A blockchain-based aquaculture aquatic organism asset value evaluation method

Abstract

The application discloses a kind of based on aquaculture aquatic organism asset value evaluation method of block chain, belong to the technical field of block chain asset evaluation.This method includes: in response to the request of upper chain of aquaculture pond monitoring data, monitoring data is stored in block chain network and triggers smart contract to execute initial value anchoring operation;According to the result of initial value anchoring operation, call the multi-source monitoring data stored in chain, the consistency check of the growth stage of aquaculture aquatic organism is carried out, and the growth state data after check is obtained;Based on the growth state data after check and in combination with on-chain historical transaction records, the dynamic value adjustment factor of current aquaculture aquatic organism asset is determined;The result of initial value anchoring operation is modified using dynamic value adjustment factor and is handled, and the modified asset value evaluation result is written back to block chain network.The method improves the credibility of value evaluation and market fitting degree through growth stage consistency check and dynamic adjustment based on market transaction.

Description

Technical Field

[0001] This invention relates to the field of blockchain asset valuation technology, specifically a blockchain-based method for valuing aquatic biological assets. Background Technology

[0002] Valuation of aquatic biological assets is fundamental for asset mortgages, transactions, and insurance claims in the aquaculture industry. Current valuation methods typically rely on manual on-site sampling, where appraisers randomly collect samples to measure parameters such as body length and weight, then estimate the overall biomass value based on experience or static growth curve models. While some digital methods have introduced water quality sensors and cameras for environmental monitoring, this data is mostly stored in centralized databases, posing a risk of unilateral data tampering. The valuation process still relies on data from a single sampling point, failing to reflect the dynamic growth status throughout the aquaculture cycle. Furthermore, market prices fluctuate continuously, and current valuation methods only refer to historical average selling prices or fixed price lists to determine a benchmark unit price. This results in a significant lag and deviation between the valuation results and real-time market conditions, leading to either overvaluation or undervaluation of assets, impacting the reliability of decisions made by financial institutions and transaction participants. One approach is to store monitoring data on a blockchain to ensure immutability, but simply storing the data does not solve the credibility problem of determining the growth stage. Water quality parameters and biomass images collected by sensors may produce outliers due to equipment malfunctions or environmental interference. Directly using these for growth stage identification can lead to distorted anchor values. Furthermore, if smart contracts rely solely on a single data source for initial value calculations, the lack of cross-validation with multi-dimensional data makes it difficult to avoid value anchoring biases caused by input errors. On the other hand, current technology lacks a mechanism to dynamically extract market reference value from on-chain historical transaction records and correct assessment results in real time. Asset values ​​cannot be adjusted to reflect fluctuations in actual transaction prices, and assessment results remain at a static anchoring level.

[0003] One pressing issue in the valuation of aquatic biological assets is how to automatically cross-validate growth stages using multi-source sensor data after data is stored on the blockchain, thereby eliminating misjudgments of growth stages caused by anomalies in single-source data. Another issue is how to incorporate on-chain historical transaction information onto immutable data, transforming real market transaction dynamics into quantifiable value adjustment factors, thus enabling the assessed value to respond responsively to market conditions. Summary of the Invention

[0004] This invention provides a blockchain-based method for assessing the value of aquatic biological assets. By triggering a smart contract after monitoring data is uploaded to the blockchain, the method calls on-chain multi-source monitoring data to verify the consistency of the growth stage, and calculates a dynamic value adjustment factor by combining on-chain historical transaction records to correct the initial value and write it back to the blockchain. This solves the problems of insufficient credibility in determining the growth stage and the deviation of the assessed value from market conditions in the existing technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a blockchain-based method for evaluating the value of aquatic biological assets, which combines blockchain notarization, automatic execution of smart contracts, multi-source data consistency verification and dynamic value adjustment to achieve objective, transparent and real-time evaluation of the value of aquatic biological assets.

[0006] In this method, upon receiving a request to upload aquaculture pond monitoring data to the blockchain, the monitoring data is stored as evidence on the blockchain network, triggering a pre-deployed smart contract to execute an initial value anchoring operation. By parsing the monitoring data packet, water quality parameter sequences and biomass image sequences are extracted. Outlier removal and interpolation repair are performed on the water quality parameter sequences, and size and pixel value normalization processing is applied to the biomass image sequences, generating cleaned water quality parameter sequences and normalized image sequences. These sequences are packaged into evidence storage data blocks, and their hash values ​​are calculated. The hash values ​​are then associated with the evidence storage data blocks and uploaded to the blockchain, achieving tamper-proof evidence storage of the monitoring data. After successful upload, the smart contract uses the evidence storage data block as input to automatically execute the initial value anchoring function, obtaining an initial evaluation result containing growth stage identifiers, ensuring the transparency and traceability of the basic evaluation process.

[0007] To improve the reliability of the evaluation results, this invention utilizes on-chain multi-source monitoring data to perform consistency verification of the growth stage. Based on the growth stage identifier in the initial value anchoring result, the corresponding standard growth curve template is queried from the blockchain network. Water quality parameter sequences and biomass image sequences from multiple consecutive time points associated with the current aquaculture pond are retrieved as multi-source monitoring data. The biomass image sequences are input into a preset growth stage discrimination model, which prioritizes dividing the image sequences into multiple image subsequences. The model extracts features from each subsequence and outputs the sub-stage discrimination results, which are then statistically analyzed to obtain the discriminated growth stage identifier. When the discriminated growth stage identifier matches the initial growth stage identifier, the multi-source monitoring data is marked as verified; otherwise, the growth stage identifier is corrected based on the discrimination results, and the corrected identifier is associated with and stored in the multi-source monitoring data. This mechanism utilizes objective on-chain data to cross-validate the initial judgment, effectively reducing evaluation errors caused by biases in a single data source or model.

[0008] Based on the verified growth status data, a dynamic value adjustment factor is further determined. The current biomass density and average body length parameters are extracted from the growth status data. Historical transaction records of similar aquatic organisms on the blockchain are queried, and recent transaction records with a time interval less than a preset threshold are selected. For each recent transaction, the ratio of transaction amount to the total biomass traded is calculated to obtain a unit value parameter. The unit value parameter sequence is sorted in ascending order, and the first and third quantile values ​​are extracted to form a market reference value range. Combining the current biomass density parameter, average body length parameter, and market reference value range, a dynamic value adjustment factor reflecting real-time market supply and demand and growth status is calculated. This adjustment factor integrates reliable historical market data on the blockchain with the current growth situation, making the value adjustment more closely aligned with actual market fluctuations.

[0009] The initial value anchoring result is corrected using a dynamic value adjustment factor. The underlying asset value is read, multiplied by the dynamic value adjustment factor, and then normalized to ensure the corrected value falls within a preset value fluctuation range, resulting in a corrected asset valuation result. A valuation record transaction containing this result and the current timestamp is generated, and the transaction is digitally signed to verify the sender's valuation authority. After successful verification, consensus nodes verify whether the valuation result matches the latest on-chain data. Once confirmed, the transaction is packaged into a block and appended to the end of the blockchain using a consensus algorithm. Thus, the entire valuation process is stored and verified on-chain, ensuring the valuation result is immutable, auditable, and dynamically responds to growth conditions and market changes, providing a reliable, automated, and highly timely valuation method for aquatic assets.

[0010] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0011] In response to the request to upload monitoring data from the aquaculture pond to the blockchain, the monitoring data, including water quality parameter sequences and biomass image sequences, is stored on the blockchain, triggering a smart contract to execute the initial value anchoring operation. Based on the growth stage identifier in the initial anchoring result, the corresponding standard growth curve template is queried from the blockchain network. Simultaneously, water quality parameter sequences and biomass image sequences from multiple consecutive time points associated with the current aquaculture pond are retrieved as multi-source monitoring data. The biomass image sequences from the multi-source monitoring data are input into a preset growth stage discrimination model to obtain a discriminant growth stage identifier, which is then compared with the initial growth stage identifier. When they match, the multi-source monitoring data is marked as valid growth state data; when they do not match, the result of the initial value anchoring operation is corrected based on the discriminant growth stage identifier. By automatically cross-validating the growth stage using tamper-proof multi-source monitoring data on the blockchain, erroneous growth stage information introduced by relying solely on a single data source or manual judgment is avoided. This ensures that the initial value anchoring is based on the verified true growth state, eliminating the distortion of the assessed value caused by misjudgments of the growth stage due to equipment malfunctions or environmental interference. Based on the verified growth status data, the current biomass density parameter and the current average body length parameter are extracted. The historical transaction records corresponding to similar aquatic organisms stored on the blockchain are queried, and recent transaction records with a time difference of less than a preset time threshold are selected. The unit value parameter of each recent transaction is calculated, and statistical processing of the unit value parameter yields a market reference value range. Then, a dynamic value adjustment factor is calculated based on the current biomass density parameter, the current average body length parameter, and the market reference value range. The dynamic value adjustment factor is used to perform product correction and normalization constraint processing on the underlying asset value value in the initial value anchoring result, generating a corrected asset value assessment result. The assessment record transaction containing the assessment result and the current timestamp is broadcast to the blockchain consensus node, verified, and packaged into a block. By directly driving the calculation of the value adjustment factor through on-chain historical transaction records, discrete transaction prices are transformed into continuous market reference value ranges. This allows the assessment result to reflect recent market transaction conditions in real time, avoiding assessment lag and market deviation caused by static unit price settings, and enhancing the reference value of the assessment value for actual transaction scenarios. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0013] Figure 1 This is a flowchart of a blockchain-based method for assessing the value of aquatic biological assets in aquaculture.

[0014] Figure 2 This is a flowchart for verifying the consistency of growth stages of aquatic organisms in culture.

[0015] Figure 3 This is a schematic diagram of outlier removal and cleaning of water quality parameter sequences;

[0016] Figure 4 This is a schematic diagram showing the distribution of unit value parameters and market reference value range of recently traded aquatic organisms. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1 This invention provides a blockchain-based method for assessing the value of aquatic biological assets, comprising: responding to a request to upload monitoring data from an aquaculture pond to the blockchain, storing the monitoring data on the blockchain network, and triggering a smart contract to perform an initial value anchoring operation; based on the result of the initial value anchoring operation, calling the multi-source monitoring data stored on the blockchain to perform consistency verification on the growth stage of the aquatic organisms, obtaining verified growth status data; based on the verified growth status data, combined with on-chain historical transaction records, determining a dynamic value adjustment factor for the current aquatic biological asset; using the dynamic value adjustment factor to correct the result of the initial value anchoring operation, generating a corrected asset value assessment result, and writing the assessment result back to the blockchain network.

[0019] In practice, the process of responding to the request to upload the monitoring data of the aquaculture pond to the blockchain and storing the monitoring data on the blockchain network and triggering the smart contract to execute the initial value anchoring operation includes monitoring data packet parsing and sequence extraction, data cleaning and standardization processing, generation of evidence data blocks and triggering of smart contracts.

[0020] During the monitoring data packet parsing and sequence extraction process, the monitoring data packet carried in the on-chain request is deserialized to obtain the data packet header and body. The data type identifier is read from the data packet header. This identifier is an encoded value that distinguishes water quality parameters from biomass images; a value of 0x01 indicates water quality parameter data, and a value of 0x02 indicates biomass image data. The corresponding data area in the data packet body is located based on the data type identifier: if the identifier is 0x01, the location is the water quality parameter data area; if the identifier is 0x02, the location is the biomass image data area. Water quality parameter values ​​are read sequentially from the water quality parameter data area at preset time intervals to generate a water quality parameter sequence. The preset time interval is set according to the reporting cycle of the acquisition device; for example, if the sampling cycle of the dissolved oxygen sensor in the aquaculture pond is 30 seconds, the preset time interval is set to 30 seconds. Multiple frames of biomass images are read from the biomass image data area and arranged from earliest to latest according to the acquisition timestamp of each frame to generate a biomass image sequence. The biomass images are acquired and uploaded by the underwater camera device at a frequency of one frame per minute.

[0021] During data cleaning and normalization, outlier removal was performed on the water quality parameter sequence to obtain a cleaned water quality parameter sequence. Simultaneously, the biomass image sequence was normalized to obtain a normalized image sequence. When removing outliers from the water quality parameter sequence, the mean and standard deviation of all water quality parameter values ​​were first calculated. For each water quality parameter value in the sequence, the deviation between that value and the mean was calculated. Water quality parameter values ​​with a deviation greater than three times the standard deviation were marked as outliers. The basis for setting three times the standard deviation is that, under the assumption of normal distribution, the probability of data points exceeding three times the standard deviation being extreme outliers is less than 0.3%, which can effectively identify abnormal readings caused by sensor momentary malfunctions or transmission errors. The water quality parameter values ​​marked as outliers were deleted from the water quality parameter sequence, and linear interpolation was used to fill the deleted positions. The calculation method for linear interpolation is as follows:

[0022]

[0023] in, This indicates the water quality parameter values ​​generated during the filling process. This indicates the most recent water quality parameter value that was not marked as an outlier before the deletion location. This represents the nearest water quality parameter value that was not marked as an outlier after the deletion location. This indicates the time sequence number corresponding to the filling position. express The corresponding time sequence number, express The corresponding time sequence number. If the deleted outlier is located at the beginning of the water quality parameter sequence and there are no valid values ​​before it, then use the time sequence number directly. Fill in; if the deleted outlier is located at the end of the water quality parameter sequence and there are no valid values ​​after it, then use directly. Filling. After filling is complete, the water quality parameter sequence after cleaning is obtained.

[0024] When normalizing the biomass image sequence, firstly, a size normalization operation is performed on each frame of the biomass image sequence, uniformly adjusting the pixel size of all biomass images to a standard size of 224 pixels × 224 pixels. This size setting is based on adapting to the standard input layer requirements of the convolutional neural network model used for subsequent growth stage discrimination. After size normalization, a pixel value normalization operation is performed on each frame of the biomass image, linearly mapping the pixel value of each pixel in the size-normalized biomass image from the original value range [0,255] to the interval [0,1]. The mapping method is to divide each pixel value by 255 to obtain the normalized image sequence.

[0025] During the generation of the evidence data block and the triggering of the smart contract, the cleaned water quality parameter sequence and the standardized image sequence are packaged into a single evidence data block. A structured data format is used for packaging; the cleaned water quality parameter sequence is stored as an array, and each frame of the standardized image sequence is Base64 encoded and stored as a string array. A data collection timestamp and aquaculture pond identifier are appended to the evidence data block. The hash value of the evidence data block is calculated using the SHA-256 hash algorithm. The calculated hash value is then associated with the evidence data block, and a blockchain transaction is initiated to write the association into the distributed ledger of the blockchain network, completing the monitoring data notarization. Once the blockchain network confirms the successful on-chain storage of the evidence data block, the initial value anchoring function in the pre-deployed smart contract is called, with the hash value of the evidence data block passed as an input parameter. The smart contract reads the stored data block from the chain based on the input hash value, performs the initial value anchoring operation, which includes identifying the types and densities of aquatic organisms based on standardized image sequences, comprehensively evaluating the growth stage by combining the cleaned water quality parameter sequence, and outputting the initial value anchoring operation result containing the growth stage identifier and the value of the underlying asset.

[0026] See Figure 3The horizontal axis of the graph represents time sequence, and the vertical axis represents dissolved oxygen concentration, in mg / L. The gray dashed curve represents the original water quality parameter sequence, which contains several outliers that significantly deviate from the overall trend; these outliers are marked with red crosses. The black solid curve represents the cleaned water quality parameter sequence after outlier removal and linear interpolation. The cleaned data curve is smooth and without significant abrupt changes, more accurately reflecting the actual changes in aquaculture pond water quality. Outlier determination is based on deviation thresholds; the concentration values ​​corresponding to the multiple red crosses in the graph are significantly higher or lower than the overall mean, verifying the effectiveness of the outlier removal method. The cleaned water quality parameter sequence in the graph exhibits periodic fluctuations within the time sequence range of approximately 0 to 800, with concentration values ​​mostly stable between 5 and 9 mg / L, meeting the dissolved oxygen concentration reference range requirements for aquatic organism growth.

[0027] In specific implementation, please refer to Figure 2 Based on the results of the initial value anchoring operation, the process of calling on-chain stored multi-source monitoring data to perform consistency verification on the growth stages of aquatic organisms and obtain the verified growth status data includes obtaining growth stage identifiers and standard growth curve templates, retrieving multi-source monitoring data, performing consistency comparison based on the growth stage discrimination model, and processing the comparison results.

[0028] The growth stage identifier is extracted from the result of the initial value anchoring operation. This identifier is a string value output by the smart contract during the initial value anchoring operation, corresponding to different growth stages of the cultured aquatic organisms. For example, "STAGE_LARVAE" represents the larval stage, "STAGE_JUVENILE" represents the juvenile stage, and "STAGE_ADULT" represents the adult stage. After obtaining the growth stage identifier, it is used as the query key to retrieve a pre-stored standard growth curve template from the world state database of the blockchain network. The standard growth curve template is reference data developed by aquaculture experts based on the biological characteristics of specific cultured aquatic organisms. The template includes the dissolved oxygen concentration threshold range, ammonia nitrogen concentration threshold range, biomass density reference range, and the duration range of the corresponding growth stage. The standard growth curve template is written to the blockchain after being signed by the management node, and its storage hash forms a mapping relationship with the growth stage identifier.

[0029] Water quality parameter sequences and biomass image sequences from multiple consecutive time points associated with the current aquaculture pond are retrieved from the blockchain network as multi-source monitoring data. The retrieval process uses the unique identifier of the current aquaculture pond and a preset backtracking time window as parameters to query all stored data blocks on the blockchain whose timestamps fall within the backtracking time window. The length of the preset backtracking time window is set based on the growth rate of the aquatic organisms being cultured; for fish with moderate growth rates, the preset backtracking time window is set to 14 days. The retrieved stored data blocks are arranged in ascending order of timestamps. From each stored data block, the cleaned water quality parameter sequence and standardized image sequence are parsed. All cleaned water quality parameter sequences are then concatenated in chronological order to form a water quality parameter sequence, and all standardized image sequences are concatenated in chronological order of collection time to form a biomass image sequence. The resulting concatenated water quality parameter sequence and biomass image sequence constitute the multi-source monitoring data.

[0030] Before inputting the biomass image sequence from multi-source monitoring data into the preset growth stage discrimination model, the biomass image sequence is first divided into image subsequences. Multiple image subsequences are extracted from the biomass image sequence using a sliding window method, according to the acquisition time sequence. Each image subsequence contains a fixed number of consecutive biomass images, F, where F is set to 8. The reason for setting F to 8 is that significant morphological features of a complete behavioral expression of cultured aquatic organisms can usually be observed within the time span of 8 images, and the computational latency and memory usage of 8 frames of data are within a balanced range during subsequent inference of the growth stage discrimination model. The sliding window step size is set to 4 frames, resulting in a 50% overlap between adjacent image subsequences. This overlap setting helps smooth the fluctuations in the recognition of temporal boundary features by the growth stage discrimination model. Taking a biomass image sequence containing 8000 biomass images as an example, after division with a sliding step size of 4, approximately 1999 image subsequences are obtained.

[0031] The growth stage discrimination model is a deep neural network that integrates spatial feature encoding and time series modeling. The core architecture of the model consists of three modules connected in series: a spatial feature encoding network, a temporal feature aggregation network, and a growth stage classifier. The spatial feature encoding network uses a ResNet-34 convolutional neural network with fully connected layers removed, containing one 7×7 convolutional layer, one max-pooling layer, 16 stacked residual modules, and one average pooling layer, generating a spatial feature vector with a dimension of 512 at the output. The temporal feature aggregation network uses a two-layer bidirectional long short-term memory network, with each layer having 128 hidden units. The two bidirectional long short-term memory layers are stacked along the time dimension, mapping the input spatial feature vector sequence to a fixed-length context aggregation vector, with an output dimension of 256. The growth stage classifier consists of a fully connected layer and a softmax activation layer. The input dimension of the fully connected layer is 256, and the output dimension is equal to the total number of categories M of the growth stages of aquatic organisms. The value of M is 5, corresponding to the five stages of shrimp larvae, juvenile shrimp, intermediate shrimp, adult shrimp, and broodstock shrimp. After the fully connected layer outputs the logits vector, the softmax activation layer calculates the confidence probability of each growth stage.

[0032] The training process of the growth stage discrimination model is as follows: Biomass image sequence data with labeled growth stage categories are collected from the historical aquaculture monitoring system. Professional annotators label the growth stage category of the image subsequence corresponding to each frame window with a reference to aquaculture logs and underwater observation records. The labeled dataset is randomly divided into training, validation, and test sets in a 7:2:1 ratio. During training, the multi-class cross-entropy loss function is used, the momentum stochastic gradient descent optimizer is used, the momentum coefficient is set to 0.9, the weight decay coefficient is set to 0.0001, the initial learning rate is set to 0.01, the batch size is set to 32, and the total number of training iterations is set to 80. After each iteration, the classification accuracy is calculated on the validation set. If the validation set accuracy does not improve for four consecutive iterations, the learning rate is decayed to 0.5 times the current value. An early stopping mechanism is also adopted during training. When the validation set accuracy does not exceed the historical best value for 10 consecutive iterations, the training is terminated early, and the model parameters at the optimal validation set accuracy are saved. After training, the generalization ability of the growth stage discrimination model is verified using a test set, and the model structure and parameters are solidified.

[0033] After obtaining the growth stage discrimination model, each segmented image subsequence is sequentially fed into the model. The spatial feature encoding network independently extracts a 512-dimensional spatial feature vector for each frame of biomass image in the subsequence, generating a spatial feature sequence of size F×512. The temporal feature aggregation network inputs the spatial feature sequence step-by-step, and after computation by a two-layer bidirectional long short-term memory network, outputs a 256-dimensional context aggregation vector carrying the contextual information of the entire image subsequence. The growth stage classifier receives the 256-dimensional context aggregation vector, maps it to a 5-dimensional logits vector through a fully connected layer, and the softmax activation layer converts the logits vector into five confidence probability values. The growth stage category corresponding to the maximum confidence probability is taken as the sub-stage discrimination result for the corresponding image subsequence.

[0034] Voting statistics are performed on the sub-stage discrimination results of all image sub-sequences. The frequency of each growth stage category in all sub-stage discrimination results is counted, and the category with the highest frequency is determined. When the category with the highest frequency is unique, it is determined as the growth stage identifier. When multiple categories have the same highest frequency, for each category with the same frequency, the softmax confidence probability values ​​corresponding to all image sub-sequences classified as that category are accumulated to obtain the total category confidence score. The category with the highest total category confidence score is selected as the growth stage identifier. If the total category confidence scores are still equal, based on the time priority principle, the category identified by the last corresponding image sub-sequence appearing on the biomass image sequence timeline is selected as the growth stage identifier.

[0035] After obtaining the growth stage identifier, the string corresponding to the growth stage identifier is compared with the string corresponding to the growth stage identifier contained in the initial value anchoring operation result. In the comparison, a case-sensitive exact match is used to ensure that the encoded values ​​are completely identical.

[0036] When the growth stage identifier is found to be consistent with the growth stage identifier in the initial value anchoring operation result, the retrieved multi-source monitoring data is directly marked as the verified growth state data. The verified growth state data includes the unmodified original growth stage identifier, water quality parameter sequence and biomass image sequence. These data are stored as a whole in the temporary data cache area as an integrated record.

[0037] When the discrepancy between the discrepancy between the growth stage identifier and the growth stage identifier in the initial value anchoring operation result is found, the discrepancy growth stage identifier is used as the corrected growth stage identifier, and the result of the initial value anchoring operation is corrected. The correction process includes replacing the growth stage identifier field in the initial value anchoring operation result with the corrected growth stage identifier, triggering the blockchain network to query the standard growth curve template mapped to the corrected growth stage identifier, and packaging the corrected growth stage identifier and the queried standard growth curve template to form a correction record. Subsequently, the correction record is correlated with the retrieved multi-source monitoring data to generate verified growth status data. The verified growth status data includes the corrected growth stage identifier, the correlated standard growth curve template data, water quality parameter sequences, and biomass image sequences. The verified growth status data is then persistently stored in the verification result database of the blockchain network, completing all operations for growth stage consistency verification.

[0038] In practice, the process of determining the dynamic value adjustment factor of aquatic biological assets based on verified growth status data and on-chain historical transaction records includes extracting current biomass density parameters and current average body length parameters, screening recent transaction records, statistically calculating market reference value ranges, and calculating dynamic value adjustment factors.

[0039] The current biomass density parameter and the current average body length parameter are extracted from the validated growth status data. The validated growth status data stores a water quality parameter sequence and a biomass image sequence confirmed through consistency verification. The water quality parameter sequence includes continuous values ​​for dissolved oxygen concentration, ammonia nitrogen concentration, temperature, and pH, while the biomass image sequence contains temporally continuous underwater images of aquatic organisms. The current biomass density parameter is extracted as follows: The latest timestamped single-frame biomass image from the biomass image sequence is processed using a deep learning-based biomass counting model. The biomass counting model uses the CenterNet object detection framework as its core and Hourglass-104 as its backbone feature extraction network. The input layer of the backbone feature extraction network receives tensors of size 512×512×3. Multi-scale heatmap features are extracted step-by-step through stacked hourglass modules, outputting the center point coordinates and bounding box regression parameters of the detected target. The biomass counting model is trained on an image dataset containing 150,000 labeled aquatic organisms. During training, the input data consists of biomass images of the aquaculture pond captured by an underwater camera. The labeled data consists of the center point coordinates and bounding box of each aquatic organism in the image. The loss function is composed of the weighted sum of the focal loss from center point prediction and the L1 loss from bounding box regression. The focal loss weight coefficient is set to 1.0, and the L1 loss weight coefficient is set to 0.1. After the biomass counting model is trained, it performs inference on a single frame of biomass image, counts the total number of detected aquatic organisms, and divides the total number of aquatic organisms by the area of ​​the aquaculture pond to obtain the current biomass density parameter. The area of ​​the aquaculture pond is read from the basic archive data of the aquaculture pond stored on the blockchain. The current method for extracting the average body length parameter is as follows: For each aquatic organism detected by the biomass counting model, the estimated body length of each aquatic organism is calculated by using the pixel size of the bounding box and the pre-calibrated pixel-to-physical size conversion ratio. The pre-calibrated pixel-to-physical size conversion ratio is obtained by calibrating the underwater camera device using a calibration plate with known physical size during installation and has been written into the basic data of the aquaculture pond and stored on the blockchain. The arithmetic mean of the estimated body lengths of all aquatic organisms is taken to obtain the current average body length parameter.

[0040] After obtaining the current biomass density and average body length parameters, the system queries the blockchain network for the set of historical transaction records corresponding to similar aquatic organisms, using these parameters as search criteria. The criteria for determining similar aquatic organisms are that both the species identifier and growth stage identifier match. The species identifier is a unique code for the aquatic organism species, and the growth stage identifier is the one confirmed in the verified growth status data. In the blockchain network, each aquatic organism transaction record is stored as a transaction in the distributed ledger. Each transaction includes a transaction timestamp, species identifier, growth stage identifier, total biomass, total transaction amount, biomass density parameter at the time of the transaction, and average body length parameter at the time of the transaction. When querying the set of historical transaction records, a query request is constructed using the species identifier and growth stage identifier as a combined index key. The world state database of the blockchain node returns a list of all matching transaction records. From the list of transactions, the time difference between the current system time and the timestamp of each transaction is calculated. A preset time threshold of 30 days is set. The rationale for setting 30 days is that the price fluctuation cycle of aquatic organisms in the aquaculture market typically maintains relative continuity within 30 days, and the correlation between historical transaction prices and current market supply and demand weakens significantly after 30 days. Transactions with a time difference of less than 30 days between their timestamp and the current system time are selected, and these transactions constitute the recent transaction records.

[0041] For each transaction in the recent transaction records, calculate the unit value parameter. Read the total transaction amount and total biomass fields from the transaction. Divide the total transaction amount by the total biomass to obtain the unit value parameter, which represents the transaction price per unit of biomass. The unit of biomass can be one fish or one kilogram, depending on the species being farmed. Sort all unit value parameters in ascending order using the quicksort algorithm, resulting in a unit value parameter sequence. Based on the sorting result, extract the first and third quantile values ​​from the unit value parameter sequence. The first quantile value is extracted by calculating the position index. ,in, This represents the position index corresponding to the first quantile. This represents the total number of unit value parameters in the unit value parameter sequence; when When the value is an integer, the first quantile is taken from the unit value parameter sequence. The unit value parameter value for each location; when When it is a non-integer, let The integer part is The decimal part is The first quantile value is derived from the unit value parameter sequence. The unit value parameter value of the position is the same as that of the position. The unit value parameter value at each position is calculated using linear interpolation, where the first quantile value equals the value at the second quantile value. The unit value parameter value of each position plus Multiply by the first The unit value parameter value of the position is the same as that of the position. The difference in unit value parameter values ​​at each location. The third quantile value is extracted by calculating the location index. ,in, Indicates the position index corresponding to the third quantile; when When the value is an integer, the third quantile is taken from the unit value parameter sequence. The unit value parameter value for each location; when When it is a non-integer, let The integer part is The decimal part is The third quantile value is derived from the unit value parameter sequence. The unit value parameter value of the position is the same as that of the position. The unit value parameter value at each position is calculated using linear interpolation, where the third quantile value equals the value at the 1st position. The unit value parameter value of each position plus Multiply by the first The unit value parameter value of the position is the same as that of the position. The difference in unit value parameter values ​​at each position. The closed interval formed by the first quantile value and the third quantile value is defined as the market reference value interval, with the lower boundary of the market reference value interval being the first quantile value and the upper boundary being the third quantile value.

[0042] After obtaining the market reference value range, the dynamic value adjustment factor is calculated by combining the current biomass density parameter and the current average body length parameter. The standard biomass density reference range and the standard average body length reference range are read from the standard growth curve template associated with the validated growth status data. The current biomass density parameter is compared with the median of the standard biomass density reference range, and the current average body length parameter is compared with the median of the standard average body length reference range to obtain the degree of density deviation and body length deviation, respectively. The formula for calculating the dynamic value adjustment factor is:

[0043]

[0044] in, Indicates the dynamic value adjustment factor. The value is limited to the range [0.7, 1.3]. When the calculated result is less than 0.7, 0.7 is used; when the calculated result is greater than 1.3, 1.3 is used. The basis for limiting the value range is that the adjustment range of the asset value within a single appraisal period shall not exceed ±30%. This represents the species influence coefficient, a preset constant related to the aquatic organisms being cultured. It is determined by a publicly available industry price index for the species, specifically for Australian lobster species. The value is 0.95, for the grouper species. The value is 1.05, for varieties not explicitly specified. Use the default value of 1.0; It represents the median of the market reference value range, and is calculated by dividing the sum of the lower and upper boundaries of the market reference value range by 2. This represents the interquartile range of the market reference value interval, calculated by subtracting the lower boundary of the market reference value interval from the upper boundary. When it is 0, The value is adjusted to 0.01 times the mean of the unit value parameter sequence to avoid division by zero. This represents the density weighting coefficient, which is set based on the growth characteristics of the cultured aquatic organisms. For species whose culture density has a greater impact on value, the density weighting coefficient is 0.7; for species whose culture density has a smaller impact on value, the density weighting coefficient is 0.3; and for other species, the density weighting coefficient is 0.5. This represents the body length weighting coefficient, which satisfies the following condition: the body length weighting coefficient and the density weighting coefficient satisfy... The constraint relationship, The value is determined by Calculated; This represents the current biomass density parameter; This represents the median of the standard biomass density reference range; This represents the current average body length parameter; This represents the median of the standard average body length reference range. After calculating the dynamic value adjustment factor, it is written into the extended field of the validated growth status data for storage, and then read and used during the asset value correction process.

[0045] See Figure 4 In the graph, the horizontal axis represents the unit value parameter in "yuan / tail", and the vertical axis represents the probability density of the unit value parameter. The light blue bar chart shows the frequency distribution of the unit value parameter corresponding to recent on-chain transaction records, and the dark blue curve is the probability density estimation curve based on this distribution. The graph uses three vertical dashed lines to indicate the statistically obtained first quantile, third quantile, and median values, specifically the first quantile value of 2.69 yuan / tail (orange dashed line), the median of 3.43 yuan / tail (green dashed line), and the third quantile value of 4.31 yuan / tail (red dashed line). The light gray area represents the market reference value range, defined by the closed interval [2.69, 4.31] formed by the first and third quantile values.

[0046] This diagram illustrates the specific technical details of calculating the market reference value range based on recent transaction records in one of the above embodiments. By collecting transaction data of similar farmed aquatic organisms at the same growth stage stored on the blockchain, the unit value parameter of each transaction is extracted. The unit value parameters are then sorted in ascending order using a quicksort (or mergesort) algorithm to calculate the first and third quantiles, forming a reference range reflecting the recent market price fluctuations of this type of farmed aquatic organism. This market reference value range provides a statistical basis for the subsequent calculation of dynamic value adjustment factors, ensuring that the asset valuation process takes into account reasonable market fluctuations.

[0047] In practice, the process of using dynamic value adjustment factors to correct the results of the initial value anchoring operation, generating a corrected asset value assessment result, and writing the assessment result back to the blockchain network includes reading the value of the basic asset and initial correction calculation, normalization constraint processing, generation of assessment record transactions, consensus verification, and block writing.

[0048] This function retrieves the underlying asset value from the result of the initial value anchoring operation. The result of the initial value anchoring operation is generated and stored in the world state database of the blockchain network during the execution of the smart contract. The underlying asset value is a field in the result of the initial value anchoring operation, representing the initial valuation of a unit of aquatic organism asset under the current monitoring data and standard growth curve template reference. The currency unit of the underlying asset value is RMB. By calling the smart contract's constant read function and passing the hash value of the stored data block as the index parameter, the smart contract returns the precise floating-point value of the underlying asset value.

[0049] The initial revised value is obtained by multiplying the read base asset value by the dynamic value adjustment factor. This multiplication is performed on the off-chain evaluation node, using double-precision floating-point arithmetic to preserve computational accuracy. The dynamic value adjustment factor is calculated by combining biomass density parameters, average body length parameters, and market reference value ranges. As a dimensionless adjustment coefficient, the initial revised value obtained by multiplying the base asset value by the dynamic value adjustment factor reflects a preliminary revised valuation that incorporates actual growth conditions and market trends.

[0050] The initial revised value is normalized to ensure it falls within a preset value fluctuation range, resulting in a normalized revised value. The preset value fluctuation range is centered on the underlying asset value, with preset fluctuation tolerance ratios extending both above and below it. The preset fluctuation tolerance ratio is set at 20%, based on the assumption that the value fluctuation of aquatic biological assets between two adjacent valuation periods, caused by normal growth fluctuations and market changes, typically does not exceed ±20%. Valuation changes exceeding this range are considered abnormal deviations and require constraint correction. The lower bound of the preset value fluctuation range is 0.8 times the underlying asset value, and the upper bound is 1.2 times the underlying asset value. The normalization constraint processing uses a piecewise constraint mapping function, calculated as follows:

[0051]

[0052] in, This represents the normalized adjusted value. This represents the initial correction value. This indicates the lower bound of the preset value fluctuation range. , This represents the numerical value of the underlying assets. This indicates the upper bound of the preset value fluctuation range. When the initial correction value is... Less than the lower bound At that time, Assigned to the normalized corrected value When the initial correction value is... Greater than the upper bound At that time, Assigned to the normalized corrected value When the initial correction value is... Falling and In between, the original value is kept unchanged as the normalized corrected value. After the normalization constraint is completed, the normalized adjusted value is taken as the adjusted asset valuation result.

[0053] After obtaining the revised asset valuation result, an valuation record transaction is generated, containing the revised asset valuation result and the current timestamp. The valuation record transaction is constructed using a unified transaction data structure on the blockchain network. This transaction data structure includes standard fields and extended payload fields. The standard fields include the sender's address, receiver's address, transaction type identifier, and digital signature. The extended payload fields are serialized in JSON format. The key-value pairs written in the extended payload fields include: the "Revised Asset Value" field stores the numeric string of the revised asset valuation result; the "Valuation Timestamp" field stores the Unix millisecond-level timestamp representation of the current timestamp; the "Evidence Block Hash" field stores the hash value of the evidence block on which this valuation is based; the "Dynamic Value Adjustment Factor" field stores the numeric string of the dynamic value adjustment factor used; and the "Initial Value Anchoring Identifier" field stores the execution transaction hash of the initial value anchoring operation. The sender address of the assessment record transaction is set to the blockchain account address of the assessment computing node, the receiver address is set to the blockchain account address of the aquatic biological asset owner, and the transaction type identifier is set to "0xA3", where 0xA3 is a predefined asset assessment result notarization transaction type code.

[0054] Before the evaluation record transaction is broadcast to the consensus nodes in the blockchain network, the evaluation computing node uses its own private key to perform elliptic curve digital signature on the entire extended data payload field of the evaluation record transaction, generates a digital signature value, and fills the digital signature value into the digital signature field of the evaluation record transaction.

[0055] The process of broadcasting the evaluation record transaction to the consensus nodes in the blockchain network, and then having the consensus nodes verify the evaluation record transaction and package it into a block, includes digital signature verification, permission verification, data matching verification, and block packaging consensus.

[0056] During the digital signature verification process, after receiving the broadcast evaluation record transaction, the consensus node extracts the sender's blockchain account address from the sender's address field of the evaluation record transaction. Based on the sender's blockchain account address, it queries the sender's public key storage record and uses the retrieved sender's public key to verify the digital signature field of the evaluation record transaction. The verification operation recovers the digest value of the extended data payload field and compares it with the recalculated digest value of the received extended data payload field. If the verified digest values ​​match, it confirms that the evaluation record transaction has not been tampered with and the sender's identity is genuine.

[0057] During the process of confirming the sender's identity and assessment authority, the consensus node queries the assessment node whitelist stored in the authority management contract. The assessment node whitelist contains the blockchain account addresses of all nodes authorized to perform asset valuation operations. The sender's blockchain account address is compared one by one with the addresses in the assessment node whitelist. If the sender's blockchain account address exists in the assessment node whitelist, the sender's identity and assessment authority are confirmed.

[0058] After confirming the sender's identity and assessment authority, a verification operation is performed to match the revised asset valuation result with the latest monitoring data stored in the blockchain network. The consensus node reads the corresponding stored data block from the blockchain distributed ledger based on the "stored data block hash" field value in the extended data payload field of the assessment record transaction. It verifies whether the content of the stored data block is still the most up-to-date stored data block related to the current aquaculture pond in the current blockchain network by querying the latest stored data block hash value indexed by the aquaculture pond identifier and comparing it with the "stored data block hash" field value in the assessment record transaction. If the two hash values ​​match, it indicates that the monitoring data on which the revised asset valuation result is based matches the latest stored data in the blockchain network.

[0059] When the revised asset valuation result matches the latest monitoring data, the consensus node places the valuation record transaction into its pending transaction pool. The pending transaction pool is a transaction cache queue maintained in the consensus node's memory, arranging all legitimate transactions in order of receipt time. When multiple pending transactions exist, the consensus node sorts them from highest to lowest Gas price, prioritizing transactions offering higher fees.

[0060] Consensus nodes extract transactions from the pool of transactions to be packaged according to the block packaging cycle. After executing the consensus algorithm, a new block containing the evaluation record transaction is appended to the end of the blockchain. The block packaging cycle is set to 10 seconds. The reason for setting 10 seconds is that, while balancing transaction confirmation latency and blockchain network bandwidth overhead, the 10-second block interval allows nodes in the distributed network sufficient time to complete transaction synchronization and consensus information exchange. The number of transactions extracted from the pool of transactions to be packaged is determined by the block capacity limit, which is set to 5000 transactions or 16 megabytes, whichever comes first. Consensus nodes construct a new block, which includes a block header and a list of transactions. The block header includes the hash value of the previous block, the Merkle root, the timestamp, and the block height. The list of transactions includes all transactions to be confirmed, including the extracted evaluation record transaction. Consensus nodes execute a practical Byzantine fault-tolerant consensus algorithm. Through a three-phase message exchange of pre-preparation, preparation, and commit, consensus is reached when more than two-thirds of the consensus nodes agree. The new block is then appended to the end of the blockchain, completing the write-back of the evaluation results. Once the write-back is complete, the revised asset valuation results are permanently recorded on the blockchain network, making them available for query and verification by owners of aquatic life assets, regulators, and transaction participants.

[0061] In practice, the process of parsing the monitoring data packets carried in the on-chain request and extracting the water quality parameter sequences and biomass image sequences from the monitoring data packets is as follows: The monitoring data packets carried in the on-chain request are deserialized. These data packets are generated by the edge gateway at the aquaculture pond using the Protocol Buffers serialization protocol, and the encoding format is a binary byte stream. During the deserialization operation, the deserializer corresponding to the encoding protocol is called. The deserializer first reads the first 4 bytes of the byte stream as a total data packet length identifier, and then reads the next 2 bytes as a data packet header length identifier. Based on the data packet header length identifier, the end position of the data packet header and the start position of the data packet body are located, thus separating the monitoring data packet into a data packet header and a data packet body. The data type identifier field is read from the data packet header. The data type identifier field occupies 1 byte. When the data type identifier field is 0x01, it indicates that the data in the data packet body contains water quality parameter data; when the data type identifier field is 0x02, it indicates that the data in the data packet body contains biomass image data. The corresponding data area is located from the data packet body based on the value of the data type identifier field: when the data type identifier field is 0x01, the area from offset 0 to half the total length of the data packet body is read as the water quality parameter data area; when the data type identifier field is 0x02, the area from half the total length of the data packet body to the end is read as the biomass image data area. Water quality parameter values ​​are read sequentially from the water quality parameter data area at preset time intervals. The preset time interval is determined by the sampling frequency of the aquaculture pond monitoring equipment and is set to 60 seconds. The water quality parameter values ​​are stored continuously in groups of 60 seconds. Each group contains an ordered arrangement of four indicators: dissolved oxygen concentration, ammonia nitrogen concentration, temperature, and pH value. All the read water quality parameter values ​​are arranged in chronological order to generate a water quality parameter sequence. Multiple biomass images are read from the biomass image data area. The biomass images are stored continuously frame by frame in JPEG compression format. The frames are separated by a two-byte separator 0xFFD9 followed by 0xFFD8. The complete JPEG image data of each frame is extracted by scanning the separator and decoded into bitmap format. The acquisition timestamp metadata of each biomass image is read. The decoded multiple biomass images are arranged in order from earliest to latest acquisition timestamp to generate a biomass image sequence.

[0062] In the specific implementation of outlier removal processing of the water quality parameter sequence to obtain a cleaned water quality parameter sequence, and normalization processing of the biomass image sequence to obtain a normalized image sequence, the deviation value of each water quality parameter value in the water quality parameter sequence is first calculated. For each water quality parameter value in the sequence, the arithmetic mean of all water quality parameter values ​​in the sequence is calculated. Then, the absolute value of the difference between each water quality parameter value and the arithmetic mean is taken to obtain the deviation value corresponding to each water quality parameter value. The preset deviation threshold is set to 2.5 times the standard deviation of all water quality parameter values ​​in the sequence. The basis for setting 2.5 times the standard deviation is that, under the premise that the environmental parameters of the aquaculture water body approximately follow a normal distribution, the probability of data exceeding 2.5 times the standard deviation is less than 1.24%, which can effectively cover outliers caused by occasional sensor drift or short-term abnormal fluctuations. Water quality parameter values ​​with deviation values ​​greater than the preset deviation threshold are marked as outliers. The water quality parameter values ​​marked as outliers in the sequence are deleted, and linear interpolation is used to fill the deleted positions to obtain the cleaned water quality parameter sequence. The calculation formula for linear interpolation is as follows:

[0063]

[0064] in, This represents the filling water quality parameter values ​​calculated through linear interpolation; This represents the most recent water quality parameter value that was not marked as an outlier before the location where the outlier was deleted; This represents the nearest water quality parameter value that was not marked as an outlier after the location of the deleted outlier. This indicates the time sequence number corresponding to the location of the deleted outlier. The time sequence number is the index number of the water quality parameter value in the water quality parameter sequence. express The corresponding time sequence number; express The corresponding time sequence number. If the deleted outlier is located before the beginning of the water quality parameter sequence, it does not exist. Then take directly As filler water quality parameter values; if the deleted outlier is located at the end of the water quality parameter sequence and does not exist thereafter. Then take directly As a parameter value for filling water.

[0065] When normalizing the biomass image sequence, a size normalization operation is performed on each frame of the biomass image sequence. The pixel width and pixel height of each frame of the biomass image are uniformly adjusted to a standard size of 224 pixels × 224 pixels. The adjustment method adopts a bilinear interpolation scaling algorithm to ensure that the contour features of the cultured aquatic organisms in the adjusted image are not distorted. Then, a pixel value normalization operation is performed on each frame of the size-normalized biomass image. The original pixel values ​​of the size-normalized biomass image are integers between 0 and 255. Each pixel value is divided by 255.0 to map the pixel value to a preset numerical range between 0.0 and 1.0, resulting in a normalized image sequence.

[0066] In the specific implementation of inputting biomass image sequences from multi-source monitoring data into a preset growth stage discrimination model to obtain growth stage identifiers and comparing these identifiers with the actual growth stage identifiers, the biomass image sequences are first divided into multiple image subsequences according to the acquisition time sequence. The division method uses a fixed frame-length sliding window, with the window frame count set to 8 frames. The basis for setting 8 frames is that a complete respiratory opening and closing cycle or thoracic limb swinging cycle of aquatic organisms can usually be fully presented within the time span corresponding to 8 frames, and the behavioral features contained in 8 frames are sufficient to support the growth stage discrimination model in extracting discriminative features. The sliding step size is set to 4 frames, with 4 frames overlapping between adjacent image subsequences, to enhance the growth stage discrimination model's ability to capture stage transition boundaries.

[0067] The pre-defined growth stage discrimination model adopts a three-dimensional convolutional residual network architecture. The core architecture of the model consists of a three-dimensional convolutional feature extraction module, a three-dimensional residual aggregation module, and a stage classification output module connected in series. The three-dimensional convolutional feature extraction module contains a 7×7×7 three-dimensional convolutional layer and a 3×3×3 three-dimensional max pooling layer. The three-dimensional convolutional layer has 3 input channels and 64 output channels, with a stride of 2×1×1. The three-dimensional max pooling layer has a stride of 2×2×2. The 3D residual aggregation module consists of four stacked residual block groups. The first residual block group contains two basic 3D residual blocks, each consisting of two 3×3×3 3D convolutional layers and a shortcut connection, with 64 output channels. The second residual block group contains two basic 3D residual blocks with 128 output channels. Upon entering the second residual block group, the number of channels is adjusted using a 1×1×1 3D convolution and spatial downsampling is performed, with a spatial downsampling factor of 2. The third residual block group contains two basic 3D residual blocks with 256 output channels, and spatial downsampling is also performed upon entry. The fourth residual block group contains two basic 3D residual blocks with 512 output channels, and spatial downsampling is performed upon entry. The 3D residual aggregation module outputs a feature tensor with dimensions T×H×W×512, where T represents the time dimension, H represents the height dimension, and W represents the width dimension. A three-dimensional adaptive average pooling layer is connected at the end of the three-dimensional residual aggregation module to pool the feature tensor to a size of 1 in the time, height, and width dimensions, resulting in a feature vector of length 512. The stage classification output module consists of a fully connected layer and a softmax activation function. The fully connected layer has an input dimension of 512 and an output dimension of 5, corresponding to the five growth stages of farmed aquatic organisms: larval stage, juvenile stage, intermediate stage, adult stage, and broodstock stage. The softmax activation function exponentially normalizes the output of the fully connected layer to obtain the confidence probability value for each growth stage.

[0068] The training process of the growth stage discrimination model is as follows: A dataset of image sequences representing the growth stages of aquatic organisms is constructed. Each sample in the dataset is a normalized biomass image sequence of 8 consecutive frames. Aquaculture technicians label each sample with a growth stage category based on aquaculture logs and actual underwater observation records. The dataset contains 60,000 samples and is randomly divided into training, validation, and test sets in an 8:1:1 ratio. During training, the multi-class cross-entropy loss function is used, and the adaptive moment estimation optimizer is selected. The initial learning rate is set to 0.001, the first-order moment decay coefficient is set to 0.9, the second-order moment decay coefficient is set to 0.999, the batch size is set to 16, and the training iterations are set to 60 rounds. After each round of training, the classification accuracy is calculated on the validation set. If the validation set accuracy does not improve for 5 consecutive rounds, the learning rate is decayed to 0.1 times the current value. Decay stops when the learning rate falls below 1×10^-6. An early stopping strategy is adopted. When the validation set accuracy does not exceed the historical best accuracy within 10 consecutive rounds, the training is terminated early, and the model parameters with the highest validation set accuracy are saved as the final discriminant model parameters for the growth stage.

[0069] Each segmented image subsequence is input into the trained growth stage discrimination model. The 3D convolutional feature extraction module and the 3D residual aggregation module extract spatiotemporal features from the eight consecutive frames of biomass images, outputting a 512-dimensional feature vector. The stage classification output module performs a fully connected mapping on the 512-dimensional feature vector and calculates the confidence probability values ​​for each of the five growth stages using the softmax activation function. The growth stage category with the highest confidence probability value is taken as the sub-stage discrimination result for the corresponding image subsequence. A voting statistics process is performed on the sub-stage discrimination results for all image subsequences. All sub-stage discrimination results are iterated over, and a counting mapping table is used to accumulate the number of times each growth stage category appears as a sub-stage discrimination result. After the statistics are completed, the growth stage category with the most accumulated occurrences is selected as the growth stage identifier. When multiple growth stage categories have the highest and equal accumulated occurrences, all image subsequences corresponding to these categories are extracted. The average confidence probability value of each image subsequence belonging to that category, output by the growth stage discrimination model, is calculated for each category. The growth stage category with the largest average confidence probability value is taken as the growth stage identifier. After obtaining the growth stage identifier, the string of the growth stage identifier is compared character by character with the string of the growth stage identifier contained in the initial value anchoring operation result. A strict matching method that distinguishes between uppercase and lowercase is used. If the two strings are completely identical, they are judged to be identical; otherwise, they are judged to be inconsistent.

[0070] In the specific implementation of calculating the unit value parameter corresponding to each transaction in recent transaction records and statistically processing the unit value parameter to obtain the market reference value range, for each transaction in recent transaction records, the total transaction amount and the total number of traded organisms are read from the extended data load field of the transaction. The total transaction amount is stored as a floating-point value in RMB, and the total number of traded organisms is an integer value representing the total number of aquatic organisms traded. The total transaction amount is divided by the total number of traded organisms to obtain the unit value parameter corresponding to that transaction. The unit value parameter is a floating-point value representing the market transaction price of a single aquatic organism. All unit value parameters are sorted in ascending order. During the sorting, the merge sort algorithm is used to sort each unit value parameter from smallest to largest, resulting in a unit value parameter sequence. The first quantile value and the third quantile value are extracted from the unit value parameter sequence. The position index of the first quantile is calculated using the formula: ,in This represents the index of the first quantile. This represents the total number of unit value parameters in the unit value parameter sequence. If... If it is an integer, then take the first value in the unit value parameter sequence. The unit value parameter value at each position is taken as the first quantile value; if Let be a non-integer. The integer part is The decimal part is Then the first quantile value is determined by the 1st quantile. The unit value parameter value of the position is the same as that of the position. The unit value parameter value at the i-th position is obtained by linear interpolation, and the interpolation method is as follows: The unit value parameter value of each position plus Multiply by the first The unit value parameter value of the position is the same as that of the position. The difference in unit value parameter values ​​at each location. Calculate the third quantile location index; the formula for calculating the location index is... ,in, This represents the index of the third quantile. If... If it is an integer, then take the first value in the unit value parameter sequence. The unit value parameter value at each position is taken as the third quantile value; if Let be a non-integer. The integer part is The decimal part is Then the third quantile value is derived from the first quantile. The unit value parameter value of the position is the same as that of the position. The unit value parameter value at each position is obtained by linear interpolation. The closed interval formed by the first quantile value and the third quantile value is set as the market reference value interval, with the left endpoint of the market reference value interval being the first quantile value and the right endpoint being the third quantile value.

[0071] In the specific implementation of broadcasting evaluation record transactions to consensus nodes in the blockchain network, where the consensus nodes verify the evaluation record transactions and package them into blocks, digital signature verification processing is performed on the evaluation record transactions. The digital signature field in the evaluation record transaction is generated by the evaluation computing node using its elliptic curve secp256k1 private key to sign the hash value of the extended data payload field of the evaluation record transaction. The consensus node obtains the sender's blockchain account address from the sender address field of the evaluation record transaction, queries the corresponding secp256k1 public key in the blockchain account state database based on the sender's blockchain account address, and performs a signature recovery operation on the digital signature field using the queried secp256k1 public key to obtain the original hash value. Simultaneously, the consensus node recalculates the hash value of the received extended data payload field of the evaluation record transaction using the SHA-256 hash algorithm. The recovered original hash value is compared with the recalculated hash value. If the two are completely identical, the digital signature verification is successful, confirming that the evaluation record transaction has not been tampered with during transmission and that the sender's identity is genuine. The consensus node queries the whitelist of evaluation permissions maintained in the permission management contract. This whitelist is a set of blockchain account addresses with evaluation permissions. The node compares the sender's blockchain account address with each address in the whitelist. If the sender's address is present in the whitelist, the sender's identity is confirmed to have evaluation permissions. After confirming the sender's identity and evaluation permissions, the consensus node verifies whether the corrected asset value assessment result in the evaluation record transaction matches the latest monitoring data stored in the blockchain network. The consensus node extracts the hash of the evidence storage data block stored in the extended data payload field of the evaluation record transaction, uses this hash to query the corresponding evidence storage data block in the blockchain distributed ledger, and reads the aquaculture pond identifier and timestamp recorded in the evidence storage data block. Simultaneously, the consensus node uses the current aquaculture pool identifier as the key to query the latest notarized data block hash mapping maintained in the blockchain world state database for that aquaculture pool. It compares the hash of the latest notarized data block with the hash of the notarized data block in the evaluation record transaction. If the two hash values ​​are the same, it determines that the monitoring data on which the corrected asset value assessment result is based matches the latest monitoring data notarized in the blockchain network. After confirming the match, the consensus node places the evaluation record transaction into a pending transaction pool in its memory. The pending transaction pool is an ordered queue, and the consensus node sorts the evaluation record transactions in the pending transaction pool according to their Gas price. The consensus node extracts transactions from the pending transaction pool according to the block packaging cycle, which is set to 10 seconds. The 10-second cycle is designed to balance block generation latency and network load, ensuring sufficient transaction accumulation within 10 seconds to improve block utilization. The number of extracted transactions does not exceed 5000, and the total size does not exceed 16 megabytes.Consensus nodes construct a new block by combining the extracted evaluation record transactions with other transactions to form a transaction list for the new block. A Merkle tree is then built on this transaction list, generating the Merkle tree root. The block header contains the hash value of the previous block, the Merkle tree root, the current timestamp, and the block height. Consensus nodes execute a practical Byzantine fault-tolerant consensus algorithm. During the pre-preparation phase, the master node broadcasts a pre-preparation message. After more than two-thirds of the nodes have received the pre-preparation message during the preparation phase, a commit message is broadcast. During the commit phase, after more than two-thirds of the nodes have received the commit message, consensus is reached, and the new block is appended to the end of the blockchain, completing the packaging of the evaluation record transactions into the block and writing back the evaluation results.

[0072] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A blockchain-based method for assessing the value of aquatic biological assets in aquaculture, characterized in that, include: In response to the request to upload the monitoring data of the aquaculture pond to the blockchain, the monitoring data is stored on the blockchain network, and the smart contract is triggered to perform the initial value anchoring operation. Based on the result of the initial value anchoring operation, the multi-source monitoring data stored on the chain is called to perform consistency verification on the growth stage of the aquatic organisms and obtain the verified growth status data. Based on the verified growth status data and combined with on-chain historical transaction records, the dynamic value adjustment factor of the current aquatic biological assets is determined. The result of the initial value anchoring operation is corrected using the dynamic value adjustment factor to generate a corrected asset value assessment result, which is then written back to the blockchain network.

2. The method for evaluating the value of aquatic biological assets based on blockchain according to claim 1, characterized in that, In response to the request to upload the monitoring data from the aquaculture pond to the blockchain, the monitoring data is stored on the blockchain network, and a smart contract is triggered to perform an initial value anchoring operation, including: Parse the monitoring data packet carried in the on-chain request, and extract the water quality parameter sequence and biomass image sequence from the monitoring data packet; The water quality parameter sequence is subjected to outlier removal processing to obtain a cleaned water quality parameter sequence, and the biomass image sequence is normalized to obtain a normalized image sequence. The cleaned water quality parameter sequence and the standardized image sequence are packaged into a data block for evidence storage, the hash value of the data block is calculated, and the hash value is associated with the data block and stored in the blockchain network. After the evidence data block is successfully uploaded to the blockchain, the initial value anchoring function in the pre-deployed smart contract is called, and the evidence data block is used as the input parameter to perform the initial value anchoring operation.

3. The method for evaluating the value of aquatic biological assets based on blockchain according to claim 2, characterized in that, Based on the result of the initial value anchoring operation, the multi-source monitoring data stored on the blockchain is invoked to perform consistency verification on the growth stages of the aquatic organisms, obtaining verified growth status data, including: Obtain the growth stage identifier contained in the result of the initial value anchoring operation, and query the corresponding standard growth curve template from the blockchain network based on the growth stage identifier; The water quality parameter sequence and biomass image sequence associated with the current aquaculture pond at multiple consecutive time points are retrieved from the blockchain network as the multi-source monitoring data; The biomass image sequence from the multi-source monitoring data is input into a preset growth stage discrimination model to obtain a growth stage identifier, and the discriminant growth stage identifier is compared with the growth stage identifier. When the growth stage identifier matches the growth stage identifier, the multi-source monitoring data is marked as growth status data that has passed the verification. When the growth stage identifier is inconsistent with the growth stage identifier, the result of the initial value anchoring operation is corrected according to the growth stage identifier, and the corrected growth stage identifier is associated with the multi-source monitoring data and stored as verified growth status data.

4. The method for evaluating the value of aquatic biological assets based on blockchain according to claim 3, characterized in that, The dynamic value adjustment factor for the current aquatic biological assets is determined based on the verified growth status data and combined with on-chain historical transaction records, including: Extract the current biomass density parameter and the current average body length parameter from the verified growth status data; Based on the current biomass density parameter and the current average body length parameter, query the set of historical transaction records stored on the chain corresponding to the same type of farmed aquatic organisms, and filter out recent transaction records whose transaction time is less than a preset time threshold from the set of historical transaction records; Calculate the unit value parameter corresponding to each transaction in the recent transaction records, and perform statistical processing on the unit value parameter to obtain the market reference value range; The dynamic value adjustment factor is calculated based on the current biomass density parameter, the current average body length parameter, and the market reference value range.

5. The method for evaluating the value of aquatic biological assets based on blockchain according to claim 4, characterized in that, The step of using the dynamic value adjustment factor to correct the result of the initial value anchoring operation, generating a corrected asset value assessment result, and writing the assessment result back to the blockchain network includes: Read the underlying asset value from the result of the initial value anchoring operation; The initial adjusted value is obtained by multiplying the value of the underlying asset with the dynamic value adjustment factor. The initial corrected value is subjected to normalization constraint processing to make the initial corrected value fall within a preset value fluctuation range, thus obtaining the normalized corrected value. The normalized adjusted value is used as the adjusted asset valuation result, and an valuation record transaction containing the adjusted asset valuation result and the current timestamp is generated. The evaluation record transaction is broadcast to the consensus nodes in the blockchain network. The consensus nodes verify the evaluation record transaction and package it into a block to complete the evaluation result write-back.

6. The method for evaluating the value of aquatic biological assets based on blockchain according to claim 2, characterized in that, The step of parsing the monitoring data packet carried in the on-chain request and extracting the water quality parameter sequence and biomass image sequence from the monitoring data packet includes: The monitoring data packets are deserialized to obtain the data packet header and data packet body; Read the data type identifier from the header of the data packet, and locate the water quality parameter data area and the biomass image data area from the body of the data packet based on the data type identifier; Water quality parameter values ​​are read sequentially from the water quality parameter data area at preset time intervals to generate the water quality parameter sequence. Multiple biomass images are read from the biomass image data area and arranged in chronological order of acquisition time to generate the biomass image sequence.

7. The method for evaluating the value of aquatic biological assets based on blockchain according to claim 2, characterized in that, The process of removing outliers from the water quality parameter sequence to obtain a cleaned water quality parameter sequence, and normalizing the biomass image sequence to obtain a normalized image sequence, includes: Calculate the deviation between each water quality parameter value in the water quality parameter sequence and the mean of the water quality parameter sequence, and mark the corresponding water quality parameter value with a deviation value greater than a preset deviation threshold as an outlier; Delete the water quality parameter values ​​marked as outliers in the water quality parameter sequence, and fill the deleted positions with water quality parameter values ​​using linear interpolation to obtain the cleaned water quality parameter sequence; A size normalization operation is performed on each frame of biomass image in the biomass image sequence to make the pixel size of each frame of biomass image uniform to a preset standard size; Pixel value normalization is performed on each frame of biomass image after size normalization, mapping the pixel values ​​to a preset numerical range to obtain a normalized image sequence.

8. The method for evaluating the value of aquatic biological assets based on blockchain according to claim 3, characterized in that, The step of inputting the biomass image sequence from the multi-source monitoring data into a preset growth stage discrimination model to obtain a growth stage identifier, and comparing the discriminant growth stage identifier with the growth stage identifier, includes: The biomass image sequence is divided into multiple image subsequences according to the acquisition time sequence, and each image subsequence contains a number of consecutive frames of biomass images; Each image subsequence is input into the growth stage discrimination model, which performs feature extraction processing on each image subsequence and outputs the sub-stage discrimination result corresponding to each image subsequence. A voting and statistical process is performed on the multiple sub-stage discrimination results, and the sub-stage discrimination result with the highest number of votes is determined as the discrimination growth stage identifier; The growth stage identifier is compared with the growth stage identifier to determine whether they are consistent.

9. The method for evaluating the value of aquatic biological assets based on blockchain according to claim 4, characterized in that, The calculation of the unit value parameter corresponding to each transaction in the recent transaction records, and the statistical processing of the unit value parameter to obtain the market reference value range, includes: For each transaction in the recent transaction records, read the total transaction amount and the total number of transaction organisms, and divide the total transaction amount by the total number of transaction organisms to obtain the unit value parameter of the transaction. Arrange all the unit value parameters in ascending order to obtain the unit value parameter sequence; The first quantile and the third quantile are extracted from the unit value parameter sequence, and the numerical range between the first quantile and the third quantile is determined as the market reference value range.

10. A method for evaluating the value of aquatic biological assets based on blockchain according to claim 5, characterized in that, The step of broadcasting the evaluation record transaction to consensus nodes in the blockchain network, and having the consensus nodes verify the evaluation record transaction and package it into a block, includes: The evaluation record transaction is digitally signed to verify whether the sender of the evaluation record transaction has evaluation authority. When the sender has the assessment authority, verify whether the revised asset value assessment result in the assessment record transaction matches the latest monitoring data stored in the blockchain network; When the revised asset valuation result matches the latest monitoring data, the valuation record transaction is placed into the transaction pool to be packaged in the consensus node; The consensus node extracts transaction transactions from the transaction pool to be packaged according to the block packaging cycle, and after executing the consensus algorithm, appends a new block containing the evaluation record transaction to the end of the blockchain.

Images