unit executes a series of correction algorithms on the standardized format data transmitted from the data receiving unit. The radiometric calibration process reads the sensor calibration parameters embedded in the data file, converting the raw digital quantization values ​​recorded by the sensor into physically meaningful radiance values, eliminating inherent errors such as sensor photoelectric response nonlinearity and dark current noise. The atmospheric parameter inversion correction stage, based on the radiative transfer model, inverts key atmospheric parameters such as aerosol optical thickness and water vapor content from the image data. These parameters are used to construct an atmospheric correction model, calculating the scattering and absorption effects of the atmosphere on electromagnetic wave signals, and eliminating these atmospheric influences from the apparent reflectance to obtain the true surface reflectance. The geometric calibration process utilizes digital elevation model data. By constructing a rigorous imaging geometric model or a rational polynomial coefficient model, geometric calibration accurately locates the geographic coordinates of each pixel in the image, correcting image geometric distortions caused by platform attitude changes, Earth curvature, and terrain undulations, thus ensuring precise registration of the image with the standard map projection.

[0064] In practice, the quality assessment unit performs quantitative quality checks on the images output by the correction processing unit. The unit calculates the signal-to-noise ratio (SNR) of the image, obtained by analyzing the ratio of the statistical variance of pixel values ​​in a uniform region to the mean of the signal. It also calculates the spatial resolution, evaluated by analyzing the spread function of edge targets or point targets in the image. The unit compares the calculated SNR and spatial resolution values ​​with preset quality thresholds, setting a lower limit of 30 dB for SNR and an upper limit of 1 meter for spatial resolution. Image data with an SNR below 30 dB or a spatial resolution below 1 meter is marked as unqualified and transferred to a reprocessing queue. Image data meeting both threshold requirements is marked as a standardized remote sensing image and output to the feature extraction module. The unit generates a detailed quality assessment report, recording the detection values, judgment results, and timestamp information for each image. It is understandable that the multi-source interface component of the data receiving unit needs to have protocol extension capabilities. This component adapts to the data transmission protocols of new remote sensing devices that may emerge in the future by loading different protocol driver plugins. The multi-level caching mechanism of the data caching component uses the Least Recently Used (LRU) algorithm for data replacement scheduling. The data caching component monitors the access frequency and response time of each level of storage and dynamically adjusts caching strategy parameters to optimize overall data throughput performance. When the time synchronization component cannot connect to an external reference time server, it can switch to using an internal high-stability crystal oscillator to maintain a short-term time reference and re-synchronize after network recovery.

[0065] It is understandable that the execution flow of the calibration processing unit is configurable, allowing users to select radiometric calibration models, atmospheric correction models, and geometric correction models of varying complexity based on specific application scenarios. For atmospheric parameter inversion correction, in the absence of field atmospheric measurement data, the calibration processing unit can employ approximate correction using the dark pixel method or dense vegetation method based on the statistical characteristics of the image itself. The geometric fine correction stage can select to use digital elevation model data of varying precision. For plain areas, a lower-precision global digital elevation model can be used, while for mountainous areas, a higher-precision regional high-precision digital elevation model is required.

[0066] Optionally, the quality assessment unit can integrate more image quality evaluation indicators, such as radiometric resolution, spectral distortion, and cloud cover. The unit can support user-defined combinations of quality thresholds and allows for differentiated quality acceptance standards for remote sensing data from different sources and time phases. The quality assessment report can be output in a structured data format, facilitating integration into the upper-level data management system for traceability and statistical analysis. Optionally, the entire remote sensing data acquisition module can be deployed on distributed computing nodes, simultaneously receiving, correcting, and evaluating multiple remote sensing images through a parallel processing pipeline. The module's operational status and performance indicators are monitored in real time, including key indicators such as data reception rate, correction processing time, and image pass rate. The module's log system records detailed operation logs and error information, providing support for system maintenance and fault diagnosis.

[0067] In practice, the geometric fine-calibration stage processes the surface reflectance image output from the atmospheric correction step. The core task of this stage is to convert image pixel coordinates into geographic coordinates by constructing a precise mathematical model. The geometric fine-calibration stage reads auxiliary data accompanying the remote sensing data file. This auxiliary data includes platform orbit parameters, attitude parameters, and in-lens orientation elements at the time of sensor imaging. The stage uses these parameters to construct a rigorous imaging geometry model. Based on the collinearity equation principle, the rigorous imaging geometry model correlates image point coordinates, sensor position and attitude, and ground point coordinates through strict mathematical relationships. The model solution process involves complex coordinate system transformations, including transformation chains between the sensor coordinate system, platform coordinate system, geocentric coordinate system, and map projection coordinate system. In practice, when the remote sensing data provider does not provide complete rigorous imaging geometry model parameters, the geometric fine-calibration stage can use a rational polynomial coefficient model as an alternative. The rational polynomial coefficient model approximates the mapping relationship between image point coordinates and ground coordinates using a set of polynomial coefficients. It does not directly rely on physical imaging parameters but rather obtains polynomial coefficients by fitting a large amount of ground control point data. The rational polynomial coefficient model exhibits high versatility and computational efficiency. In the geometric fine correction stage, digital elevation model (DEM) data is used in the correction calculations. DEM data provides elevation information for the correction area and is used to compensate for image point displacement effects caused by terrain undulations, achieving true orthorectification.

[0068] Example 2: See Figure 3In practical implementation, the image segmentation unit of the feature extraction module receives standardized remote sensing images from the remote sensing data acquisition module. The image segmentation unit uses a watershed algorithm to segment the standardized remote sensing images into regions to identify the boundaries of crop planting areas. The preprocessing component of the image segmentation unit performs Gaussian filtering on the input standardized remote sensing images for noise reduction. Gaussian filtering smooths random noise in the image by convolving it with a convolution kernel, while relatively preserving edge information. After the preprocessing component, the segmentation algorithm component begins its work. This component employs a multi-resolution segmentation algorithm, a region merging-based segmentation method. Starting from a single pixel, the multi-resolution segmentation algorithm gradually merges adjacent regions based on the similarity criteria of spectral and spatial features. During the merging process, the segmentation algorithm component calculates the spectral and shape heterogeneity of the regions. When the heterogeneity metric between regions falls below a set merging threshold, the segmentation algorithm component performs the merging operation. The image processed by the multi-resolution segmentation algorithm generates an initial segmentation map containing multiple homogeneous regions. The boundary optimization component then performs morphological processing on the region boundaries of the initial segmentation map. The boundary optimization component applies morphological opening operations to eliminate small false boundaries and burrs, and applies morphological closing operations to fill small holes and breaks on the region boundaries. Finally, the boundary optimization component generates independent crop units with smooth boundaries.

[0069] In practical implementation, the feature calculation unit performs feature quantization on each independent crop unit output by the image segmentation unit. The feature calculation unit calculates the gray-level co-occurrence matrix (GLCM) texture features for each crop unit. The GLCM extracts texture information by analyzing the co-occurrence frequency of pixel pairs of gray values ​​under specific spatial relationships. The feature calculation unit calculates statistics such as contrast, correlation, homogeneity, and entropy from the GLCM. Simultaneously, the feature calculation unit extracts the spectral features of the crop units, including multiple vegetation indices such as the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EGI). Spectral features highlight the spectral response characteristics of vegetation by combining reflectance values ​​from different bands. The feature calculation unit also calculates the morphological shape features of the crop units, including parameters such as perimeter, area, perimeter-to-area ratio, compactness, aspect ratio, and shape index. Shape features are used to describe the geometric morphology and spatial structure characteristics of crop patches.

[0070] In specific implementation, the vector generation unit integrates various feature values ​​output by the feature calculation unit. The vector generation unit first normalizes the texture, spectral, and shape feature values. Normalization linearly transforms feature values ​​with different dimensions and ranges to a unified numerical interval, such as [0,1]. The vector generation unit uses a min-max normalization method, calculating the minimum and maximum values ​​for each feature dimension and then mapping the original feature values ​​to the target interval through linear transformation. After normalization, the vector generation unit weights and fuses different types of feature values ​​according to preset weights. Appropriate weight coefficients are assigned to texture, spectral, and shape features, reflecting the relative importance of different feature types in crop identification. The vector generation unit concatenates and combines the weighted feature values ​​in a fixed order to generate a fixed-dimensional multi-dimensional feature vector. This multi-dimensional feature vector, as a digital representation of the crop unit, is output to the blockchain storage module. In some embodiments, the multi-resolution segmentation algorithm of the image segmentation unit includes scale parameter selection, heterogeneity calculation, and region merging loop. The scale parameter selection step determines the overall coarseness of the segmentation. The scale parameter is a comprehensive threshold that controls the termination condition for region merging. The heterogeneity calculation step evaluates the spectral and shape heterogeneity among candidate regions in each merging iteration. Spectral heterogeneity is measured by calculating the standard deviation of spectral values ​​between regions, while shape heterogeneity is measured by comparing the compactness or smoothness of regions. The region merging loop step starts at the pixel level and iteratively searches for neighboring region pairs with the lowest heterogeneity. Merging is performed when heterogeneity falls below the scale parameter, and the process terminates when no more regions meet the merging criteria.

[0071] In some embodiments, the gray-level co-occurrence matrix texture feature extraction process of the feature calculation unit includes matrix generation, feature calculation, and parameter optimization. The matrix generation step constructs the gray-level co-occurrence matrix by defining the spatial relationships between pixel pairs. This step requires specifying distance and orientation parameters, typically considering four directions: 0°, 45°, 90°, and 135°. The feature calculation step calculates a series of second-order statistics from the generated gray-level co-occurrence matrix, including contrast calculation, correlation calculation, homogeneity calculation, and entropy calculation. The parameter optimization step searches for the most distinguishable combination of texture features for different crop types by experimenting with different distance parameters and quantization levels. It is understood that the watershed algorithm of the image segmentation unit is very sensitive to image gradients and is prone to oversegmentation due to noise or local texture changes. The watershed algorithm usually needs to be used in conjunction with a preprocessing step, using Gaussian filtering preprocessing to suppress spurious gradient extrema caused by noise. The label-controlled improvement of the watershed algorithm can alleviate the oversegmentation problem. The label-controlled improvement of the watershed algorithm constrains the formation of catchment basins through prior knowledge or preprocessing results.

[0072] It is understandable that the selection of spectral features for the feature calculation unit needs to consider the crop growth cycle and species characteristics, as different crops have unique spectral response patterns at different growth stages. The feature calculation unit needs to select the most discriminative combination of vegetation indices based on the specific application scenario. The feature calculation unit may simultaneously calculate multiple vegetation indices and then select the optimal subset using feature selection methods. The shape features of the feature calculation unit are highly sensitive to spatial resolution; shape features provide more detailed information in high-resolution images, but their discriminative power may decrease in medium- and low-resolution images.

[0073] Optionally, the feature weighting scheme of the vector generation unit can employ either a static or dynamic weight allocation strategy. A static weight allocation strategy assigns fixed weight coefficients to different feature types based on domain expert experience; while simple to implement, it may not be suitable for all scenarios. A dynamic weight allocation strategy automatically learns the optimal weights based on the statistical characteristics of the training samples, optimizing feature weights through machine learning algorithms to maximize classification accuracy. The vector generation unit can also integrate feature selection functionality, filtering redundant or irrelevant features by evaluating the correlation between features and class labels, thus reducing the dimensionality of the feature vectors. Optionally, the entire feature extraction module can be configured with multiple sets of preset parameters to adapt to different remote sensing data sources and application requirements. The feature extraction module can set one set of parameters for high-resolution UAV imagery and another set for medium-resolution satellite imagery. The parameter configuration of the feature extraction module includes segmentation scale, feature type, normalization method, and weight settings, allowing users to flexibly adjust these parameters through configuration files or a graphical interface. The feature extraction module's operation log records the parameters used in each processing iteration and intermediate results, facilitating result traceability and parameter tuning.

[0074] In practical implementation, the preprocessing component of the image segmentation unit employs a separable convolution method to optimize the computational efficiency of Gaussian filtering. Separable convolution decomposes the two-dimensional Gaussian kernel into two consecutive one-dimensional kernels. The multi-resolution segmentation algorithm of the image segmentation unit efficiently manages the adjacency relationships between regions through a region adjacency graph data structure, which supports fast lookup and merging operations. The boundary optimization component of the image segmentation unit uses structuring element size parameters to control the intensity of morphological operations; these parameters are set according to the image spatial resolution and the size of the target features.

[0075] In practical implementation, the gray-level co-occurrence matrix calculation of the feature calculation unit adopts gray-level compression technology to balance computational complexity and texture description capability, quantizing the original 256 gray levels into fewer levels. The spectral feature calculation of the feature calculation unit optimizes the band operation expression, avoiding redundant calculations of intermediate results to improve efficiency. The shape feature calculation of the feature calculation unit uses a boundary tracking algorithm to accurately extract the region contour, calculating the perimeter and various shape indicators based on the contour point sequence. In practical implementation, the normalization processing of the vector generation unit adopts a batch normalization strategy. The vector generation unit collects a batch of training samples to statistically determine the global minimum and maximum values, using global statistics rather than single-sample statistics for normalization to improve consistency. The weight fusion step of the vector generation unit supports real-time adjustment of weight coefficients, updating weight values ​​through a configuration interface without recalculating feature values. The multidimensional feature vectors output by the vector generation unit are stored in an efficient binary format, reducing storage space and transmission overhead. The feature extraction module ultimately generates a multidimensional feature vector that creates a comprehensive digital fingerprint for each crop unit, containing texture, spectral, and shape information. This multidimensional feature vector provides high-quality input data for subsequent blockchain storage and intelligent classification.

[0076] Example 3: In a specific implementation, the hash operation unit of the blockchain storage module receives a multi-dimensional feature vector from the feature extraction module. The hash operation unit uses the SHA-256 algorithm to perform hash calculations on the multi-dimensional feature vector to generate a unique feature fingerprint. The SHA-256 algorithm converts input data of arbitrary length into a fixed-length 256-bit binary hash value through steps such as padding, message segmentation, and loop operation. The hash operation unit first performs standardized serialization processing on the input multi-dimensional feature vector, converting the multi-dimensional feature vector into a unique byte stream representation, eliminating hash inconsistencies caused by differences in data structure. The hash operation unit then performs SHA-256 hash calculations on the serialized byte stream. The hash calculation process includes stages such as message padding, message segmentation to initialize hash values, main loop processing, and final hash value generation. The hash operation unit finally outputs a 64-bit hexadecimal string as the digital fingerprint of the feature vector. This feature fingerprint has collision resistance and unidirectionality, ensuring that any small change in the original feature vector will lead to a significant change in the feature fingerprint.

[0077] The block building unit combines the feature fingerprints generated by the hashing unit with other metadata to construct a new block. The feature fingerprints collected by the block building unit, the data acquisition timestamp, the hash value of the previous block, and a random number (nonce) together constitute the basic elements of the new block. The block building unit organizes these elements according to a predefined block structure, which includes a block header and a block body. The block header contains fields such as version number, previous block hash value, timestamp, random number (nonce), and Merkle root hash value. The block body stores feature vector metadata and its serialized representation of feature fingerprints. The block building unit calculates the Merkle root hash value of all feature fingerprints in the block and fills the corresponding field in the block header with the Merkle root hash value. The block building unit adjusts the value of the random number (nonce) so that the hash value of the entire block header meets a specific difficulty target requirement; this process is called proof-of-work. The difficulty target requirement can be expressed as a mathematical inequality:

[0078]

[0079] in: This represents the SHA-256 hash value of the block header. This represents the current network difficulty target value. After finding a random number (nonce) that meets the conditions, the block building unit broadcasts the complete new block to the blockchain network for verification.

[0080] The consensus verification unit uses an improved Byzantine fault-tolerant algorithm to verify new blocks, achieving distributed consensus through a node voting mechanism. The node management component of the consensus verification unit maintains a list of all verification nodes in the blockchain network, recording each node's network address, public key certificate, and reputation value. When a new block is broadcast to the network, the voting mechanism component organizes multiple rounds of voting by verification nodes, ensuring that honest nodes in the network reach a consensus on the validity of the new block. During the voting process, each verification node independently verifies the correctness of the new block's digital signature, fingerprint, timestamp, and the matching of the hash value of the preceding block. Verification nodes indicate their approval or rejection of the new block through signed voting messages. After receiving approval votes from more than two-thirds of the verification nodes, the block confirmation component considers the new block to have reached consensus. The block confirmation component adds a valid digital signature to the new block to confirm its legitimacy and adds the new block to the end of the blockchain, forming an immutable data chain.

[0081] In some embodiments, the specific implementation of SHA-256 hash calculation includes message preprocessing, hash initialization, compression function iteration, and result output. The message preprocessing step pads the input byte stream to make its length a multiple of 512 bits, adding padding bits and original data length information to the end of the data. The hash initialization step initializes eight 32-bit hash variables, which are taken from the first 32 decimal places of the main hash constant. The compression function iteration step divides the padded message into 512-bit message blocks, performing 64 rounds of compression function processing on each message block, using a different constant and message scheduling table in each round. The result output step concatenates the last eight hash variables to generate the final 256-bit hash value. In some embodiments, the proof-of-work process of the block building unit involves dynamic adjustment of the difficulty target and a search strategy for a random number nonce. The difficulty target is a 256-bit value that specifies that the block header hash value must be less than or equal to the target value to be valid. The search strategy for random numbers (nonces) typically employs a sequentially increasing approach. Block building units start from 0 and progressively increase the nonce value, calculating the hash value of the entire block header each time and comparing it to the difficulty target. The block building unit can optimize the search process by using parallel computing or hardware acceleration to increase the speed of nonce searching. Once a matching nonce is found, the block building unit immediately stops searching and broadcasts the complete block containing that nonce.

[0082] In some embodiments, the consensus process of the improved Byzantine Fault Tolerance algorithm includes four phases: request, pre-preparation, preparation, and commit. In the request phase, the master node receives a new block proposal and broadcasts a pre-preparation message to backup nodes. In the pre-preparation phase, backup nodes verify the legality of the pre-preparation message and the validity of its signature; upon successful verification, they broadcast a preparation message. In the preparation phase, nodes collect enough preparation messages and then enter the commit phase, broadcasting a commit message. In the commit phase, after receiving enough commit messages, nodes confirm the new block and write it to their local blockchain. The improved Byzantine Fault Tolerance algorithm ensures that all honest nodes agree on the order and content of the new block through a three-phase protocol, guaranteeing the system's liveness and security even if some nodes fail or engage in malicious behavior.

[0083] It is understandable that the deterministic nature of the SHA-256 hash algorithm guarantees that identical input data will inevitably produce the same feature fingerprint, providing a foundation for data integrity verification. The collision resistance of the SHA-256 hash algorithm makes the probability of finding two different feature vectors that produce the same feature fingerprint extremely low, ensuring the uniqueness of feature fingerprints. The one-way nature of the SHA-256 hash algorithm prevents the possibility of retrieving the original feature vector from the feature fingerprint, protecting the privacy of sensitive agricultural data. It is also understandable that the Proof-of-Work mechanism balances block generation speed and network security through its computational difficulty setting. The high computational cost of the Proof-of-Work mechanism effectively prevents malicious nodes from tampering with the blockchain. The difficulty target of the Proof-of-Work mechanism needs to be dynamically adjusted according to network computing power, maintaining the relative stability of the block generation interval. The Proof-of-Work mechanism, combined with an improved Byzantine fault-tolerant algorithm, improves transaction processing efficiency while ensuring security, making it suitable for the needs of agricultural big data management scenarios.

[0084] Optionally, the blockchain storage module supports flexible configuration of various consensus algorithms, allowing the selection of a suitable consensus mechanism based on network size and security requirements. For small-scale private blockchain networks, the module can be configured to use a simple signature-based voting mechanism to improve efficiency. For large-scale public blockchain networks, the module needs to be configured to use consensus mechanisms resistant to Sybil attacks, such as Proof-of-Work or Proof-of-Stake. The blockchain storage module provides interface specifications for consensus algorithm plugins, allowing for adaptation to different application scenarios by loading different consensus algorithm plugins.

[0085] In practical implementation, the node management component evaluates node reputation based on historical behavior records. It tracks each validator node's participation in voting, the correctness of its votes, and the network's online stability. The component assigns higher reputation values ​​to nodes with a long history of good performance and lowers the reputation value or removes nodes with a history of malicious behavior from the node list. The voting mechanism component employs a multi-round voting timeout mechanism to prevent consensus deadlock due to network latency or node failures. The block confirmation component uses an elliptic curve digital signature algorithm to sign confirmed blocks, ensuring the verifiability of block origins and the non-repudiation of content. The blockchain storage module periodically performs blockchain compression and archiving operations, moving early, infrequently accessed block data to cold storage and optimizing the utilization efficiency of online storage space. The blockchain storage module provides a data retrieval interface, supporting rapid queries of agricultural data records in the blockchain based on criteria such as feature fingerprints, time ranges, or geographical locations.

[0086] See Figure 4 This visualization showcases the core working mechanism of the blockchain storage module in agricultural remote sensing data management. The charts visually represent the linear expansion process of the blockchain through the increasing block height trend, while the mining difficulty curve demonstrates the dynamic network security guarantee provided by the proof-of-work mechanism. The hash value distribution characteristics verify the collision resistance and determinism of the SHA-256 algorithm, and the unique color identifier of each block reflects the uniqueness of the digital fingerprint. The entire visualization fully demonstrates the superior performance of blockchain technology in ensuring the integrity and tamper-proof nature of agricultural data, providing a reliable storage foundation for massive amounts of remote sensing data and supporting the stable operation of key agricultural applications such as crop monitoring and growth analysis.

[0087] Example 4: In specific implementation, the path extraction unit of the path analysis module reads complete blockchain data from the blockchain storage module. The path extraction unit reconstructs the storage path sequence of feature vectors by parsing the header information of each block. The path extraction unit sequentially traverses the blockchain starting from the genesis block, recording the block hash value, timestamp, and reference relationship in previous blocks corresponding to each feature vector. The path extraction unit maintains a path index table, which maps the unique identifier of each feature vector to its position sequence in the blockchain. The path extraction unit uses a graph data structure to store path information, representing each block as a node in the graph and the hash reference relationship between blocks as directed edges. The path extraction unit traverses the entire blockchain graph using a depth-first search or breadth-first search algorithm, identifying all complete paths from the genesis block to the last block. The path extraction unit generates a complete storage path record for each feature vector, containing information such as the hash value sequence, timestamp sequence, and path length of all blocks in the path. The similarity calculation unit receives the feature storage path set generated by the path extraction unit. It then uses a dynamic time warping algorithm to calculate the similarity between different paths. Dynamic time warping can handle path sequences of varying lengths and measures similarity by finding the optimal alignment path between two sequences. The similarity calculation unit first converts the path sequence into a feature vector sequence, with each path point corresponding to a feature vector containing attributes such as block timestamps, path depth, and branch factors. Next, it constructs a cumulative cost matrix, filling each element of the matrix using a dynamic programming algorithm to find the alignment path with the minimum cumulative cost. Finally, the unit normalizes the minimum cumulative cost to a similarity score between 0 and 1; a similarity score closer to 1 indicates greater similarity between the two paths. The similarity calculation unit supports various distance metrics, including Euclidean distance, Manhattan distance, and cosine distance, selecting the appropriate metric based on the characteristics of the path features.

[0088] Referring to Table 1, the path filtering unit filters key paths based on the output of the similarity calculation unit. The path filtering unit sets a configurable similarity threshold, marking path pairs with similarity scores higher than this threshold as highly correlated path pairs. The path filtering unit constructs a path similarity graph, where nodes represent feature storage paths and edges represent similarity relationships between paths. The path filtering unit applies a graph clustering algorithm to identify densely connected subgraphs in the path similarity graph, grouping all paths belonging to the same subgraph into a key path set. The path filtering unit generates summary statistics for each key path set, including set size, average similarity, and center path. The path filtering unit outputs the key path sets to the intelligent classification module for further analysis.

[0089] Table 1: Path Similarity Calculation Parameter Configuration Table

[0090]

[0091] In some embodiments, the blockchain traversal process of the path extraction unit employs multi-threaded parallel processing technology to improve efficiency. The path extraction unit divides the blockchain into multiple segments according to time range, and allocates an independent processing thread for each segment to extract paths. The path extraction unit uses a thread-safe queue structure to store intermediate results, and merges the extraction results from each sub-thread in the main thread. The path extraction unit implements an incremental update mechanism; when a new block is added to the blockchain, the path extraction unit only performs path analysis on the newly added block, and integrates the new path with the existing path index. The path extraction unit provides a path query interface, supporting the rapid retrieval of relevant paths based on conditions such as time range, path length, or specific blocks.

[0092] It is understandable that the Dynamic Time Warping (VTW) algorithm is robust to local time distortions in path sequences and can effectively handle path time offsets caused by blockchain forks or network latency. The computational complexity of the VTW algorithm is proportional to the product of path lengths, and it may require significant computational resources when handling long paths. The VTW algorithm balances computational accuracy and efficiency by limiting the size of the warping window, which restricts the maximum time offset of path alignment. Optionally, the critical path set generation process of the path filtering unit supports multiple clustering algorithm configurations. The path filtering unit can be configured to use the density-based DBSCAN clustering algorithm, which can automatically discover clusters of arbitrary shapes and identify noise points. The path filtering unit can be configured to use the hierarchical agglomerative clustering algorithm, which generates tree-like cluster structures for multi-granularity analysis. The path filtering unit can be configured to use the spectral clustering algorithm, which utilizes the eigenvectors of the graph Laplacian matrix for dimensionality reduction clustering. The path filtering unit provides an interface for tuning clustering parameters, allowing users to adjust the clustering granularity according to specific needs.

[0093] Optionally, the path analysis module includes comprehensive performance monitoring and logging capabilities. It monitors performance metrics such as path extraction throughput, similarity calculation speed, and path filtering accuracy. The module records detailed operation logs, including the path analysis start time, the number of paths processed, similarity distribution statistics, and abnormal path detection results. It provides a visual debugging interface, displaying a visual representation of path similarity graphs, clustering results, and the set of critical paths. The module supports exporting analysis results, outputting the set of critical paths in a structured data format to a file or database.

[0094] In its implementation, the dynamic time warping algorithm of the similarity calculation unit incorporates several optimization measures. The similarity calculation unit uses a diagonal-filling strategy for the dynamic programming matrix to reduce memory usage. It employs an early termination technique, terminating the calculation prematurely when the accumulated cost exceeds a threshold. The similarity calculation unit implements a batch processing mode, calculating the similarity between multiple path pairs at once to leverage cache locality. It supports symmetry optimization of the similarity matrix, avoiding redundant calculations of similarity values ​​at symmetrical positions. In its implementation, the post-processing of the critical path set in the path filtering unit includes path representativeness selection and set quality evaluation. The path filtering unit selects a central path from each critical path set as the representative of that set, choosing the path with the highest average similarity to other paths. The path filtering unit calculates the internal cohesion and external separation indices for each critical path set to evaluate the set quality. The path filtering unit further optimizes low-quality sets by segmenting or merging them, thereby improving the discriminative power of the critical path sets. The path filtering unit establishes a mapping relationship between the key path set and the original feature vectors, providing a complete analytical foundation for the intelligent classification module. The path analysis module, through the system's path extraction, similarity calculation, and path filtering processes, mines highly correlated feature path patterns from blockchain data, providing crucial structural information for the intelligent classification of agricultural data.

[0095] In its implementation, the process of the path filtering unit applying graph clustering algorithms to identify densely connected subgraphs in the path similarity graph begins with the construction of the similarity graph. The path filtering unit represents the path similarity graph as a weighted undirected graph structure, where each node corresponds to a feature storage path, and the edges between nodes store the similarity scores of the two paths. The path filtering unit filters out edges with lower weights based on a preset similarity threshold, retaining only meaningful connections with similarity scores exceeding the threshold. The path filtering unit then calculates the degree centrality and weighted degree centrality of each node, identifying core and edge nodes in the network.

[0096] See Figure 5This paper presents the results of in-depth analysis of feature storage paths by the path analysis module. Through multi-dimensional visualization techniques, the charts clearly show the distribution patterns of different paths in terms of time span, complexity, and similarity. Paths 0, 3, 6, 9, 12, 15, and 18 in the chart are labeled at fixed intervals. These paths represent different feature storage paths in the scatter plot, with paths 0, 3, 6, 9, 12, 15, and 18 labeled every three paths to avoid overcrowding. Each path has unique features such as time span, complexity, similarity score, and path length, visualized through the position of the points (X-axis: time span, Y-axis: complexity), color (similarity score), and size (path length). Differences in path length reflect the temporal characteristics of data storage, while color intensity intuitively shows the strength of the association between paths. The analysis results validate the effectiveness of the dynamic time warping algorithm in processing blockchain path sequences, successfully identifying a set of highly similar feature paths. These critical path patterns provide important structural basis for subsequent intelligent classification, supporting the construction of a multi-level classification system for agricultural data and precision agricultural decision analysis.

[0097] Example 5: In specific implementation, the tag extraction unit of the intelligent classification module obtains terminal block information from the key path set provided by the path analysis module. The tag extraction unit parses the block body data content of the terminal block to extract the stored crop feature tags. The tag types processed by the tag extraction unit include three major categories: crop variety tags, growth cycle tags, and pest and disease tags. Crop variety tags record crop type information such as rice, corn, and wheat; growth cycle tags record growth stage information such as sowing period, seedling stage, heading period, and maturity period; and pest and disease tags record pest and disease type information such as rust, aphids, and rice planthoppers. The tag extraction unit parses and decodes the raw data in the block body, converting binary or JSON format serialized data into structured tag objects. The tag extraction unit performs data cleaning and standardization processing, merging synonym tags, expanding abbreviation tags to full names, and eliminating inconsistencies in tag descriptions. The tag extraction unit assigns a confidence score to each tag, which comes from the quality assessment results or expert verification information in the feature extraction stage. The label extraction unit establishes a source association between the label and the original feature vector, and records the block hash value and path identifier corresponding to each label.

[0098] The association analysis unit receives structured label data output by the label extraction unit and constructs a network of association rules between labels. It employs a frequent itemset mining algorithm to analyze label co-occurrence patterns, calculating the frequency and support of different label combinations appearing together in terminal blocks. The unit sets minimum support and minimum confidence thresholds and filters out statistically significant strong association rules. It constructs a label association network graph, representing each label as a network node and the relationships between labels as edges, with edge weights determined by the confidence or lift of the association rule. The unit calculates centrality indices for network nodes, including degree centrality, betweenness centrality, and eigenvector centrality, identifying key label nodes in the network. Finally, the unit discovers hierarchical and parallel relationships between labels, providing a relational foundation for subsequent classification system construction.

[0099] The classification generation unit, based on the label association network output by the association analysis unit, employs a community detection algorithm to identify densely connected subgraph structures within the network. The community identification component of the classification generation unit applies the Leiden algorithm for community detection, which divides the network into communities by optimizing the modularity function. The community identification component calculates the gain value of node movement and reassigns nodes to different communities to improve the overall modularity. It performs iterative community aggregation and refinement until the modularity no longer significantly improves. The hierarchical construction component builds a classification hierarchy based on the community detection results, analyzing the connection strength between different communities. It merges tightly connected communities into higher-level classification nodes, forming a tree-like hierarchical classification system. The index generation component generates a unique identifier for each node in the classification system and establishes a mapping table between classification nodes and their original labels. The index generation component implements serialization and persistent storage of the classification system, supporting version management and incremental updates of the classification tree.

[0100] In some embodiments, the data cleaning process of the tag extraction unit includes several standardized steps. The tag extraction unit accesses a domain ontology library, matching and mapping the extracted raw tags with standard terms in the ontology library. For tags that cannot be matched, the tag extraction unit initiates a manual review process, submitting new tags to domain experts for standardized naming. The tag extraction unit maintains a tag thesaurus, automatically merging variant expressions into standard tags. The tag extraction unit performs unified translation processing on multilingual tags, converting foreign language tags into system-defined primary language expressions. In some embodiments, the frequent itemset mining of the association analysis unit uses the FP-Growth algorithm to optimize computational efficiency. The association analysis unit constructs an FP-tree data structure to compress and store tag co-occurrence information. The association analysis unit avoids generating candidate itemsets through conditional pattern base mining. The association analysis unit uses a parallel computing framework to process large-scale tag data. The association analysis unit dynamically adjusts the minimum support threshold and adaptively sets mining parameters based on tag density.

[0101] It is understandable that the Leiden algorithm improves the quality of community segmentation during community detection by introducing a refinement stage. The Leiden algorithm first performs a fast move optimization to initially segment communities. Then, it performs a refinement move optimization to adjust community boundaries. Finally, it performs community aggregation iterations to improve modularity. Compared to the Louvain algorithm, the Leiden algorithm avoids the arbitrary connection problem, ensuring the tightness of connections within each community. Optionally, the classification generation unit can support multiple hierarchical construction strategies. The classification generation unit can use a bottom-up aggregation hierarchical clustering method, merging upwards layer by layer starting from the leaf nodes. The classification generation unit can use a top-down splitting hierarchical clustering method, recursively dividing from the root node. The classification generation unit can incorporate domain knowledge to constrain the clustering process, ensuring the rationality of the classification structure. The classification generation unit provides an interactive adjustment interface, allowing experts to manually adjust the automatically generated classification system.

[0102] Optionally, the intelligent classification module can integrate online learning capabilities to adapt to data changes. The intelligent classification module monitors the frequency and distribution changes of new labels. It periodically reruns the association analysis and classification generation processes. The intelligent classification module supports incremental updates to the classification system. It maintains the classification version history, supporting backtracking and comparison of differences between versions. Optionally, the implementation of the intelligent classification module includes a complete quality assessment and verification mechanism. The intelligent classification module calculates the internal consistency index of the classification system. It evaluates the stability of the classification through cross-validation. The intelligent classification module provides a visual representation of the classification results. It generates a classification quality report, recording key parameters and performance indicators during the classification process. In specific implementation, the hierarchical construction component's classification system generation includes multiple post-processing steps. The hierarchical construction component smooths the automatically generated classification tree, eliminating overfitting branch structures. It generates descriptive text for each classification node, extracting the most representative label combinations from association rules as node descriptions. It establishes referential relationships between classification nodes and handles cross-classification and multiple attribution cases. The hierarchical construction component optimizes the balance between depth and breadth in the classification tree, improving the usability of the classification system. In its implementation, the identifier design of the index generation component adopts a hierarchical encoding scheme. The index generation component assigns a specific encoding segment to each classification level. The hierarchical relationship of the classification is reflected through encoding. The index generation component ensures the uniqueness of identifiers across the entire network. The index generation component establishes a mapping relationship between identifiers and blockchain addresses. The index generation component supports rapid retrieval and traceability queries based on identifiers. The intelligent classification module, through the system's tag extraction, association analysis, and classification generation processes, organizes scattered crop feature tags into a structured classification knowledge system, providing intelligent classification support for agricultural data management.

[0103] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0104] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1.A blockchain-based remote-sensing agricultural big data management system, characterized in that, The system comprises: a remote sensing data acquisition module, configured to receive original image data collected by a multi-source agricultural remote sensing device, perform standardized preprocessing operations on the original image data, including radiation correction, atmospheric correction and geometric correction, and generate standardized remote sensing images; a feature extraction module, configured to perform multi-scale segmentation processing on the standardized remote sensing images, identify crop feature regions in the images, extract texture feature values, spectral feature values and shape feature values of each feature region, and fuse to generate multi-dimensional feature vectors; a blockchain storage module, configured to perform hash operation on the multi-dimensional feature vectors to generate feature fingerprints, combine a time stamp and a previous block hash value to construct a new block, add the new block to a blockchain after verification by a consensus mechanism, and form an agricultural data chain that is not tamperable; a path analysis module, configured to traverse a block sequence in the blockchain, extract storage paths of the feature vectors, analyze cross nodes between different paths, calculate path similarity, and screen a feature path set with high correlation; an intelligent classification module, configured to extract corresponding crop feature labels from terminal blocks in the feature path set, determine the correlation between the feature labels by using a multi-dimensional matching algorithm, and generate an agricultural data classification structure; the intelligent classification module comprises: a label extraction unit, configured to extract stored crop variety labels, growth cycle labels and pest and disease labels from terminal blocks of key paths; an association analysis unit, configured to construct association rules between the labels, calculate label co-occurrence frequencies, and establish a label association network; a classification generation unit, configured to identify label communities by using a community discovery algorithm according to the label association network, and generate a multi-level agricultural data classification system; the classification generation unit comprises: a community identification component, configured to identify densely connected subgraphs in the label association network by using a Leiden algorithm; a hierarchy construction component, configured to construct a classification hierarchy structure according to connection strengths between the subgraphs; an index generation component, configured to generate unique identifiers for each classification node and establish a classification index mapping table; the feature extraction module comprises: an image segmentation unit, configured to perform region segmentation on the standardized remote sensing images by using a watershed algorithm, identify crop planting region boundaries, and extract independent crop units; a feature calculation unit, configured to calculate a gray level co-occurrence matrix texture feature of each crop unit, extract spectral features of a normalized vegetation index and an enhanced vegetation index, and calculate shape features of a region perimeter, area, perimeter-to-area ratio, compactness, aspect ratio and shape index; a vector generation unit, configured to perform normalization processing on the texture feature values, the spectral feature values and the shape feature values, perform weighted fusion according to preset weights, and generate fixed-dimensional feature vectors. 2.The blockchain-based remote-sensing agricultural big data management system according to claim 1, wherein, the remote sensing data acquisition module comprises: a data receiving unit, configured to connect satellite remote sensing devices, unmanned aerial vehicle remote sensing devices and ground sensor networks, and receive multi-spectral image data, hyperspectral image data and thermal infrared image data; The correction processing unit is configured to perform radiation scaling processing on the received image data, eliminate sensor self-error, perform atmospheric parameter inversion correction, eliminate atmospheric scattering influence, implement geometric fine correction, and eliminate image distortion caused by terrain undulation. The quality evaluation unit is configured to calculate the signal-to-noise ratio and spatial resolution of the corrected image, set a quality threshold, and screen standardized remote sensing images meeting the quality requirements. 3.The blockchain-based remote-sensing agricultural big data management system according to claim 1, wherein, The blockchain storage module comprises: The hash operation unit is configured to perform SHA-256 hash calculation on the feature vector to generate a unique feature fingerprint. The block construction unit is configured to combine the feature fingerprint with a collection timestamp and a previous block hash value, add a random number, and generate a new block. The consensus verification unit adopts an improved Byzantine fault tolerance algorithm to verify the new block, and adds the new block to the end of the blockchain after reaching a consensus through a node voting mechanism. 4.The blockchain-based remote-sensing agricultural big data management system according to claim 1, wherein, The path analysis module comprises: The path extraction unit is configured to start from the starting block of the blockchain, record the storage location sequence of each feature vector, and generate a feature storage path. The similarity calculation unit is configured to calculate the similarity between different paths using a dynamic time warping algorithm and set a similarity threshold. The path screening unit is configured to screen key paths with a similarity exceeding the threshold based on the similarity calculation result, and establish a key path set. 5.The blockchain-based remote-sensing agricultural big data management system according to claim 2, wherein, The data receiving unit comprises: The multi-source interface component is configured to adapt to different remote sensing device data transmission protocols, including satellite data receiving protocols, unmanned aerial vehicle image transmission protocols, and Internet of Things sensor communication protocols. The data cache component is configured to establish a multi-level data cache mechanism, perform hierarchical storage according to data priority, and ensure the stability of data reception. The time synchronization component is configured to use the network time protocol to synchronize the time of multi-source remote sensing data, and ensure the consistency of time series data. 6.The blockchain-based remote-sensing agricultural big data management system according to claim 1, wherein, The image segmentation unit comprises: The preprocessing component is configured to perform Gaussian filter denoising on the standardized remote sensing image to enhance image edge features. The segmentation algorithm component adopts a multi-resolution segmentation algorithm to combine spectral features and spatial features for region merging. The boundary optimization component is configured to perform morphological processing on the segmentation boundary to eliminate jagged edges and generate smooth crop region boundaries. 7.The blockchain-based remote-sensing agricultural big data management system according to claim 3, characterized in that, The consensus verification unit comprises: The node management component is configured to maintain a list of verification nodes in the blockchain network and record node reputation values. The voting mechanism component is configured to organize multiple rounds of voting on new blocks by verification nodes and count the voting results. The block confirmation component is configured to add a digital signature to the new block after reaching a consensus and confirm the validity of the block.

Images