Forestry remote sensing data ontology feature recognition method and system for blockchain right confirmation

Abstract

The invention relates to the technical field of computers, in particular to a block chain right-confirming forestry remote sensing data ontology feature recognition method and system, and the method comprises the steps: extracting the multi-scale features of forestry remote sensing image data through a pre-trained deep convolutional neural network, and generating a feature pyramid with an unchanged rotation and scale; a time-space fusion feature vector is generated by combining seasonal attention weight adjustment, and forestry remote sensing image fingerprints are formed after principal component analysis and dimension reduction; a fuzzy commitment is adopted to bind the random key after error correction coding with a forestry remote sensing image fingerprint to generate a block chain evidence storage commitment, and a storage architecture with on-chain evidence storage and under-chain metadata collaboration is constructed; and efficient right confirmation is realized. The problem that a traditional Hash method is sensitive to image deformation is solved, the problem of feature drift caused by seasonal changes is solved, verifiable right confirmation is achieved, the method has the advantages of being high in storage efficiency, high in response speed and the like, a credible ownership authentication basis is provided for forestry remote sensing data sharing, and the data circulation efficiency is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and specifically to a method and system for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification. Background Technology

[0002] Forestry remote sensing data is a common and frequently used source of forestry resource data. However, its application faces numerous challenges, primarily in areas such as difficulty in establishing data ownership, challenges in data sharing, and privacy breaches. Forestry remote sensing data is scattered across various departments, originating from different methods including manual collection, satellite remote sensing, and drone photography. The lack of clear data ownership is the first barrier to data sharing. The absence of a unified and integrated management platform creates departmental barriers when sharing data with departments such as ecology and environment and agriculture and rural affairs. Furthermore, forestry remote sensing data may involve national and regional secrets and personal privacy; secure and traceable sharing is crucial to realizing the value of forestry remote sensing data.

[0003] Current methods for determining data ownership primarily involve generating a hash value for the data and comparing it with the hash values ​​of already owned data in a database. However, in practical use, forestry remote sensing data often requires processing operations such as rotation, cropping, and scaling. This results in changes to the hash value, making hash value-based ownership determination unusable. Summary of the Invention

[0004] In view of the above problems, the present invention provides a method and system for identifying the ontological features of forestry remote sensing data with blockchain-based ownership confirmation, which solves the existing problem of ownership confirmation of forestry remote sensing data.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification, comprising: Forestry remote sensing image data is acquired, and multi-scale feature extraction is performed on the forestry remote sensing image data to generate a feature pyramid with rotation invariance and scale invariance. The multi-scale feature extraction is configured to be implemented using a pre-trained deep convolutional neural network. Hierarchical feature fusion is performed on the feature pyramid. By calculating the Euclidean distance difference vector of feature maps at different levels and adjusting the feature response in combination with seasonal attention weights, a spatiotemporal fusion feature vector is generated. The spatiotemporal fusion feature vectors are dimensionality reduced to obtain forestry remote sensing image fingerprints. The dimensionality reduction process is configured to use the principal component analysis algorithm. Construct a rights confirmation and evidence storage structure based on fuzzy commitment, generate a random key and perform error correction encoding on it, and XOR bind the error-corrected random key with the fingerprint of forestry remote sensing image to generate a blockchain evidence storage commitment. The blockchain notarization commitment and key hash value are written into the blockchain, while the corresponding remote sensing image metadata is stored off-chain. The remote sensing image metadata includes the acquisition time, geographical location and season identifier. When a rights confirmation query request is received, the ontological features of the remote sensing image to be confirmed are extracted and a query fingerprint is generated. The blockchain block where similar fingerprints are located is located through off-chain index. Perform fuzzy commitment verification in the located blockchain block, recover the error correction code through XOR operation and decode to obtain the original key, and complete the confirmation of rights verification by comparing the key hash value; The output includes ownership information and similarity scores for the ownership determination results. The ownership information includes the original data holder, the derived relationship chain, and the access authorization records.

[0006] In some embodiments, multi-scale feature extraction is performed on forestry remote sensing image data to generate a feature pyramid with rotation invariance and scale invariance, including: Forestry remote sensing image data is scaled to a standard size of 512×512 pixels using bicubic interpolation to obtain the scaled image. Zero-filling is performed on the scaled image to maintain the aspect ratio of the forest features, resulting in the first expanded image. The pixel values ​​of the first expanded image are converted to the range of [-1,1] by normalization processing to obtain the second expanded image; The second augmented image is input into the pre-trained ResNet-50 network to extract feature maps from four layers. The spatial resolutions of the four layers are 256×256, 128×128, 64×64 and 32×32, respectively. A deformable convolutional network is applied to the feature map of each layer, and geometric deformation adaptation is performed by learning the offset field to obtain the deformation-adapted feature map; An orientation-sensitive convolutional kernel is applied to the deformation adaptation feature map. The orientation-sensitive convolutional kernel is configured as a Gabor filter bank with 12 directions. The extreme values ​​of the feature response in 12 directions at each spatial location are calculated, and the orientation angle corresponding to the maximum response is recorded. The feature map is rotated and aligned based on the orientation angle corresponding to the maximum response to generate an orientation-normalized feature map; Normalized feature maps from each level are input into the feature pyramid network and fused through a top-down path and 3×3 convolutions to output feature pyramids with four scales: 256×256, 128×128, 64×64, and 32×32.

[0007] In some embodiments, hierarchical feature fusion is performed on the feature pyramid by calculating the Euclidean distance difference vector between feature maps at different levels, including: The feature maps of the four scales in the feature pyramid, 256×256, 128×128, 64×64 and 32×32, are input into a 1×1 convolutional layer to unify the channels, resulting in standardized feature maps with 256 channels each. Bilinear interpolation upsampling is performed on the standardized feature maps to adjust all feature maps to a uniform size of 256×256; Calculate the Euclidean distance difference vector between adjacent feature maps, including: Calculate the sum of squared differences in pixel values ​​at corresponding spatial locations between the 256×256 and 128×128 scale feature maps, and denote it as the first difference vector; Calculate the sum of squared differences in pixel values ​​at corresponding spatial locations between the 128×128 and 64×64 scale feature maps, and denote it as the second difference vector; Calculate the sum of squared differences in pixel values ​​at corresponding spatial locations between the 64×64 and 32×32 scale feature maps, and denote it as the third difference vector; The first difference vector, the second difference vector, and the third difference vector are input into a gated recurrent unit for time-series modeling to obtain multi-scale difference features. Based on multi-scale difference features, the feature response is adjusted by combining seasonal attention weights to generate a spatiotemporal fusion feature vector.

[0008] In some embodiments, a spatiotemporal fusion feature vector is generated by adjusting the feature response based on multi-scale difference features and seasonal attention weights, including: Construct seasonal prototype vectors containing the features of the four seasons: spring, summer, autumn, and winter. Each seasonal prototype vector has a dimension of 256. Calculate the cosine similarity between multi-scale difference features and prototype vectors of each season to generate a seasonal correlation score matrix; The seasonal relevance score matrix is ​​input into the Softmax function for normalization to obtain the seasonal attention weights; Apply corresponding seasonal attention weights to the first, second, and third difference vectors respectively to perform feature importance weighting; The weighted first difference vector, second difference vector, and third difference vector are added to the standardized feature map through skip connections and fused to obtain the fused feature map. The fused feature maps are refined using 3×3 depthwise separable convolutions. Spatial information is aggregated through a global average pooling layer, outputting a 256-dimensional spatiotemporal fusion feature vector.

[0009] In some embodiments, the spatiotemporal fusion feature vector is dimensionality reduced to obtain a forestry remote sensing image fingerprint, including: The spatiotemporal fusion feature vector is input into the principal component analysis network, which includes a feature standardization layer and a covariance calculation layer. The spatiotemporal fusion feature vector is normalized to zero mean by a feature standardization layer to obtain a standardized feature vector. Calculate the covariance matrix of the standardized eigenvectors, and perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and the corresponding initial eigenvectors. The first 128 initial eigenvectors are selected as principal component projection matrices in descending order of eigenvalues; A linear transformation is performed on the standardized eigenvectors and the principal component projection matrix to obtain a 128-dimensional intermediate eigenvector. The intermediate feature vector is binarized by converting the feature values ​​into 0 / 1 bits using a sign function; The 128 bits are concatenated in sequence to generate a 128-bit forestry remote sensing image fingerprint; Hamming distance verification is performed on forestry remote sensing image fingerprints to ensure the distinguishability between them.

[0010] In some embodiments, a fuzzy commitment-based ownership verification and storage structure is constructed, a random key is generated and error-correcting encoding is performed on it, including: A cryptographically secure pseudo-random number generator is used to generate a 256-bit raw random key; The original random key is input into the error correction encoder for encoding to obtain a 255-bit error correction encoded key. Calculate the SHA-256 hash value of the original random key as the key verification fingerprint; Hamming weight verification is performed on the error correction coding key to ensure that it meets the preset uniform distribution condition; When the Hamming weight does not meet the conditions, a random key is regenerated and the above encoding process is repeated until a verified error-corrected encoded key is generated, which is the error-corrected random key. Additionally, a mapping relationship is established between the verified error correction coding key and the key verification fingerprint, and the fingerprint is stored in the key management pool.

[0011] In some embodiments, the error-corrected coded random key is XORed with the forestry remote sensing image fingerprint to generate a blockchain-based notarization commitment, including: The fingerprint of the forestry remote sensing image is input into the fingerprint expansion module, and the image expansion fingerprint is generated through a cyclic shift operation. The error correction coding key is XORed with the image-extended fingerprint input and bound to the module. Perform an XOR operation bit by bit to generate the initial binding result; The initial binding result is verified using Hamming code to detect and correct possible binding errors, resulting in a verified binding result. The integrity of the verified binding result is checked using hash calculation, and a binding verification code is generated. The binding verification code is combined with the initial binding result to generate a blockchain-based evidence commitment.

[0012] In some embodiments, when a rights confirmation query request is received, the ontological features of the remote sensing image to be confirmed are extracted and a query fingerprint is generated. The blockchain block containing similar fingerprints is located using an off-chain index, including: Multi-scale feature extraction is performed on the remote sensing images for property rights confirmation to generate a query feature pyramid; The query feature pyramid is fused using a seasonal attention weighting algorithm to obtain a query spatiotemporal fusion feature vector. The spatiotemporal fusion feature vector of the query is reduced in dimensionality to generate a query fingerprint; An index structure based on an improved VP-Tree is built in the off-chain fingerprint database. The improved VP-Tree is configured to support Hamming distance metric. Centered on the query fingerprint, a radius expansion search is performed in the improved VP-Tree to obtain a set of candidate similar fingerprints; Sort the candidate similar fingerprint set by Hamming distance, and select the top few candidate similar fingerprints, which are denoted as identical fingerprints; Locate the blockchain block address corresponding to similar fingerprints using the blockchain transaction hash mapping table; Verify the timestamps and version identifiers of candidate blocks to ensure the timeliness and validity of block data; The output includes the matching degree between the identical fingerprint and the currently queried fingerprint, as well as the location result of the corresponding block information.

[0013] In some embodiments, fuzzy commitment verification is performed in the located blockchain block, the error-correcting code is recovered through XOR operation and decoded to obtain the original key, and the ownership verification is completed by comparing the key hash value, including: Extract blockchain-based evidence commitments from blockchain blocks; The blockchain-based notarization commitment is broken down into binding results and binding verification codes; The query fingerprint is expanded to generate an extended query fingerprint. The extended query fingerprint and the binding result are XORed to obtain the recovered error correction code; The recovered error-correcting code is decoded to obtain the candidate original key; Calculate the hash value of the candidate original key and compare it with the key hash value stored in the blockchain block to obtain the verification result; When the verification result shows a hash value match, the verification is considered successful and the ownership information retrieval process is triggered. When the verification result is that the hash values ​​do not match, the second-best candidate is selected from the set of candidate similar fingerprints and the verification process is repeated. The output includes ownership information and similarity scores, confirming ownership results. Ownership information includes the original data holder, derived relationship chains, and access authorization records, including: Retrieve the corresponding metadata index based on the successfully verified key hash value; Obtain remote sensing image metadata from off-chain storage, including acquisition time, geographic location, and season identifier; Construct a chain of ownership proofs that includes the identity identifier of the original data holder; Additionally, by querying the derivative relationship chain and access authorization records of forestry remote sensing image data through smart contracts, ownership information can be generated; Calculate the similarity score between the query fingerprint and the registered fingerprint, and generate a standardized similarity value between 0 and 1; The ownership information, similarity score, and verification timestamp are combined to generate structured ownership confirmation results; The structured rights confirmation results are digitally signed to ensure the immutability of the evidence and output the final rights confirmation results.

[0014] In a second aspect, the present invention also provides a forestry remote sensing data ontology feature recognition system for blockchain-based rights confirmation, applicable to the method described in the first aspect.

[0015] Unlike existing technologies, the above-mentioned technical solution extracts multi-scale features from forestry remote sensing image data using a pre-trained deep convolutional neural network to generate rotation- and scale-invariant feature pyramids. This is combined with seasonal attention weight adjustments to generate spatiotemporal fusion feature vectors, which are then dimensionality-reduced using principal component analysis to form forestry remote sensing image fingerprints. Fuzzy commitments are used to bind error-corrected coded random keys to these fingerprints, generating blockchain-based notarization commitments and constructing a storage architecture that coordinates on-chain notarization with off-chain metadata, thus achieving efficient ownership verification. This effectively solves the problem of traditional hashing methods being sensitive to image deformation and overcomes the challenge of feature drift caused by seasonal changes. While protecting data privacy, it achieves verifiable ownership verification, offering advantages such as high storage efficiency and fast response speed. It provides a trusted ownership authentication foundation for forestry remote sensing data sharing and significantly improves data circulation efficiency.

[0016] The above description of the invention is merely an overview of the technical solution of this application. In order to enable those skilled in the art to better understand the technical solution of this application and to implement it based on the description and drawings, and to make the above-mentioned objectives and other objectives, features and advantages of this application easier to understand, the following description is provided in conjunction with the specific embodiments and drawings of this application. Attached Figure Description

[0017] The accompanying drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of specific embodiments of the present invention and other related contents, and should not be considered as limitations on this application.

[0018] In the accompanying drawings of the instruction manual: Figure 1 This is a schematic diagram of steps S101 to S108 of the rights confirmation method described in the specific implementation. Detailed Implementation

[0019] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0020] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0021] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0022] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0023] In this application, 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 actual quantity, hierarchy or order relationship between these entities or operations.

[0024] Without further limitations, the use of terms such as “comprising,” “including,” “having,” or other similar open-ended expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0025] As understood in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments in this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0026] The processor described in the embodiments of this application can be implemented by hardware, firmware, software, or a combination thereof. It can be a circuit, one or more of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, or a microprocessor. It also includes other physical, biological, or chemical structures that can implement the same or equivalent functions as the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in the various embodiments of this application, or any combination of the steps mentioned therein.

[0027] The computer program involved in the embodiments can be stored in a computer device readable storage medium, which includes, but is not limited to, disks, magnetic tapes, magnetic cards, floppy disks, flash memory, optical disks, optical cards, read-only memory (ROM), random access memory (RAM), erasable programmable ROM (EPROM), and electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical, or chemical structures that can achieve the same or equivalent functions as the storage media listed above, such as DNA, RNA, proteins, and other units with information storage capabilities. In specific embodiments, the storage medium involved can be one of the above-mentioned media types, or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiments can be centrally stored in a single medium, or distributed and stored in multiple media. The memory containing the computer device readable storage medium can be non-volatile memory or random access memory. These computer device readable storage media can be built into the device, or can be connected to the device involved in the embodiments as an external device or part of an external device. In some embodiments, the memory having a computer device readable storage medium is deployed locally; in other embodiments, the memory may be deployed remotely from the processor, for example, as a network-attached memory accessed via RF circuitry or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), or a suitable combination thereof, as long as computer device access to the memory is enabled. Furthermore, the computer program involved in the embodiments may be stored in plaintext / ciphertext form, or it may be designed as training data, integrated and recombined through model training and implicitly stored in the parameter states of a deep neural network or other machine learning model.

[0028] Please see Figure 1 In a first aspect, this embodiment provides a method for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification, including: S101. Acquire forestry remote sensing image data, perform multi-scale feature extraction on the forestry remote sensing image data, and generate a feature pyramid with rotation invariance and scale invariance. The multi-scale feature extraction is configured to be implemented using a pre-trained deep convolutional neural network. S102. Perform hierarchical feature fusion on the feature pyramid. Calculate the Euclidean distance difference vector between feature maps at different levels and adjust the feature response by combining seasonal attention weights to generate a spatiotemporal fusion feature vector. S103. Dimensionality reduction is performed on the spatiotemporal fusion feature vector to obtain the forestry remote sensing image fingerprint. The dimension reduction process is configured to use the principal component analysis algorithm. S104. Construct a rights confirmation and evidence storage structure based on fuzzy commitment, generate a random key and perform error correction encoding on it, and XOR the error-corrected random key with the fingerprint of the forestry remote sensing image to generate a blockchain evidence storage commitment. S105. Write the blockchain notarization commitment and key hash value into the blockchain, and at the same time store the corresponding remote sensing image metadata off-chain. The remote sensing image metadata includes the acquisition time, geographical location and season identifier. S106. When a rights confirmation query request is received, the ontological features of the remote sensing image to be confirmed are extracted and a query fingerprint is generated. The blockchain block where similar fingerprints are located is located through off-chain index. S107. Perform fuzzy commitment verification in the located blockchain block, recover the error correction code through XOR operation and decode to obtain the original key, and compare the key hash value to complete the rights confirmation verification. S108. Output the ownership confirmation result, which includes ownership information and similarity score. The ownership information includes the original data holder, the derived relationship chain, and the access authorization record.

[0029] In step S101, forestry remote sensing image data is acquired and multi-scale feature extraction is performed. A pre-trained deep convolutional neural network is used to process the original image data, generating a feature pyramid with rotation and scale invariance. In some preferred embodiments, the deep convolutional neural network can employ classic network structures such as ResNet or VGG, capturing the feature representation of the image from local to global through multi-level convolutional operations, ensuring robustness to image rotation, scaling, and other operations.

[0030] In step S102, hierarchical feature fusion is performed on the feature pyramid. The similarity and difference between features at different scales are measured by calculating the Euclidean distance difference vector between feature maps at different levels. Seasonal attention weights are then used to adjust the feature responses, enhancing the feature representation of seasonally sensitive areas, ultimately generating a spatiotemporal fusion feature vector. Optionally, the seasonal attention weights can be dynamically adjusted according to the vegetation cover characteristics of different seasons to improve the accuracy of feature matching.

[0031] In step S103, the spatiotemporal fusion feature vector is subjected to dimensionality reduction processing. Principal component analysis is used to remove redundant information and retain the most discriminative feature components, ultimately obtaining the forestry remote sensing image fingerprint. This fingerprint has the characteristics of low dimensionality and high discriminative power, which facilitates subsequent storage and matching.

[0032] In step S104, a fuzzy commitment-based ownership verification and notarization structure is constructed, a random key is generated and error-correcting encoded to enhance its fault tolerance. The error-correcting encoded random key is XORed with the forestry remote sensing image fingerprint to generate a blockchain notarization commitment, ensuring that the privacy of the original data is protected while supporting subsequent verifiable ownership verification.

[0033] In step S105, the blockchain notarization commitment and key hash value are written into the blockchain to ensure data immutability. Simultaneously, the corresponding remote sensing image metadata, including acquisition time, geographical location, and season identifier, is stored off-chain. Optionally, off-chain storage can employ decentralized storage solutions such as distributed databases or IPFS to improve data access efficiency.

[0034] In step S106, when a rights confirmation query request is received, ontological features are extracted from the remote sensing image to be confirmed, generating a query fingerprint. The blockchain blocks containing similar fingerprints are quickly located using off-chain indexes, narrowing the search scope and improving rights confirmation efficiency.

[0035] In step S107, fuzzy commitment verification is performed in the located blockchain block. Error correction encoding is recovered through XOR operation and decoded to obtain the candidate original key. The hash value of the candidate key is calculated and compared with the key hash value stored in the blockchain to complete the rights confirmation verification. Optionally, if the initial verification fails, a suboptimal candidate can be selected from the set of candidate similar fingerprints to repeat the verification process.

[0036] In step S108, the output includes ownership information and a similarity score. Ownership information includes the original data holder, derived relationship chains, and access authorization records, providing complete ownership proof. The similarity score is a standardized value between 0 and 1, intuitively reflecting the degree of matching between the query image and the registered image.

[0037] Specifically, forestry remote sensing images inherently possess different texture features due to variations in capture time, season, and vegetation cover. These ontological features, much like human fingerprints, are highly distinctive. Ontological features are a data-driven method for describing the information and key properties of forestry remote sensing images. Feature extraction requires a focus on multi-scale keypoint detection to achieve scale invariance and rotation invariance. This addresses the issue of ontological features remaining unchanged after scaling, rotation, and cropping of forestry remote sensing data, achieving a "data fingerprint" effect and providing fundamental support for data ownership confirmation.

[0038] This embodiment utilizes a deep learning-based model to extract high-level semantic features of images and encodes these semantic features into vectors that can be matched for similarity.

[0039] Forestry remote sensing image data is input into a pre-trained ResNet network, and deep feature maps are extracted layer by layer to obtain a feature pyramid. The extracted feature pyramid can reflect the ontological features of the remote sensing image at different levels and resolutions.

[0040] Assuming the image library contains The first remote sensing image, extracted The feature pyramid of the image needs to be compared with the remaining images at different levels and resolutions. The feature pyramids of the images are compared pairwise.

[0041] After flattening the features at all scales, we denote them as eigenvectors. Calculate the first using Euclidean distance. One component and the remaining image vectors The difference values ​​at the same location, all of which constitute a new vector. , Represents the first Differences between Zhang's remote sensing image and other remote sensing images.

[0042] To enhance the distinction between the feature vectors of each remote sensing image and other feature vectors, it is possible to... Divide by the average vector of all remote sensing images To strengthen the first The proportion of each component eigenvalue.

[0043] Due to remote sensing images eigenvectors It is formed by flattening and stitching feature maps at different scales, resulting in a relatively high dimensionality. Principal Component Analysis (PAC) can be used to reduce the dimensionality to form the final remote sensing image fingerprint, denoted as... .

[0044] Existing blockchain-based rights confirmation methods typically store the hash value of the digital content to be confirmed on the blockchain and the original data in a database. However, for forestry remote sensing data, conventional hash value methods cannot achieve good originality detection results. This is because even a single bit of editing will change the hash value. Users usually focus on the content of the image, so this method cannot be applied to rights confirmation; rights confirmation detection based on ontological features is required. However, storing high-dimensional ontological features on the blockchain requires traversing all blocks for rights confirmation queries, resulting in a single query method, low efficiency, and lack of support for multi-dimensional feature data queries. This invention's blockchain rights confirmation component includes two parts: data on-chain storage and rights confirmation query.

[0045] When storing data on the blockchain, the principle of the commit process in fuzzy commitment is used to initialize an error correction capability. The error correction code is then used to generate random numbers. Use error correction codes to Encode to obtain codewords Remote sensing images Image fingerprint , and coding Bind and generate commitment ,in This represents the XOR operation. Finally, Storage on the blockchain, among which It is a hash function; you can choose a hash function such as SHA256.

[0046] To improve efficiency and avoid searching each block individually during ownership verification, a combination of off-chain and on-chain queries is used. The specific steps are as follows: Off-chain query. Forestry remote sensing images for which users need to confirm ownership. The data fingerprint is obtained by submitting it to a certain property rights confirmation agency and extracting its ontological features. The rights confirmation agency, using Euclidean distance in the comparison database, located fingerprints from similarly stored remote sensing images. Block .

[0047] On-chain query. Within the block. The principle of using the decommit phase of fuzzy commitment is employed. Recovering the code Then, regarding Perform error correction decoding to obtain Finally, hash verification is performed, comparing... Whether or not Equal. If If so, the property rights will be successfully confirmed.

[0048] The blockchain-based rights confirmation mechanism, which utilizes multi-dimensional data queries based on fuzzy commitments, leverages the fault tolerance of error-correcting codes in fuzzy commitments to address the issue of discrepancies in feature recognition. Furthermore, by combining on-chain and off-chain queries, it not only utilizes the immutability of the blockchain but also improves query efficiency.

[0049] The blockchain-based forestry remote sensing data ontology feature recognition method provided in this embodiment constructs a robust image fingerprint through multi-scale feature extraction, spatiotemporal fusion feature generation, and principal component analysis dimensionality reduction. It utilizes fuzzy commitment technology to achieve verifiable ownership verification under privacy protection; it employs on-chain evidence storage and off-chain metadata collaborative storage to balance security and efficiency; and it achieves efficient ownership verification query through a three-stage verification mechanism. This effectively solves the problem of traditional hash-based ownership verification being sensitive to image deformation and overcomes feature drift caused by seasonal changes, providing a reliable ownership authentication foundation for forestry remote sensing data sharing.

[0050] In some embodiments, multi-scale feature extraction is performed on forestry remote sensing image data to generate a feature pyramid with rotation invariance and scale invariance, including: Forestry remote sensing image data is scaled to a standard size of 512×512 pixels using bicubic interpolation to obtain the scaled image. Zero-filling is performed on the scaled image to maintain the aspect ratio of the forest features, resulting in the first expanded image. The pixel values ​​of the first expanded image are converted to the range of [-1,1] by normalization processing to obtain the second expanded image; The second augmented image is input into the pre-trained ResNet-50 network to extract feature maps from four layers. The spatial resolutions of the four layers are 256×256, 128×128, 64×64 and 32×32, respectively. A deformable convolutional network is applied to the feature map of each layer, and geometric deformation adaptation is performed by learning the offset field to obtain the deformation-adapted feature map; An orientation-sensitive convolutional kernel is applied to the deformation adaptation feature map. The orientation-sensitive convolutional kernel is configured as a Gabor filter bank with 12 directions. The extreme values ​​of the feature response in 12 directions at each spatial location are calculated, and the orientation angle corresponding to the maximum response is recorded. The feature map is rotated and aligned based on the orientation angle corresponding to the maximum response to generate an orientation-normalized feature map; Normalized feature maps from each level are input into the feature pyramid network and fused through a top-down path and 3×3 convolutions to output feature pyramids with four scales: 256×256, 128×128, 64×64, and 32×32.

[0051] In this embodiment, forestry remote sensing image data is scaled to a standard size of 512×512 pixels using bicubic interpolation to obtain the scaled image. This standardization process ensures that remote sensing data from different sources have a uniform input specification, facilitating subsequent feature extraction. Optionally, the bicubic interpolation algorithm can better preserve high-frequency detail information in the image, outperforming nearest neighbor or bilinear interpolation methods.

[0052] In this embodiment, zero-value padding is performed on the scaled image to maintain the aspect ratio of the forest features, resulting in the first expanded image. Zero-filling around the edges preserves the original geometric characteristics of the forest features, preventing distortion of important landform information due to size adjustments. In some preferred embodiments, the width of the zero-value padding can be dynamically calculated and determined based on the aspect ratio of the original image.

[0053] In this embodiment, normalization is used to convert the pixel values ​​of the first augmented image to the [-1, 1] interval to obtain the second augmented image. This makes remote sensing data under different lighting conditions comparable and improves the stability of feature extraction. The second augmented image is input into a pre-trained ResNet-50 network to extract feature maps at four levels. The spatial resolutions of the four levels are 256×256, 128×128, 64×64, and 32×32, respectively, forming the basis for multi-scale feature representation. The ResNet-50 network can be pre-trained on the ImageNet dataset and then fine-tuned using forestry remote sensing data.

[0054] In this embodiment, a deformable convolutional network is applied to the feature map of each level. Geometric deformation adaptation is achieved by learning the offset field, resulting in a deformation-adapted feature map. This automatically adapts to different forest morphological changes, enhancing the flexibility of feature representation. The offset learning for the deformable convolution is implemented using a 3×3 convolutional kernel. Orientation-sensitive convolutional kernels are applied to the deformation-adapted feature map, configured as a 12-directional Gabor filter bank. By calculating the extreme values ​​of the feature responses in the 12 directions at each spatial location and recording the orientation angle corresponding to the maximum response, directional feature quantization is achieved. In some preferred embodiments, the Gabor filter bank includes filters of different scales to capture multi-granularity directional features.

[0055] In this embodiment, the feature map is rotated and aligned according to the orientation angle corresponding to the maximum response to generate an orientation-normalized feature map. This eliminates orientation deviations caused by differences in shooting angles and enhances the rotation invariance of the features. The normalized feature maps of each level are input into a feature pyramid network and fused through a top-down path and 3×3 convolutions to output feature pyramids with four scales: 256×256, 128×128, 64×64, and 32×32. Semantic information from different levels is integrated to form a feature representation with multi-scale perception capabilities. The feature fusion process uses an element-wise addition method.

[0056] This embodiment constructs a feature pyramid with geometric deformation adaptability through standardized preprocessing, deformable convolutional networks, orientation-sensitive feature extraction, and multi-scale feature fusion. It effectively solves the feature inconsistency problem in forestry remote sensing data caused by factors such as shooting angle and seasonal changes, providing a stable and reliable feature representation foundation for subsequent land ownership confirmation. The feature extraction process takes into account both local details and global semantic information, significantly improving the discriminative ability of feature representation while ensuring rotation and scale invariance.

[0057] In some embodiments, hierarchical feature fusion is performed on the feature pyramid by calculating the Euclidean distance difference vector between feature maps at different levels, including: The feature maps of the four scales in the feature pyramid, 256×256, 128×128, 64×64 and 32×32, are input into a 1×1 convolutional layer to unify the channels, resulting in standardized feature maps with 256 channels each. Bilinear interpolation upsampling is performed on the standardized feature maps to adjust all feature maps to a uniform size of 256×256; Calculate the Euclidean distance difference vector between adjacent feature maps, including: Calculate the sum of squared differences in pixel values ​​at corresponding spatial locations between the 256×256 and 128×128 scale feature maps, and denote it as the first difference vector; Calculate the sum of squared differences in pixel values ​​at corresponding spatial locations between the 128×128 and 64×64 scale feature maps, and denote it as the second difference vector; Calculate the sum of squared differences in pixel values ​​at corresponding spatial locations between the 64×64 and 32×32 scale feature maps, and denote it as the third difference vector; The first difference vector, the second difference vector, and the third difference vector are input into a gated recurrent unit for time-series modeling to obtain multi-scale difference features. Based on multi-scale difference features, the feature response is adjusted by combining seasonal attention weights to generate a spatiotemporal fusion feature vector.

[0058] In this embodiment, feature maps of four scales—256×256, 128×128, 64×64, and 32×32—from the feature pyramid are input into a 1×1 convolutional layer for channel unification, resulting in standardized feature maps with 256 channels each. This ensures that features at different levels have the same channel dimension, providing a unified data format for subsequent feature fusion. In some preferred embodiments, the 1×1 convolutional layer employs the ReLU activation function to enhance nonlinear expressive power.

[0059] In this embodiment, bilinear interpolation upsampling is performed on the standardized feature maps to adjust all feature maps to a uniform size of 256×256. This size unification process achieves spatial alignment of features across scales, facilitating the calculation of feature differences between hierarchical levels. The sum of squared differences in pixel values ​​at corresponding spatial locations between the 256×256 and 128×128 scale feature maps is calculated and denoted as the first difference vector, quantifying the degree of feature change between the finest and second-finest scales. The sum of squared differences in pixel values ​​at corresponding spatial locations between the 128×128 and 64×64 scale feature maps is calculated and denoted as the second difference vector, reflecting the feature evolution patterns at medium scales and capturing the mesoscopic changes in vegetation cover. The sum of squared differences in pixel values ​​at corresponding spatial locations between the 64×64 and 32×32 scale feature maps is calculated and denoted as the third difference vector, characterizing feature differences from local to global perspectives and demonstrating sensitivity to changes in large areas of forest land.

[0060] In this embodiment, the first, second, and third difference vectors are input into a gated recurrent unit (GRU) for temporal modeling to obtain multi-scale difference features. A GRU network is used to learn the temporal dependencies between scale differences, establishing a cross-scale feature evolution model. Optionally, the GRU unit can employ a two-layer stacked structure to enhance modeling capabilities. Based on the multi-scale difference features, the feature response is adjusted by incorporating seasonal attention weights to generate a spatiotemporally fused feature vector. The seasonal attention mechanism dynamically strengthens the feature representation of areas with significant seasonal changes and suppresses non-seasonal interference factors.

[0061] This embodiment achieves hierarchical fusion of feature pyramids through techniques such as channel standardization, multi-scale difference calculation, and temporal modeling. It accurately quantifies feature changes at different scales using Euclidean distance difference vectors, captures inter-scale evolutionary patterns through gated recurrent units, and enhances spatiotemporal feature representation capabilities by incorporating a seasonal attention mechanism. The generated spatiotemporal fusion feature vector possesses both multi-scale perception capabilities and seasonal adaptability, providing a highly discriminative feature representation for forestry remote sensing data ownership confirmation and solving the problem of traditional single-scale feature fusion methods being insensitive to vegetation growth cycle changes.

[0062] In some embodiments, a spatiotemporal fusion feature vector is generated by adjusting the feature response based on multi-scale difference features and seasonal attention weights, including: Construct seasonal prototype vectors containing the features of the four seasons: spring, summer, autumn, and winter. Each seasonal prototype vector has a dimension of 256. Calculate the cosine similarity between multi-scale difference features and prototype vectors of each season to generate a seasonal correlation score matrix; The seasonal relevance score matrix is ​​input into the Softmax function for normalization to obtain the seasonal attention weights; Apply corresponding seasonal attention weights to the first, second, and third difference vectors respectively to perform feature importance weighting; The weighted first difference vector, second difference vector, and third difference vector are added to the standardized feature map through skip connections and fused to obtain the fused feature map. The fused feature maps are refined using 3×3 depthwise separable convolutions. Spatial information is aggregated through a global average pooling layer, outputting a 256-dimensional spatiotemporal fusion feature vector.

[0063] In this embodiment, seasonal prototype vectors containing the characteristics of spring, summer, autumn, and winter are constructed, with each seasonal prototype vector having a dimension of 256. These seasonal prototype vectors are obtained through cluster analysis of remote sensing features of typical seasons and serve as a baseline reference for seasonal features. The seasonal prototype vectors can be generated by clustering historical seasonal remote sensing data using the K-means algorithm.

[0064] The cosine similarity between multi-scale difference features and seasonal prototype vectors is calculated to generate a seasonal relevance score matrix. This quantifies the matching degree between the current feature and seasonal features, providing a basis for attention weight allocation. A temperature coefficient can be incorporated into the cosine similarity calculation to adjust the score distribution. The seasonal relevance score matrix is ​​input into a Softmax function for normalization to obtain seasonal attention weights. The Softmax function transforms the similarity scores into a probability distribution, ensuring that the sum of the weights is 1. In this embodiment, corresponding seasonal attention weights are applied to the first, second, and third difference vectors, respectively, to perform feature importance weighting, giving higher weights to seasonally sensitive feature differences and suppressing feature changes unrelated to the season.

[0065] The weighted first, second, and third difference vectors are fused with the standardized feature map via skip connections to obtain the fused feature map. Skip connections preserve the original feature information and avoid the gradient vanishing problem. A 3×3 depthwise separable convolution is applied to the fused feature map for feature refinement. This operation enhances local feature interactions and improves feature quality while reducing the number of parameters. Spatial information is aggregated through a global average pooling layer, outputting a 256-dimensional spatiotemporal fusion feature vector. Pooling compresses the two-dimensional feature map into a one-dimensional vector, preserving the channel-dimensional feature representation. Optionally, a fully connected layer can be added after pooling for further dimensionality reduction.

[0066] This embodiment achieves effective fusion of multi-scale features and seasonal information through operations such as seasonal prototype matching, attention weighting, and feature refinement. The seasonal attention mechanism dynamically adjusts the feature response, enabling the model to adapt to changes in vegetation characteristics across different seasons. The generated 256-dimensional spatiotemporal fusion feature vector possesses both seasonal discriminative power and spatial discriminative ability, providing a robust feature representation for forestry remote sensing data rights determination. Through designs such as depthwise separable convolution and skip connections, computational complexity is controlled while ensuring feature quality, making it suitable for large-scale remote sensing data processing.

[0067] In some embodiments, the spatiotemporal fusion feature vector is dimensionality reduced to obtain a forestry remote sensing image fingerprint, including: The spatiotemporal fusion feature vector is input into the principal component analysis network, which includes a feature standardization layer and a covariance calculation layer. The spatiotemporal fusion feature vector is normalized to zero mean by a feature standardization layer to obtain a standardized feature vector. Calculate the covariance matrix of the standardized eigenvectors, and perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and the initial eigenvectors corresponding to the eigenvalues; The first 128 initial eigenvectors are selected as principal component projection matrices in descending order of eigenvalues; A linear transformation is performed on the standardized eigenvectors and the principal component projection matrix to obtain a 128-dimensional intermediate eigenvector. The intermediate feature vector is binarized by converting the feature values ​​into 0 / 1 bits using a sign function; The 128 bits are concatenated in sequence to generate a 128-bit forestry remote sensing image fingerprint; Hamming distance verification is performed on forestry remote sensing image fingerprints to ensure the distinguishability between them.

[0068] In this embodiment, the spatiotemporal fusion feature vector is input into a principal component analysis (PCA) network, which includes a feature normalization layer and a covariance calculation layer. The most discriminative feature components are extracted through linear transformation. The spatiotemporal fusion feature vector is normalized to zero mean by the feature normalization layer to obtain a standardized feature vector, eliminating differences in feature dimensions and ensuring the accuracy of subsequent covariance calculation. The covariance matrix of the standardized feature vector is calculated, and eigenvalue decomposition is performed on the covariance matrix to obtain eigenvalues ​​and their corresponding initial eigenvectors. This reveals the intrinsic variation pattern of the feature vectors and determines the main feature directions. Optionally, the eigenvalue decomposition can be implemented using the Jacobi iterative algorithm.

[0069] In this embodiment, the first 128 initial eigenvectors are selected as the principal component projection matrix in descending order of eigenvalues. The principal component projection matrix retains the 128 orthogonal directions with the largest variance in the original eigenvalues. A linear transformation is performed on the standardized eigenvectors and the principal component projection matrix to obtain 128-dimensional intermediate eigenvectors, achieving effective compression of the feature space and removing redundant information.

[0070] In this embodiment, the intermediate feature vector is binarized by converting the feature values ​​into 0 / 1 bits using a sign function, enhancing the robustness of the features and facilitating fingerprint comparison and storage. The 128 bits are sequentially concatenated to generate a 128-bit forestry remote sensing image fingerprint. This compact representation significantly reduces storage space requirements and improves retrieval efficiency. Hamming distance verification is performed on the forestry remote sensing image fingerprints to ensure their distinguishability. This verification process validates the uniqueness of the fingerprints and avoids duplicate collisions. In some preferred embodiments, a fingerprint database can be established to achieve rapid comparison.

[0071] This embodiment utilizes techniques such as principal component analysis, feature binarization, and Hamming verification to transform high-dimensional spatiotemporal features into a compact 128-bit image fingerprint. While preserving feature discriminative power, it achieves efficient storage and rapid comparison, resulting in a forestry remote sensing image fingerprint with robust properties against noise and transformations. Eigenvalue decomposition selects the 128 most representative principal components to ensure the fingerprint contains maximum information. Binarization further enhances the fingerprint's practicality and operability.

[0072] In some embodiments, a fuzzy commitment-based ownership verification and storage structure is constructed, a random key is generated and error-correcting encoding is performed on it, including: A cryptographically secure pseudo-random number generator is used to generate a 256-bit raw random key; The original random key is input into the error correction encoder for encoding to obtain a 255-bit error correction encoded key. Calculate the SHA-256 hash value of the original random key as the key verification fingerprint; Hamming weight verification is performed on the error correction coding key to ensure that it meets the preset uniform distribution condition; When the Hamming weight does not meet the conditions, a random key is regenerated and the above encoding process is repeated until a verified error-corrected encoded key is generated, which is the error-corrected random key. Additionally, a mapping relationship is established between the verified error correction coding key and the key verification fingerprint, and the fingerprint is stored in the key management pool.

[0073] In this embodiment, a cryptographically secure pseudo-random number generator is used to generate a 256-bit original random key, which serves as the basic encryption material for rights confirmation and evidence storage, exhibiting unpredictability and high entropy. In some preferred embodiments, the pseudo-random number generator can be implemented in AES-CTR mode. The original random key is input into an error correction encoder for encoding, resulting in a 255-bit error correction encoded key. This encoding process enhances the key's fault tolerance by adding redundant check bits. The SHA-256 hash value of the original random key is calculated as the key verification fingerprint, serving as the key's unique identifier for subsequent verification stages.

[0074] In this embodiment, the error-correcting coding key undergoes Hamming weight verification to ensure it meets a preset uniform distribution condition. This verification prevents key deviation and guarantees statistical randomness. If the Hamming weight does not meet the condition, a random key is regenerated and the above encoding process is repeated until a verified error-correcting coding key is generated, which is the error-corrected random key, ensuring that the final key meets cryptographic strength requirements. A mapping relationship is established between the verified error-correcting coding key and the key verification fingerprint, and the fingerprint is stored in the key management pool for easy key retrieval and management. Optionally, the mapping relationship can be organized using a Merkle tree structure.

[0075] This embodiment constructs a secure and reliable key management system through pseudo-random generation, error-correcting coding, and multiple verifications. The error-correcting coded key has the ability to resist noise interference and achieves two-way verification in conjunction with key verification fingerprints. Hamming weight verification ensures key quality, and the key management pool enables ordered storage, providing a solid cryptographic foundation for forestry remote sensing data ownership confirmation and making it suitable for data evidence storage scenarios in a distributed environment.

[0076] In some embodiments, the error-corrected coded random key is XORed with the forestry remote sensing image fingerprint to generate a blockchain-based notarization commitment, including: The fingerprint of the forestry remote sensing image is input into the fingerprint expansion module, and the image expansion fingerprint is generated through a cyclic shift operation. The error correction coding key is XORed with the image-extended fingerprint input and bound to the module. Perform an XOR operation bit by bit to generate the initial binding result; The initial binding result is verified using Hamming code to detect and correct possible binding errors, resulting in a verified binding result. The integrity of the verified binding result is checked using hash calculation, and a binding verification code is generated. The binding verification code is combined with the initial binding result to generate a blockchain-based evidence commitment.

[0077] In this embodiment, the fingerprint of the forestry remote sensing image is input into the fingerprint expansion module. An extended image fingerprint is generated through a cyclic shift operation. Bit operations are used to increase the randomness and complexity of the fingerprint, improving the security of subsequent binding operations. Optionally, a variable step size strategy can be used for the cyclic shift to enhance the expansion effect. The error correction coding key is XORed with the extended image fingerprint input into the binding module as the core binding unit, achieving a secure association between cryptographic materials and biometrics. Optionally, a temporal noise protection mechanism can be added to the binding module.

[0078] In this embodiment, an XOR logical operation is performed bit-by-bit to generate the initial binding result, ensuring the irreversibility and uniform distribution of the binding process. Hamming code verification is applied to the initial binding result to detect and correct potential binding errors, resulting in a verified binding result that guarantees the reliability of the bound data and prevents bit errors during transmission and storage. Integrity verification is performed on the verified binding result using a hash operation to generate a binding verification code, which serves as a digital fingerprint of data integrity for subsequent verification. Optionally, the hash operation can be implemented using the SHA-3 algorithm. In this embodiment, the binding verification code is concatenated with the initial binding result to generate a blockchain-based notarized commitment, containing the original binding information and a verification marker, facilitating blockchain storage and verification.

[0079] This embodiment achieves secure association between forestry remote sensing data and cryptographic keys through fingerprint extension, XOR binding, and multiple verifications. The generated blockchain-based notarized commitment is immutable and verifiable, and data integrity is ensured through Hamming checksum and hash verification. Fingerprint extension enhances binding security, and XOR operation guarantees reversibility, balancing security and storage efficiency, making it suitable for data ownership confirmation applications in a blockchain environment.

[0080] In some embodiments, when a rights confirmation query request is received, the ontological features of the remote sensing image to be confirmed are extracted and a query fingerprint is generated. The blockchain block containing similar fingerprints is located using an off-chain index, including: Multi-scale feature extraction is performed on the remote sensing images for property rights confirmation to generate a query feature pyramid; The query feature pyramid is fused using a seasonal attention weighting algorithm to obtain a query spatiotemporal fusion feature vector. The spatiotemporal fusion feature vector of the query is reduced in dimensionality to generate a query fingerprint; An index structure based on an improved VP-Tree is built in the off-chain fingerprint database. The improved VP-Tree is configured to support Hamming distance metric. Centered on the query fingerprint, a radius expansion search is performed in the improved VP-Tree to obtain a set of candidate similar fingerprints; Sort the candidate similar fingerprint set by Hamming distance, and select the top few candidate similar fingerprints, which are denoted as identical fingerprints; Locate the blockchain block address corresponding to similar fingerprints using the blockchain transaction hash mapping table; Verify the timestamps and version identifiers of candidate blocks to ensure the timeliness and validity of block data; The output includes the matching degree between the identical fingerprint and the currently queried fingerprint, as well as the location result of the corresponding block information.

[0081] In this embodiment, upon receiving a rights confirmation query request, the feature extraction process of the remote sensing image to be confirmed is initiated as the initial step in the rights confirmation verification, ensuring that the input data for subsequent operations has sufficient representational power. Multi-scale feature extraction is performed on the remote sensing image to be confirmed, generating a query feature pyramid. The feature pyramid structure contains image features at different resolutions, forming a hierarchical feature representation system. Multi-scale feature extraction can be implemented using the Gaussian pyramid decomposition method. Furthermore, the query feature pyramid is fused using a seasonal attention weighting algorithm to obtain a query spatiotemporal fusion feature vector. This embodiment can dynamically adjust the weight coefficients of different seasonal features to enhance the discriminative power of spatiotemporal features. The query spatiotemporal fusion feature vector is then subjected to dimensionality reduction processing to generate a query fingerprint. The query fingerprint, as a compact representation of the image, facilitates subsequent similarity comparison operations.

[0082] In this embodiment, an index structure based on an improved VP-Tree is constructed in the off-chain fingerprint database. The improved VP-Tree is configured to support Hamming distance metric. This index structure optimizes the node partitioning strategy of the traditional VP-Tree, improving the retrieval efficiency of binary fingerprints. For example, the improved VP-Tree can adopt a dynamic vantage point selection strategy. Centered on the query fingerprint, a radius expansion search is performed in the improved VP-Tree to obtain a set of candidate similar fingerprints. By gradually expanding the search range, the retrieval accuracy and computational cost are balanced. The candidate similar fingerprint set is sorted by Hamming distance, and the top few candidate similar fingerprints are selected and marked as identical fingerprints, ensuring that the returned candidate fingerprints have the highest matching probability.

[0083] In this embodiment, the blockchain block address corresponding to a similar fingerprint is located through a blockchain transaction hash mapping table to maintain the correspondence between fingerprint hashes and blockchain transactions, achieving fast address resolution. Optionally, a Bloom filter can be used in the blockchain transaction hash mapping table to accelerate queries. The timestamp and version identifier of the candidate block are verified to ensure the timeliness and validity of the block data and prevent the use of expired or invalid block data for rights confirmation. The output includes the matching degree between the similar fingerprint and the currently queried fingerprint, as well as the location result of the corresponding block information, providing a complete chain of evidence required for rights confirmation. Optionally, the output result can be in a standardized JSON format.

[0084] This embodiment achieves efficient rights confirmation queries for remote sensing images through multi-scale feature extraction, seasonal attention-weighted fusion, and improved VP-Tree retrieval. The generated query fingerprint has strong representational capabilities, and combined with off-chain indexing and blockchain verification mechanisms, it ensures the accuracy and reliability of the rights confirmation results. The feature pyramid enhances multi-scale analysis capabilities, seasonal attention optimizes spatiotemporal feature representation, and the improved VP-Tree improves retrieval efficiency, balancing query accuracy and system performance, making it suitable for rights confirmation management scenarios involving large-scale remote sensing data.

[0085] In some embodiments, fuzzy commitment verification is performed in the located blockchain block, the error-correcting code is recovered through XOR operation and decoded to obtain the original key, and the ownership verification is completed by comparing the key hash value, including: Extract blockchain-based evidence commitments from blockchain blocks; The blockchain-based notarization commitment is broken down into binding results and binding verification codes; The query fingerprint is expanded to generate an extended query fingerprint. The extended query fingerprint and the binding result are XORed to obtain the recovered error correction code; The recovered error-correcting code is decoded to obtain the candidate original key; Calculate the hash value of the candidate original key and compare it with the key hash value stored in the blockchain block to obtain the verification result; When the verification result shows a hash value match, the verification is considered successful and the ownership information retrieval process is triggered. When the verification result is that the hash values ​​do not match, the second-best candidate is selected from the set of candidate similar fingerprints and the verification process is repeated. The output includes ownership information and similarity scores, confirming ownership results. Ownership information includes the original data holder, derived relationship chains, and access authorization records, including: Retrieve the corresponding metadata index based on the successfully verified key hash value; Obtain remote sensing image metadata from off-chain storage, including acquisition time, geographic location, and season identifier; Construct a chain of ownership proofs that includes the identity identifier of the original data holder; Additionally, by querying the derivative relationship chain and access authorization records of forestry remote sensing image data through smart contracts, ownership information can be generated; Calculate the similarity score between the query fingerprint and the registered fingerprint, and generate a standardized similarity value between 0 and 1; The ownership information, similarity score, and verification timestamp are combined to generate structured ownership confirmation results; The structured rights confirmation results are digitally signed to ensure the immutability of the evidence and output the final rights confirmation results.

[0086] In this embodiment, fuzzy commitment verification is performed in the located blockchain block. The error-correcting code is recovered through XOR operation and decoded to obtain the original key. The key hash value is compared to complete the ownership verification. This ensures the verifiability of data ownership while protecting the confidentiality of the original key. The blockchain evidence commitment is extracted from the blockchain block. The blockchain evidence commitment includes the binding result and binding verification code, serving as the basic data for ownership verification. The blockchain evidence commitment is decomposed into the binding result and binding verification code, recovering the data components generated in the initial binding stage, providing input for subsequent XOR operations. The query fingerprint is extended to generate an extended query fingerprint, and the matching compatibility of the fingerprint is enhanced through cyclic shifting or hash transformation. The extended query fingerprint is XORed with the binding result to obtain the recovered error-correcting code. This operation reverses the unbinding process, restoring the initial error-correcting code data.

[0087] In this embodiment, the recovered error-correcting code is decoded to obtain candidate original keys. This decoding process utilizes error-correcting algorithms to repair potential transmission or storage errors. The hash value of the candidate original key is calculated and compared with the key hash value stored in the blockchain block to obtain a verification result. This comparison ensures that the recovered key is consistent with the registered key. Optionally, a timed delay strategy can be used for hash comparison to prevent side-channel attacks. When the verification result shows a hash value match, the verification is considered successful, and the ownership information retrieval process is triggered, automatically associating the key with ownership metadata. When the verification result shows a hash value mismatch, a second-best candidate is selected from the set of candidate similar fingerprints, and the verification process is repeated to improve the fault tolerance of the ownership confirmation system.

[0088] In this embodiment, the output includes ownership information and a similarity score. Ownership information includes the original data holder, the derived relationship chain, and access authorization records. The ownership confirmation result provides a complete chain of ownership evidence. The corresponding metadata index is retrieved based on the successfully verified key hash value, establishing a link between the key and off-chain metadata. Remote sensing image metadata, including acquisition time, geographical location, and seasonal identifier, is obtained from off-chain storage to enhance the credibility of the ownership information. An ownership proof chain containing the original data holder's identity is constructed, providing a traceable ownership history. The derived relationship chain and access authorization records of forestry remote sensing image data are queried via smart contracts to generate ownership information, ensuring the immutability of authorization records. A similarity score is calculated between the query fingerprint and the registered fingerprint, generating a standardized similarity value between 0 and 1 to quantify the degree of matching of image content.

[0089] In this embodiment, ownership information, similarity scores, and verification timestamps are combined to generate a structured ownership confirmation result, standardizing the output format of the ownership confirmation evidence. The structured ownership confirmation result is digitally signed to ensure the immutability of the evidence, and the final ownership confirmation result is output using a digital certificate from a verification institution.

[0090] This embodiment implements a complete ownership verification process for forestry remote sensing images through fuzzy commitment verification, key hash comparison, and smart contract querying. While protecting the security of the original key, key information is recovered through XOR operations and error correction decoding, and the credibility of the verification process is ensured by combining blockchain evidence storage. The ownership proof chain provides a traceable ownership history, smart contracts enable automated authorization management, and standardized similarity scoring quantifies the degree of image matching, balancing verification accuracy and privacy protection, making it suitable for data ownership authentication needs in a distributed environment.

[0091] In a second aspect, this embodiment also provides a forestry remote sensing data ontology feature recognition system with blockchain-based ownership confirmation, which is applicable to the method described in the first aspect.

[0092] Unlike existing technologies, the above-mentioned technical solution extracts multi-scale features from forestry remote sensing image data using a pre-trained deep convolutional neural network to generate rotation- and scale-invariant feature pyramids. This is combined with seasonal attention weight adjustments to generate spatiotemporal fusion feature vectors, which are then dimensionality-reduced using principal component analysis to form forestry remote sensing image fingerprints. Fuzzy commitments are used to bind error-corrected coded random keys to these fingerprints, generating blockchain-based notarization commitments and constructing a storage architecture that coordinates on-chain notarization with off-chain metadata, thus achieving efficient ownership verification. This effectively solves the problem of traditional hashing methods being sensitive to image deformation and overcomes the challenge of feature drift caused by seasonal changes. While protecting data privacy, it achieves verifiable ownership verification, offering advantages such as high storage efficiency and fast response speed. It provides a trusted ownership authentication foundation for forestry remote sensing data sharing and significantly improves data circulation efficiency.

[0093] Finally, it should be noted that although the above embodiments have been described in the text and drawings of this application, this should not limit the scope of patent protection of this application. Any technical solutions that are based on the essential concept of this application and utilize the content described in the text and drawings of this application, resulting in equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this application.

Claims

1. A method for identifying the ontological features of forestry remote sensing data using blockchain-based rights confirmation, characterized in that, include: Forestry remote sensing image data is acquired, and multi-scale feature extraction is performed on the forestry remote sensing image data to generate a feature pyramid with rotation invariance and scale invariance. The multi-scale feature extraction is configured to be implemented using a pre-trained deep convolutional neural network. The feature pyramid is subjected to hierarchical feature fusion. By calculating the Euclidean distance difference vector of the feature maps at different levels and adjusting the feature response in combination with seasonal attention weights, a spatiotemporal fusion feature vector is generated. The spatiotemporal fusion feature vector is subjected to dimensionality reduction processing to obtain a forestry remote sensing image fingerprint. The dimensionality reduction processing is configured to use a principal component analysis algorithm. A fuzzy commitment-based rights confirmation and evidence storage structure is constructed, a random key is generated and error-corrected encoding is performed on it, and the error-corrected random key is XORed with the fingerprint of the forestry remote sensing image to generate a blockchain evidence storage commitment. The blockchain notarization commitment and key hash value are written into the blockchain, while the corresponding remote sensing image metadata is stored off-chain. The remote sensing image metadata includes the acquisition time, geographical location and season identifier. When a rights confirmation query request is received, the ontological features of the remote sensing image to be confirmed are extracted and a query fingerprint is generated. The blockchain block where similar fingerprints are located is located through off-chain index. Perform fuzzy commitment verification in the located blockchain block, recover the error correction code through XOR operation and decode to obtain the original key, and complete the confirmation of rights verification by comparing the key hash value; The output includes ownership information and similarity scores, and the ownership information includes the original data holder, the derived relationship chain, and access authorization records.

2. The method for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification according to claim 1, characterized in that, Multi-scale feature extraction is performed on the forestry remote sensing image data to generate a feature pyramid with rotation invariance and scale invariance, including: Forestry remote sensing image data is scaled to a standard size of 512×512 pixels using bicubic interpolation to obtain the scaled image. Zero-filling is performed on the scaled image to maintain the aspect ratio of the forest features, resulting in the first expanded image. The pixel values ​​of the first expanded image are converted to the range of [-1, 1] by normalization processing to obtain the second expanded image; The second augmented image is input into the pre-trained ResNet-50 network to extract feature maps from four layers. The spatial resolutions of the four layers are 256×256, 128×128, 64×64 and 32×32, respectively. A deformable convolutional network is applied to the feature map of each layer, and geometric deformation adaptation is performed by learning the offset field to obtain the deformation-adapted feature map; An orientation-sensitive convolutional kernel is applied to the deformation adaptation feature map. The orientation-sensitive convolutional kernel is configured as a Gabor filter bank with 12 directions. The extreme values ​​of the feature responses in 12 directions at each spatial location are calculated, and the orientation angle corresponding to the maximum response is recorded. The feature map is rotated and aligned according to the orientation angle corresponding to the maximum response to generate an orientation-normalized feature map; Normalized feature maps from each level are input into the feature pyramid network and fused through a top-down path and 3×3 convolutions to output feature pyramids with four scales: 256×256, 128×128, 64×64, and 32×32.

3. The method for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification according to claim 1, characterized in that, The feature pyramid is subjected to hierarchical feature fusion by calculating the Euclidean distance difference vector between feature maps at different levels, including: The feature maps of the four scales (256×256, 128×128, 64×64, and 32×32) in the feature pyramid are respectively input into a 1×1 convolutional layer for channel unification, resulting in standardized feature maps with 256 channels each. Bilinear interpolation upsampling is performed on the standardized feature maps to adjust all feature maps to a uniform size of 256×256; Calculate the Euclidean distance difference vector between adjacent feature maps, including: Calculate the sum of squared differences in pixel values ​​at corresponding spatial locations between the 256×256 and 128×128 scale feature maps, and denote it as the first difference vector; Calculate the sum of squared differences in pixel values ​​at corresponding spatial locations between the 128×128 and 64×64 scale feature maps, and denote it as the second difference vector; Calculate the sum of squared differences in pixel values ​​at corresponding spatial locations between the 64×64 and 32×32 scale feature maps, and denote it as the third difference vector; The first difference vector, the second difference vector, and the third difference vector are input into a gated recurrent unit for time-series modeling to obtain multi-scale difference features. Based on the multi-scale difference features, and combined with seasonal attention weights to adjust the feature response, a spatiotemporal fusion feature vector is generated.

4. The method for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification according to claim 3, characterized in that, Based on the multi-scale difference features, and by adjusting the feature response using seasonal attention weights, a spatiotemporal fusion feature vector is generated, including: Construct seasonal prototype vectors containing the features of the four seasons: spring, summer, autumn, and winter. Each seasonal prototype vector has a dimension of 256. Calculate the cosine similarity between multi-scale difference features and prototype vectors of each season to generate a seasonal correlation score matrix; The seasonal relevance score matrix is ​​input into the Softmax function for normalization to obtain the seasonal attention weights; Apply corresponding seasonal attention weights to the first, second, and third difference vectors respectively to perform feature importance weighting; The weighted first difference vector, second difference vector, and third difference vector are added to the standardized feature map through skip connections and fused to obtain the fused feature map. The fused feature maps are refined using 3×3 depthwise separable convolutions. Spatial information is aggregated through a global average pooling layer, outputting a 256-dimensional spatiotemporal fusion feature vector.

5. The method for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification according to claim 1, characterized in that, The spatiotemporal fusion feature vector is subjected to dimensionality reduction processing to obtain a forestry remote sensing image fingerprint, including: The spatiotemporal fusion feature vector is input into a principal component analysis network, which includes a feature standardization layer and a covariance calculation layer. The spatiotemporal fusion feature vector is normalized to zero mean by a feature standardization layer to obtain a standardized feature vector. Calculate the covariance matrix of the standardized eigenvectors, and perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and the initial eigenvectors corresponding to the eigenvalues; The first 128 initial eigenvectors are selected as principal component projection matrices in descending order of eigenvalues; A linear transformation is performed on the standardized eigenvectors and the principal component projection matrix to obtain a 128-dimensional intermediate eigenvector. The intermediate feature vector is binarized by converting the feature values ​​to 0 / 1 bits using a sign function; The 128 bits are concatenated in sequence to generate a 128-bit forestry remote sensing image fingerprint; Hamming distance verification is performed on forestry remote sensing image fingerprints to ensure the distinguishability between them.

6. The method for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification according to claim 1, characterized in that, Constructing a rights confirmation and evidence storage structure based on fuzzy commitments, generating a random key and performing error-correcting encoding on it, including: A cryptographically secure pseudo-random number generator is used to generate a 256-bit raw random key; The original random key is input into the error correction encoder for encoding to obtain a 255-bit error correction encoded key; Calculate the SHA-256 hash value of the original random key as the key verification fingerprint; Hamming weight verification is performed on the error correction coding key to ensure that it meets the preset uniform distribution condition; When the Hamming weight does not meet the conditions, a random key is regenerated and the above encoding process is repeated until a verified error-corrected encoded key is generated, which is the error-corrected random key. Additionally, a mapping relationship is established between the verified error correction coding key and the key verification fingerprint, and the fingerprint is stored in the key management pool.

7. The method for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification according to claim 6, characterized in that, The error-corrected, encoded random key is XORed with the forestry remote sensing image fingerprint to generate a blockchain-based notarization commitment, including: The fingerprint of the forestry remote sensing image is input into the fingerprint expansion module, and the image expansion fingerprint is generated through a cyclic shift operation. The error correction coding key is XOR-bound to the image-extended fingerprint input module; Perform an XOR operation bit by bit to generate the initial binding result; The initial binding result is validated using Hamming codes to detect and correct binding errors, resulting in a validated binding result. The integrity of the verified binding result is checked using hash calculation, and a binding verification code is generated. The binding verification code is combined with the initial binding result to generate a blockchain-based evidence commitment.

8. The method for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification according to claim 1, characterized in that, Upon receiving a rights confirmation query request, the ontological features of the remote sensing image to be confirmed are extracted and a query fingerprint is generated. The blockchain block containing similar fingerprints is located using an off-chain index, including: Multi-scale feature extraction is performed on the remote sensing images for property rights confirmation to generate a query feature pyramid; The query feature pyramid is fused using a seasonal attention weighting algorithm to obtain a query spatiotemporal fusion feature vector. The spatiotemporal fusion feature vector of the query is reduced in dimensionality to generate a query fingerprint; An index structure based on an improved VP-Tree is constructed in the off-chain fingerprint database, the improved VP-Tree being configured to support Hamming distance metric; Centered on the query fingerprint, a radius expansion search is performed in the improved VP-Tree to obtain a set of candidate similar fingerprints; Sort the candidate similar fingerprint set by Hamming distance, and select the top few candidate similar fingerprints, which are denoted as identical fingerprints; Locate the blockchain block address corresponding to similar fingerprints using the blockchain transaction hash mapping table; Verify the timestamps and version identifiers of candidate blocks to ensure the timeliness and validity of block data; The output includes the matching degree between the identical fingerprint and the currently queried fingerprint, as well as the location result of the corresponding block information.

9. The method for identifying the ontological features of forestry remote sensing data with blockchain-based ownership verification according to claim 1, characterized in that, Perform fuzzy commitment verification within the located blockchain block, recover the error-correcting code through XOR operation and decode to obtain the original key, and complete the rights confirmation verification by comparing the key hash value, including: Extract the blockchain evidence commitment from the blockchain block; The blockchain-based notarization commitment is broken down into a binding result and a binding verification code; The query fingerprint is expanded to generate an extended query fingerprint; The extended query fingerprint and the binding result are XORed to obtain the recovered error correction code; The recovered error correction code is decoded to obtain the candidate original key; Calculate the hash value of the candidate original key and compare it with the key hash value stored in the blockchain block to obtain the verification result; When the verification result shows a hash value match, the verification is considered successful and the ownership information retrieval process is triggered. When the verification result is that the hash values ​​do not match, the second-best candidate is selected from the set of candidate similar fingerprints and the verification process is repeated. The output includes ownership information and a similarity score, confirming the data's ownership. The ownership information includes the original data holder, derived relationship chains, and access authorization records, including: Retrieve the corresponding metadata index based on the successfully verified key hash value; Obtain remote sensing image metadata from off-chain storage, including acquisition time, geographic location, and season identifier; Construct a chain of ownership proofs that includes the identity identifier of the original data holder; Additionally, by querying the derived relationship chain and access authorization records of the forestry remote sensing image data through smart contracts, ownership information is generated; Calculate the similarity score between the query fingerprint and the registered fingerprint, and generate a standardized similarity value between 0 and 1; The ownership information, similarity score, and verification timestamp are combined to generate a structured ownership confirmation result; The structured ownership confirmation result is digitally signed to ensure the immutability of the evidence, and the final ownership confirmation result is output.

10. A forestry remote sensing data ontology feature recognition system for blockchain-based rights confirmation, characterized in that, The method applicable to any one of claims 1 to 9.

Images