Unmanned feeding ship multi-source aquaculture data low-latency anti-interference communication method and system

Abstract

The present application relates to the technical field of aquaculture, in particular to a multi-source aquaculture data low-latency anti-interference communication method and system for unmanned feeding ship, comprising the following steps: multi-source aquaculture data acquisition; edge data fusion and instruction layered coding encapsulation, real-time fusion processing through a lightweight neural network at the unmanned feeding ship end to generate decision instructions corresponding to the aquaculture event; multi-node anti-interference networking; data transmission scheduling; data reception and state feedback. By deploying a MobileNet-SSD lightweight neural network at the unmanned feeding ship end, the present application realizes edge real-time cross-modal fusion of aquaculture water environment data and feeding state data, combines layered coding design driven by aquaculture events and channel quality change rate adaptive transmission strategy, avoids data redundancy, guarantees complete transmission of key instruction information when channel quality deteriorates, and effectively solves the problem of insufficient data fusion and transmission protocol adaptability.

Description

Technical Field

[0001] This invention relates to the field of aquaculture technology, and more specifically, to a method and system for low-latency, interference-resistant communication of multi-source aquaculture data from unmanned feeding vessels. Background Technology

[0002] In the process of intelligent upgrading of aquaculture, unmanned feeding vessels, as the core execution equipment for precision feeding, need to simultaneously acquire environmental data of the aquaculture waters (such as water quality, water temperature, dissolved oxygen, etc.) and feeding status data of the farmed organisms. Data transmission to the shore-based control center and feedback of decision-making instructions are achieved through a multi-node communication network. The low latency and anti-interference capability of this communication directly determine the feeding accuracy, the utilization rate of aquaculture resources, and the survival rate of the farmed organisms. However, aquaculture waters exhibit complex characteristics such as severe electromagnetic signal attenuation, dynamic fluctuations in channel quality, and dispersed distribution of sensing nodes. This makes the efficient collaborative transmission of multi-source heterogeneous aquaculture data and stable communication in a dynamic network environment the core challenges for technology implementation.

[0003] In existing technologies, the transmission of aquaculture data often adopts a "distributed collection-centralized transmission" model. This involves collecting single-type aquaculture data through distributed sensing nodes, processing it with a fixed encoding format, and then directly uploading it to the cloud or control center. Data fusion processing largely relies on cloud computing power. In terms of communication networking, static networks are mostly built based on fixed relay nodes, using single-channel transmission or simple channel switching strategies. Some technologies introduce basic link redundancy design to improve stability. Data transmission scheduling mostly adopts the conventional "first-come, first-served" mechanism without differentiating command priorities. However, these technologies have significant shortcomings: First, insufficient data fusion and transmission protocol adaptability: real-time edge fusion of aquaculture water environment data and feeding status data is not achieved; data transmission lacks layered adaptation design and cannot dynamically adjust transmission content according to channel quality, resulting in excessive data redundancy or missing key information; transmission latency is insufficient to meet the real-time requirements of precise feeding. Second, inadequate network mechanism and mobile scenario adaptability: the network method is not adapted to the mobile characteristics of unmanned feeding vessels; the static network architecture and fixed channel switching strategy lack flexibility, making it difficult to quickly respond to dynamic changes in link quality in aquaculture waters, and limiting anti-interference capabilities. Third, weak transmission reliability and closed-loop management capabilities: the lack of an instruction priority scheduling mechanism based on event urgency allows critical aquaculture safety-related instructions to easily consume transmission resources due to routine data; and single-link or simple redundancy designs cannot guarantee communication reliability in extreme scenarios, failing to form closed-loop management. Therefore, we propose a low-latency, anti-interference communication method and system for multi-source aquaculture data from unmanned feeding vessels. Summary of the Invention

[0004] The purpose of this invention is to provide a low-latency, anti-interference communication method and system for multi-source aquaculture data from unmanned feeding vessels, in order to solve the problems mentioned in the background art, such as insufficient data fusion and transmission protocol adaptability, lack of adaptability of networking mechanisms and mobile scenarios, and weak transmission reliability and closed-loop management capabilities.

[0005] To address the aforementioned technical problems, one objective of this invention is to provide a low-latency, interference-resistant communication method for multi-source aquaculture data from unmanned feeding vessels, comprising the following steps:

[0006] S1. Multi-source aquaculture data collection: Real-time collection of aquatic environmental data and feeding status data is achieved through sensing node units at different monitoring points in the water area and sensing units carried by unmanned feeding boats; each sensing node unit is equipped with a unique identification and positioning function.

[0007] S2. Edge Data Fusion and Layered Encoding and Encapsulation of Instructions: Based on the environmental data and feeding status data collected in S1, a lightweight neural network is used for real-time fusion processing at the unmanned feeding vessel to generate decision instructions for corresponding aquaculture events. The decision instructions adopt a layered encoding method driven by aquaculture events: the basic layer contains information on the type of aquaculture event and the amount of feed, and the enhancement layer contains the event basis of the decision instructions. Communication channel parameters are continuously collected for a preset number of periods, and the channel quality change rate is calculated. If the channel quality change rate exceeds a preset trigger threshold, it is determined that the channel quality is deteriorating, and only the content of the basic layer is transmitted. If the channel quality change rate does not exceed the preset trigger threshold, it is determined that the channel quality is stable, and the joint content of the basic layer and the enhancement layer is transmitted. The encoded instructions are then encapsulated according to a preset frame structure.

[0008] S3. Multi-node anti-interference networking: Using the unmanned feeding boat as a mobile communication relay, a dynamic network is built with each sensing node unit to select effective communication links and autonomously switch to the optimal channel.

[0009] S4. Data transmission scheduling: The encoded instruction content is transmitted in time-sharing according to the priority of the decision instruction, and forwarded by the unmanned feeding vessel to the shore-based control center through a dual-mode redundant link;

[0010] S5. Data reception and status feedback: After receiving data, the shore-based control center verifies the integrity of the command and initiates retransmission as needed; each node provides real-time feedback on its own status, forming a communication closed loop.

[0011] As a further improvement to this technical solution, in step S1, the multi-source aquaculture data acquisition includes the following steps:

[0012] S11. Deploy the sensing node units at different monitoring points in the aquaculture water area at preset intervals, assign a unique identification code to each sensing node unit, and enable the positioning function of the sensing node unit to obtain the real-time location information of the sensing node unit.

[0013] S12, the sensing node unit controlling the water monitoring points and the sensing unit carried by the unmanned feeding boat synchronously start the data acquisition operation according to the preset time cycle;

[0014] S13. Store the aquatic environment data collected by the sensing node unit and the feeding status data collected by the unmanned feeding boat sensing unit to the local cache module of the corresponding acquisition terminal.

[0015] As a further improvement to this technical solution, in step S2, a lightweight neural network is used for real-time fusion processing to generate decision instructions for the corresponding aquaculture event, including the following steps:

[0016] S211. Select MobileNet-SSD as a lightweight neural network and deploy it on the edge computing unit of the unmanned feeding boat, and configure the network inference frame rate and input resolution to adapt to real-time data processing.

[0017] S212. Retrieve the aquatic environment data and feeding status data collected in S1, perform normalization processing on the environmental data, and perform standardization processing on the feeding status data.

[0018] S213. Input the preprocessed data into the neural network according to the dimension of "environmental feature group - feeding feature group", extract single-modal features through the DepthwiseConv layer, complete cross-modal feature fusion through the Bottleneck layer, and then output preliminary decision information containing the category code of breeding event and the numerical code of feeding amount through the classifier.

[0019] S214. Based on the single feeding amount of the unmanned feeding vessel and the upper limit of the average daily feed intake of the cultured species (the upper limit of the average daily feed intake of the cultured species is a quantitative indicator of the physiological characteristics of the cultured species), a double validity check is performed. After removing abnormal data, a standard decision instruction that conforms to the preset format is generated.

[0020] As a further improvement to this technical solution, in step S2, the decision instruction adopts a hierarchical encoding method driven by aquaculture events, including the following steps:

[0021] S215. Divide aquaculture events into multiple core types, assign a unique binary code to each type, and form a fixed mapping between the code and the event type;

[0022] S216. Extract the aquaculture event type and feeding amount information from the decision instruction, use the binary code of S215 as the event type field, set the feeding amount quantification unit based on the accuracy of the unmanned feeding boat feeding equipment, generate the feeding amount code field, and splice them in the order of "event type code - feeding amount code" to form the basic layer.

[0023] S217. Taking the event basis of the decision-making instruction as the core, extract the normalized aquaculture water environment data and the quantitative feeding status data, generate summaries for the normalized aquaculture water environment data and the quantitative feeding status data respectively, and splice them in the order of "environmental data summary - feeding status data summary" to form an enhancement layer.

[0024] S218. Using a preset end-order encoding rule, the encoding module of the edge computing unit encodes the content of the base layer and the enhancement layer separately. The total length of the encoded instruction is as follows: Characterization, when transmitting the base layer alone During joint transmission ,in The length of the encoded base layer data. This represents the length of the encoded enhancement layer data.

[0025] As a further improvement to this technical solution, in step S2, communication channel parameters are continuously collected for a preset number of periods and the channel quality change rate is calculated to adapt to the transmission content, including the following steps:

[0026] S219. Through the communication unit carried by the unmanned feeding boat, the channel core parameters, including signal-to-noise ratio, bit error rate, and transmission delay, are continuously collected for a preset number of cycles, and the collection cycle is synchronized with the data collection cycle of S1.

[0027] S220. Based on the collected core channel parameters, a weighted algorithm is used to calculate the overall channel quality value for each cycle. , This is the normalized result after weighting the parameters;

[0028] S221. Channel quality composite value based on continuous acquisition period The channel quality change rate was calculated. ;

[0029] S222, Preset negative rate of change trigger threshold When the channel quality change rate When the channel quality is determined to be deteriorating, only the basic layer coded content is selected as the transmitted data; when If the channel quality remains stable, the jointly coded content of the base layer and enhancement layer is selected as the transmission data.

[0030] As a further improvement to this technical solution, in step S2, the encoded instructions are encapsulated according to a preset frame structure, including the following steps:

[0031] S223. The frame structure of the preset instruction encapsulation is "header identifier - node information - hierarchical instruction segment - timestamp - check code". Each field is seamlessly spliced ​​in order and stored in the edge computing unit of the unmanned feeding vessel.

[0032] S224. The header identifier is set to a preset binary fixed code, which is used by the shore-based control center to identify the command type; the node information includes the identification code of the sensing node unit and a summary of GPS positioning coordinates.

[0033] S225. Based on the transmission mode determined by S222, fill in the basic layer content after S218 encoding, or the joint encoding content of the basic layer and the enhancement layer. The timestamp adopts a preset format that is compatible with the data acquisition cycle of S1 and is accurate to a preset time unit.

[0034] S226. Generate a checksum using a combined CRC-32 and MD5 algorithm. First, perform a CRC-32 operation on the header identifier, node information, hierarchical instruction segment, and timestamp fields. Then, perform an MD5 operation on the CRC-32 operation result together with the header identifier, node information, hierarchical instruction segment, and timestamp fields. Finally, use the combined operation result as the checksum and append it to the end of the frame structure to complete the instruction encapsulation.

[0035] As a further improvement to this technical solution, in step S3, the multi-node anti-interference networking includes the following steps:

[0036] S31. The unmanned feeding vessel periodically sends link detection signals through the communication unit. Each sensing node unit responds with a response signal based on its own unique identifier. The unmanned feeding vessel records the signal arrival time, signal strength and link connectivity status of each sensing node unit and establishes a node link information table.

[0037] S32. Using link stability, transmission delay, and packet loss rate as the core screening indicators, quantitatively evaluate the detected communication links, eliminate invalid links with signal strength below the preset threshold or packet loss rate exceeding the preset range, retain valid links that meet data transmission requirements, and update the node link information table.

[0038] S33. For the candidate channels corresponding to the effective links, combine the core channel parameters collected in S2 to calculate the comprehensive communication quality score of each candidate channel, and determine the channel with the highest score and that meets the low latency transmission standard as the optimal channel.

[0039] S34. Monitor the quality status of the current communication link in real time. When the comprehensive link quality score is lower than the preset ratio of the optimal channel score, or when an abnormal situation occurs, trigger the autonomous switching mechanism and quickly switch to the optimal channel based on the node link information table. At the same time, repeat steps S31-S33 periodically to update the link and channel information and maintain the anti-interference stability of the dynamic network.

[0040] As a further improvement to this technical solution, in step S4, the data transmission scheduling includes the following steps:

[0041] S41. Based on the aquaculture event types generated by S2, set multiple priorities according to the urgency of the events. The decision instructions corresponding to events involving the survival safety of the aquaculture objects are classified as the highest priority, and the decision instructions corresponding to regular feeding events are classified as the basic priority. Establish a mapping relationship between priority and instruction type and store it in the edge computing unit of the unmanned feeding vessel.

[0042] S42. According to the priority mapping relationship, independent transmission time slots are allocated to instructions of different priorities. The highest priority instructions occupy the priority transmission time slot, and the basic priority instructions occupy the remaining time slots in a time-sharing manner according to a preset order. The time slot length is adapted to the amount of instruction content data after S2 encoding.

[0043] S43. The unmanned feeding vessel activates the preset dual-mode communication link, binds the main link to the optimal channel determined by S3, associates the backup link with the redundant channel, and configures the link switching trigger conditions.

[0044] S44. The unmanned feeding vessel forwards the coded instructions to the shore-based control center through the main link according to the time-sharing scheduling rules; it monitors the transmission status of the main link in real time, and immediately switches to the backup link to continue transmission when the main link status triggers the switching condition.

[0045] As a further improvement to this technical solution, in step S5, data reception and status feedback includes the following steps:

[0046] S51. The shore-based control center receives the coded instructions forwarded by the unmanned feeding vessel, extracts the checksum from the instructions, recalculates the received header identifier, node information, hierarchical instruction segment, and timestamp field using the same algorithm, and compares it with the extracted checksum.

[0047] S52. If the verification results are consistent, parse the instruction content; if the verification results are inconsistent, the shore-based control center initiates a retransmission request to the unmanned feeding vessel through a dual-mode redundant link, clarifying the identifier and transmission link of the retransmission instruction.

[0048] S53. Each sensing node unit and the unmanned feeding vessel shall provide real-time feedback on their own operating status information according to a preset cycle, including power supply status, communication link quality, data acquisition and processing status, and the feedback information shall be associated with their own unique identity identifier.

[0049] S54. The shore-based control center receives status feedback information from all sensing node units, establishes and updates the sensing node unit status table in real time, triggers an alarm mechanism for abnormal states, and stores the command parsing results in association with the sensing node unit status information.

[0050] The second objective of this invention is to provide a low-latency, interference-resistant communication system for multi-source aquaculture data from unmanned feeding vessels, used to execute the aforementioned low-latency, interference-resistant communication method for multi-source aquaculture data from unmanned feeding vessels, comprising:

[0051] The sensing node unit is deployed at different monitoring points in the aquaculture water area at a preset interval. Each sensing node unit is equipped with a unique identification code and positioning function, and has a built-in local cache module for collecting aquaculture water area environmental data in real time at a preset time period and storing it in the local cache module. It can also feed back its own operating status information at a preset period, and the feedback information is associated with its own unique identification.

[0052] The sensing unit is mounted on the unmanned feeding vessel and has a built-in local cache module. It is used to synchronously start the feeding status data collection with the sensing node unit at a preset time period, store the collected data in the local cache module, and can provide feedback on its own data collection and processing status.

[0053] An edge computing unit, deployed on an unmanned feeding vessel, is used for real-time fusion processing of multi-source aquaculture data, generating standard decision instructions, executing hierarchical coding and instruction encapsulation, and completing channel quality assessment and related data storage. It includes a lightweight neural network module, an encoding module, and a data processing module, wherein:

[0054] The lightweight neural network module is configured with an inference frame rate and input resolution adapted to real-time data processing. It is used to retrieve aquatic environmental data collected by the perception node unit and feeding status data collected by the sensor unit. It performs normalization processing on the environmental data and standardization processing on the feeding status data. It extracts single-modal features through the DepthwiseConv layer and completes cross-modal feature fusion through the Bottleneck layer. Then, it outputs preliminary decision information containing aquaculture event category code and feeding amount numerical code through the classifier. After dual validity verification of the single feeding amount of the unmanned feeding boat and the upper limit of the average daily feeding amount of the aquatic object, it generates a standard decision instruction that conforms to the preset format.

[0055] The encoding module adopts a layered encoding method driven by aquaculture events and a preset end-order encoding rule to encode the content of the basic layer and the enhancement layer respectively. It also completes the instruction encapsulation according to the preset frame structure of "header identifier - node information - layered instruction segment - timestamp - check code". The check code is generated by the CRC-32 and MD5 joint algorithm.

[0056] The data processing module is used to store the node link information table and the mapping relationship between priority and instruction type, and to calculate the comprehensive channel quality value and rate of change.

[0057] The communication unit, mounted on the unmanned feeding vessel, acts as a mobile communication relay. It periodically sends link detection signals, records the signal arrival time, signal strength, and link connectivity status of each sensing node, and establishes a node link information table. It quantitatively evaluates communication links using link stability, transmission delay, and packet loss rate as core screening indicators, eliminates invalid links, and updates the node link information table. It calculates a comprehensive communication quality score for each candidate channel based on core channel parameters, selects the optimal channel, monitors link quality in real time, and triggers an autonomous switching mechanism to maintain network anti-interference stability. Simultaneously, it enables dual-mode communication links, binding the primary link to the optimal channel and associating the backup link with redundant channels. It forwards encoded instructions according to time-sharing scheduling rules, monitors link transmission status in real time, and triggers switching.

[0058] The shore-based control center receives coded instructions forwarded by the unmanned feeding vessel through its communication unit, extracts the checksum from the instructions, and recalculates the received header identifier, node information, hierarchical instruction segment, and timestamp field using a combined CRC-32 and MD5 algorithm, comparing it with the checksum. If the checksum matches, the control center parses the instruction content; if the checksum does not match, it initiates a retransmission request to the unmanned feeding vessel through a dual-mode redundant link, specifying the identifier of the retransmission instruction and the transmission link. Simultaneously, the control center receives status feedback information from the sensing node units and the unmanned feeding vessel, establishes and updates the sensing node unit status table in real time, triggers an alarm mechanism for abnormal states, and associates and stores the instruction parsing results with the sensing node unit status information.

[0059] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0060] 1. This invention achieves real-time cross-modal fusion of aquaculture water environment data and feeding status data at the edge by deploying a lightweight MobileNet-SSD neural network on the unmanned feeding vessel. Combined with a hierarchical coding design driven by aquaculture events and a channel quality change rate adapted transmission strategy, it avoids data redundancy and ensures the complete transmission of key command information when the channel quality deteriorates. It effectively solves the problem of insufficient data fusion and transmission protocol adaptability and meets the real-time requirements of precise feeding.

[0061] 2. This invention constructs a dynamic network using unmanned feeding boats as mobile communication relays. It filters effective links through multiple indicators such as link stability, transmission latency, and packet loss rate, and determines the optimal channel by combining the comprehensive channel quality score and realizes dynamic switching. It adapts to the mobile scenario characteristics of aquaculture waters, solves the problem of insufficient adaptability of networking mechanism to mobile scenario, and improves the anti-interference stability of multi-node communication.

[0062] 3. This invention prioritizes instructions based on the urgency of aquaculture events and allocates independent transmission time slots. It also configures dual-mode redundant links and a link switching mechanism to ensure priority transmission of critical aquaculture safety-related instructions, improving data transmission reliability in extreme scenarios. Through CRC-32 and MD5 joint verification, on-demand retransmission, and real-time status feedback mechanisms for each node, a complete communication closed loop is formed, solving the problems of weak transmission reliability and closed-loop management capabilities, and realizing full-process control of aquaculture data transmission and node status.

[0063] 4. This invention further ensures the rationality of decision-making instructions by synchronously collecting data from multiple sources, storing it locally, and combining edge data preprocessing and dual validity verification. Combined with standardized instruction frame structure encapsulation, it ensures the standardization and integrity of data transmission, and fully supports the precise communication needs in intelligent aquaculture scenarios. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0066] like Figure 1 As shown, this embodiment provides a low-latency, interference-resistant communication method for multi-source aquaculture data from unmanned feeding vessels, including:

[0067] S1. Multi-source aquaculture data collection: Real-time collection of aquatic environmental data and feeding status data is achieved through sensing node units at different monitoring points in the water area and sensing units carried by unmanned feeding boats; each sensing node unit is equipped with a unique identification and positioning function.

[0068] In this step, the multi-source aquaculture data collection in S1 includes the following steps:

[0069] S11. Deploy the sensing node units at different monitoring points in the aquaculture water area at preset intervals, assign a unique identification code to each sensing node unit, and enable the positioning function of the sensing node unit to obtain the real-time location information of the sensing node unit.

[0070] Specifically, based on the actual area of ​​the aquaculture waters, the uniformity of water quality, and the required monitoring accuracy, sensing node units are evenly deployed at preset intervals of 5-20 meters at different monitoring points within the aquaculture waters, focusing on covering feeding areas, water edge areas, and areas prone to water quality fluctuations to ensure no monitoring blind spots. Each sensing node unit is assigned a unique 16-bit binary identifier code, with the coding rule set as follows: "the first 8 bits are the area number (corresponding to the zoning plan of the aquaculture waters), and the last 8 bits are the node sequence number (corresponding to the node deployment order within the zoning)." The code corresponds one-to-one with the physical deployment location of the sensing node unit, facilitating rapid identification of the data source by the shore-based control center. Simultaneously, the built-in GPS positioning module of the sensing node unit is activated, synchronously recording the latitude and longitude coordinates of the node after each data collection cycle. The positioning data is stored in association with subsequently collected environmental data, and the positioning accuracy meets the conventional application requirements for aquaculture monitoring, ensuring that the shore-based control center can trace the spatial source of the data.

[0071] S12, the sensing node unit controlling the water monitoring points and the sensing unit carried by the unmanned feeding boat synchronously start the data acquisition operation according to the preset time cycle;

[0072] Specifically, combining the feeding activity patterns of the aquaculture species (such as the peak feeding times in the morning and evening) and the dynamic change frequency of aquaculture water environment data (such as water temperature and dissolved oxygen), the preset time period for data collection is set to 1-5 minutes. The period length can be dynamically adjusted by the shore-based control center according to the aquaculture scenario. Synchronous data collection is triggered using a "unified timestamp command" mechanism: the shore-based control center sends a collection start command with a precise timestamp to all sensing node units and the sensing units of the unmanned feeding vessel according to a preset cycle. After receiving the command, each collection terminal synchronizes based on its local clock and starts the data collection operation at the same time. The sensing node units focus on collecting aquaculture water environment data (such as water temperature, pH value, and dissolved oxygen content), while the sensing units on the unmanned feeding vessel focus on collecting data on the feeding status of the aquaculture species (such as feeding density and remaining feed amount), ensuring the consistency of the two types of data in the time dimension and providing a foundation for subsequent fusion processing.

[0073] S13. Store the aquatic environment data collected by the sensing node unit and the feeding status data collected by the unmanned feeding boat sensing unit to the local cache module of the corresponding acquisition terminal.

[0074] Specifically, both the sensing node unit and the sensing unit of the unmanned feeding vessel are equipped with industrial-grade flash memory modules as local cache modules. The cache capacity of the sensing node unit is adapted to the data collection and storage requirements for 24 consecutive hours, while the cache unit capacity of the unmanned feeding vessel is adapted to the data collection and storage requirements for a single full-water cruise, thus avoiding data loss.

[0075] Meanwhile, the collected data is stored in a standardized binary format, seamlessly concatenated and stored according to the field order of "collection timestamp - node identification code - data type code - original data value": the environmental data stored by the sensing node unit is associated with its own identification and GPS positioning coordinates, and the feeding status data stored by the unmanned feeding vessel is associated with the shipborne sensing unit number and real-time ship position information. Simultaneously, the cache unit is configured with a cyclic overwrite and early warning mechanism: when the cache capacity reaches 90%, it automatically overwrites the oldest stored non-critical historical data according to the "first-in, first-out" principle; when the cache capacity reaches 95%, a low cache warning signal is triggered and uploaded to the shore-based control center, reminding staff to synchronize data in a timely manner to ensure the continuity and security of data storage.

[0076] S2. Edge Data Fusion and Layered Encoding and Encapsulation of Instructions: Based on the environmental data and feeding status data collected in S1, a lightweight neural network is used for real-time fusion processing at the unmanned feeding vessel to generate decision instructions for corresponding aquaculture events. The decision instructions adopt a layered encoding method driven by aquaculture events: the basic layer contains information on the type of aquaculture event and the amount of feed, and the enhancement layer contains the event basis of the decision instructions. Communication channel parameters are continuously collected for a preset number of periods, and the channel quality change rate is calculated. If the channel quality change rate exceeds a preset trigger threshold, it is determined that the channel quality is deteriorating, and only the content of the basic layer is transmitted. If the channel quality change rate does not exceed the preset trigger threshold, it is determined that the channel quality is stable, and the joint content of the basic layer and the enhancement layer is transmitted. The encoded instructions are then encapsulated according to a preset frame structure.

[0077] In this step, in S2, a lightweight neural network is used to perform real-time fusion processing to generate decision instructions for the corresponding aquaculture event, including the following steps:

[0078] S211. Select MobileNet-SSD as a lightweight neural network and deploy it on the edge computing unit of the unmanned feeding boat, and configure the network inference frame rate and input resolution to adapt to real-time data processing.

[0079] Specifically, the unmanned feeding vessel is equipped with an industrial-grade edge computing unit as the carrier for the neural network. The MobileNet-SSD model is compiled into an inference engine adapted to the edge computing unit using TensorRT. The network inference frame rate is configured at 15fps—this frame rate matches the 1-5 minute data acquisition cycle set by S1, ensuring that each round of data acquisition can be processed in real time without data backlog. The input resolution is set to 320×320 pixels. This parameter balances feature extraction accuracy and inference speed: compared to higher resolutions (such as 480×480), it can reduce inference time by about 40%, while compared to lower resolutions (such as 224×224), it can retain more detailed features of feeding status. The network weights are finely tuned based on a real-world dataset of aquaculture scenarios (containing 50,000 sets of environment-feeding matching data), ensuring that the model adapts to the feature distribution of aquaculture data and improving decision accuracy.

[0080] S212. Retrieve the aquatic environment data and feeding status data collected in S1, perform normalization processing on the environmental data, and perform standardization processing on the feeding status data.

[0081] Specifically, to eliminate the magnitude differences and distribution shifts between different types of data and ensure the consistency of neural network input, targeted preprocessing was performed on environmental data and feeding status data respectively. The specific operations are as follows:

[0082] First, environmental data normalization is performed:

[0083] The min-max normalization method is used to map the original environmental data to the [0,1] interval. The calculation formula is as follows:

[0084] ;

[0085] in, This represents the normalized environmental data (the output range is strictly controlled within [0,1]).

[0086] This represents the raw values ​​of environmental data (such as the collected values ​​of a single environmental parameter, such as water temperature, pH value, dissolved oxygen content, ammonia nitrogen concentration, nitrite concentration, and transparency). This represents the normal minimum value of the environmental parameter (based on historical monitoring data of the aquaculture water area over the past year and the aquaculture industry standard). This indicates the normal maximum value of the environmental parameter (based on historical monitoring data of the aquaculture water area over the past year and the aquaculture industry standard).

[0087] Standardization of feeding status data:

[0088] The Z-score standardization method is used to transform the data into a standard normal distribution with a mean of 0 and a variance of 1. The calculation formula is as follows:

[0089] ;

[0090] in, This represents the standardized feeding status data (output mean approaches 0, variance approaches 1); This represents the raw values ​​of feeding status data (such as the collected values ​​of a single feeding parameter, such as feeding density, feeding activity, amount of remaining feed, and feeding duration). This represents the historical average of the feeding status parameter (calculated based on valid data collected over the past 30 days, with outliers removed and the arithmetic mean taken). This represents the standard deviation of historical data for this feeding status parameter (calculated based on valid data collected over the past 30 days, reflecting the degree of data dispersion).

[0091] S213. Input the preprocessed data into the neural network according to the dimension of "environmental feature group - feeding feature group", extract single-modal features through the DepthwiseConv layer, complete cross-modal feature fusion through the Bottleneck layer, and then output preliminary decision information containing the category code of breeding event and the numerical code of feeding amount through the classifier.

[0092] Specifically, the preprocessed environmental data and feeding status data are combined according to a fixed dimension and input into the neural network. Feature fusion and decision output are completed through hierarchical operations. The specific operation is as follows:

[0093] Feature group dimension definition: The environmental feature group consists of 6 parameters: water temperature, pH value, dissolved oxygen content, ammonia nitrogen concentration, nitrite concentration, and transparency. Each parameter corresponds to a 2-dimensional feature (original value + change trend), for a total of 12-dimensional feature vectors. The feeding feature group consists of 4 parameters: feeding density, feeding activity, amount of remaining food, and feeding duration. Each parameter corresponds to a 2-dimensional feature (original value + change rate), for a total of 8-dimensional feature vectors. The two types of feature groups are concatenated in the order of "environmental feature group first, feeding feature group second" to form a 20-dimensional input vector.

[0094] The specific operational logic of the neural network layer is as follows:

[0095] DepthwiseConv layer: Uses 3×3 convolution kernels, with a total of 20 kernels (consistent with the dimension of the input vector), stride set to 1, padding mode set to "same", and performs local feature extraction on environmental feature group and feeding feature group respectively. This layer only performs convolution on a single channel, which can reduce the amount of computation by 75% compared to traditional convolution, making it suitable for edge computing power.

[0096] Bottleneck layer: Set the compression factor to 0.5. First, use 1×1 convolution to compress the single-modal feature dimension to 50% of the original dimension (environment feature group is compressed to 6 dimensions, feeding feature group is compressed to 4 dimensions). Then, use matrix multiplication to complete the cross-modal feature association operation. Finally, use 1×1 convolution to restore the fused feature dimension to 20 dimensions, realizing efficient cross-modal fusion of "compression-fusion-expansion".

[0097] Classifier and Regression Branch: The classifier consists of two fully connected layers (the first layer outputs 64-dimensional features, and the second layer outputs 6-dimensional features). The output layer uses the Softmax activation function to map the 6-dimensional features to the probability distribution of 6 types of aquaculture events. The 3-bit binary code corresponding to the category with the highest probability is taken as the event category code. The regression branch operates in parallel with the classifier. It outputs a continuous value through the fully connected layer, which is quantized and converted into a 16-bit binary code for the feeding amount. The two are combined to form preliminary decision information.

[0098] S214. Based on the single feeding amount of the unmanned feeding vessel and the upper limit of the average daily feed intake of the cultured species (the upper limit of the average daily feed intake of the cultured species is a quantitative indicator of the physiological characteristics of the cultured species), a double validity check is performed. After removing abnormal data, a standard decision instruction that conforms to the preset format is generated.

[0099] Specifically, to avoid overfeeding or underfeeding due to abnormal decision-making instructions, a dual verification is performed based on equipment hardware limitations and the physiological characteristics of the farmed animals. The specific operation is as follows:

[0100] Single feeding amount setting: Based on the feeding bin capacity of the unmanned feeding vessel (e.g., a 100kg feeding bin) and the needs of the aquaculture scenario (to avoid frequent refeeding and excessive feeding at a single time leading to water quality deterioration), the single feeding amount is set as follows: If the initial decision is the amount of feed... Beyond this range ( or This is considered abnormal.

[0101] Calculation of daily feed intake limit: Taking the target species, Litopenaeus vannamei, as an example, the average daily feed intake per shrimp is 5%-8% of its body weight (a physiological characteristic and technical quantitative indicator). If the stocking density is... 10,000 fish / mu, average weight g / tail, then the upper limit of daily food intake per acre is:

[0102] ;

[0103] in, Indicates the upper limit of average daily feed intake in the breeding area (unit: kg); Indicates stocking density (unit: fish / acre); This indicates the average weight of the farmed animals (unit: kg / animal, taken here). ); This represents the average daily feeding rate per tail (taking the upper limit of physiological characteristic indicators). );

[0104] Substituting the previous example data into the calculation, we get:

[0105] ;

[0106] That is, the average daily feed intake in this breeding area shall not exceed 80 kg; if the cumulative daily feed intake is... Compared with the current preliminary decision feeding amount The sum exceeds (Right now This is considered abnormal.

[0107] Anomaly handling logic: If the initial decision data is abnormal, adjust the feeding amount based on the reasonable decision-making amounts over the past three periods. (Sorted by time from most recent to oldest) The standard decision feed amount is generated using a weighted average method, and the formula is:

[0108] ;

[0109] in, Indicates the standard decision-making feeding amount (unit: kg); This indicates the reasonable feeding amount for the most recent cycle; This indicates the reasonable feeding amount for the most recent two cycles; This indicates the reasonable feeding amount for the most recent three cycles;

[0110] Give higher weight to recent data to ensure timely decision-making; if the preliminary decision data is normal, it is directly converted into a standard decision instruction in the format of "event category code-feed amount code-data collection timestamp".

[0111] In this step, in S2, the decision instruction adopts a hierarchical encoding method driven by aquaculture events, including the following steps:

[0112] S215. Divide aquaculture events into multiple core types, assign a unique binary code to each type, and form a fixed mapping between the code and the event type;

[0113] Specifically, based on common feeding-related events in aquaculture scenarios, aquaculture events are divided into 6 core types, each assigned a unique 3-bit binary code, forming a fixed mapping table stored in the edge computing unit. The mapping relationship is as follows:

[0114] Routine feeding event (code 001): Environmental parameters and feeding status are within the normal range, and feeding is carried out according to the routine plan;

[0115] Emergency feeding adjustment event due to abnormal water quality (code 010): If the pH value or ammonia nitrogen concentration exceeds the normal range, reduce the feeding amount by 50%;

[0116] Insufficient feed intake supplementation event (code 011): When the amount of remaining feed is below the threshold (e.g., 5%), supplement with 30% of the normal amount;

[0117] Overfeeding stoppage event (code 100): When the amount of remaining feed is higher than the threshold (e.g., 30%), the current feeding is suspended;

[0118] Water temperature change event (code 101): If the water temperature changes by more than 3°C in 24 hours, the feeding amount should be adjusted to 70% of the normal amount.

[0119] Dissolved oxygen deficiency feeding suspension event (code 110): When the dissolved oxygen content is below 5 mg / L, suspend feeding until it returns to normal.

[0120] S216. Extract the aquaculture event type and feeding amount information from the decision instruction, use the binary code of S215 as the event type field, set the feeding amount quantification unit based on the accuracy of the unmanned feeding boat feeding equipment, generate the feeding amount code field, and splice them in the order of "event type code - feeding amount code" to form the basic layer.

[0121] Specifically, the base layer focuses on core instruction information to ensure that critical content can still be transmitted even when the channel deteriorates:

[0122] Feeding quantity quantification unit setting: Minimum feeding accuracy based on the unmanned feeding vessel feeding equipment (e.g., ), set the quantification unit to 0.1kg, and change the original feeding amount Convert to quantized units The calculation formula is:

[0123] ;

[0124] in, This represents the number of units (non-negative integer) of the amount of feed given. Indicates the initial feeding amount (unit: kg); Indicates the minimum precision of the feeding equipment (unit: kg); This represents the floor function (ensuring the quantization result matches the actual feeding accuracy of the equipment); for example, the original feeding amount. ,but ,correspond Each quantification unit.

[0125] Basic layer splicing rules: Seamless splicing is performed in the order of "event type code (3-bit binary) - feeding amount code (16-bit binary)"—the 16-bit binary code can cover 0-65535 quantization units, corresponding to a feeding amount range of 0-6553.5kg, fully meeting the single feeding amount requirement of 5-50kg; after splicing, a 19-bit binary basic layer data is formed, and the encoded data length is... Bit.

[0126] S217. Taking the event basis of the decision-making instruction as the core, extract the normalized aquaculture water environment data and the quantitative feeding status data, generate summaries for the normalized aquaculture water environment data and the quantitative feeding status data respectively, and splice them in the order of "environmental data summary - feeding status data summary" to form an enhancement layer.

[0127] Specifically, the enhancement layer stores a summary of the data used to support decision-making, facilitating the shore-based control center's ability to verify the rationality of decisions. The details are as follows:

[0128] Data digest generation: The SHA-256 hash algorithm is used to operate on the normalized environmental data sequence (12-dimensional). The SHA-256 algorithm outputs a 256-bit binary number. The first 32 bits are taken as the environmental data digest (which can both represent the characteristics of the original data and reduce the amount of data). The same operation is performed on the standardized feeding state data sequence (8-dimensional), and the first 32 bits are also taken as the feeding state data digest.

[0129] Enhancement layer splicing rules: The data is spliced ​​in the order of "Environmental data digest (32-bit binary) - Feeding state data digest (32-bit binary)" to form 64-bit binary enhancement layer data. The encoded data length is... .

[0130] S218. Using a preset end-order encoding rule, the encoding module of the edge computing unit encodes the content of the base layer and the enhancement layer separately. The total length of the encoded instruction is as follows: Characterization, when transmitting the base layer alone During joint transmission ,in The length of the encoded base layer data. This represents the length of the encoded enhancement layer data.

[0131] Specifically, the preset endianness encoding rule is as follows: the little-endian encoding commonly used in industrial communication is adopted, that is, the low-order byte of the data is stored at the low address and the high-order byte is stored at the high address, which is consistent with the decoding rule of the shore-based control center to avoid decoding failure due to byte order disorder; the encoding module is integrated into the FPGA chip of the edge computing unit (such as Xilinx Artix-7), and the encoding operation is completed through hardware acceleration, with the encoding latency controlled within 10ms.

[0132] Calculation of total instruction length:

[0133] When transmitting the base layer separately, the total instruction length is the base layer data length, and the formula is: ; Indicates the length of the encoded base layer data (fixed to 19 bits);

[0134] When jointly transmitting the base layer and enhancement layer, the total instruction length is the sum of the data lengths of the two layers, as shown in the formula: ; This indicates the length of the encoded enhancement layer data (fixed at 64 bits); substituting the data yields the combined transmission result. Bit.

[0135] In this step, S2, communication channel parameters are continuously collected for a preset number of periods and the channel quality change rate is calculated to adapt to the transmitted content, including the following steps:

[0136] S219. Through the communication unit carried by the unmanned feeding boat, the channel core parameters, including signal-to-noise ratio, bit error rate, and transmission delay, are continuously collected for a preset number of cycles, and the collection cycle is synchronized with the data collection cycle of S1.

[0137] Specifically, to accurately assess channel quality, key parameters are collected at fixed intervals. The specific operation is as follows:

[0138] Preset number of acquisition cycles: Set the number of continuous acquisition cycles. — Five cycles can reflect short-term channel change trends without causing decision lag due to too many acquisition cycles. The acquisition cycle is synchronized with the data acquisition cycle of S1 (e.g., if the acquisition cycle of S1 is 2 minutes, the channel parameters are acquired once every 2 minutes, for a total of 5 acquisitions and 10 minutes).

[0139] Data Acquisition Parameters and Methods: Three core parameters—signal-to-noise ratio (SNR)—were collected via the LoRa+4G dual-mode communication unit mounted on the unmanned feeding vessel. ), bit error rate ( ), transmission delay ( ); 10 parameter values ​​are collected within each acquisition cycle, and the arithmetic mean is taken as the parameter value for that cycle. The formula is:

[0140] ;

[0141] in, This represents the average value of the parameter during that period (which can be...). , or ); This indicates the number of data collections per cycle; Indicates the first [number]th ... The original values ​​of the parameters collected each time; averaging them can reduce parameter fluctuations caused by random interference and improve the accuracy of the evaluation.

[0142] S220. Based on the collected core channel parameters, a weighted algorithm is used to calculate the overall channel quality value for each cycle. , This is the normalized result after weighting the parameters;

[0143] Specifically, a weighted algorithm is used to comprehensively evaluate channel quality based on multiple parameters. The specific operation is as follows:

[0144] First, parameter normalization is performed, mapping the original values ​​of each parameter to the [0,1] interval to eliminate dimensional differences. This specifically includes:

[0145] Signal-to-noise ratio (SNR) (Positive correlation parameter: the larger the value, the better the channel quality) Normalization formula:

[0146] ;

[0147] in, This represents the normalized signal-to-noise ratio (range [0,1]). This represents the raw signal-to-noise ratio (SNR) value collected (unit: dB, range: -120dB to 0dB). =-120dB (Minimum effective value of signal-to-noise ratio); =0dB (maximum effective value of signal-to-noise ratio);

[0148] Bit error rate (BER) Negative correlation parameter: the smaller the value, the better the channel quality. Normalization formula:

[0149] ;

[0150] in, This represents the normalized bit error rate (range [0,1]). This represents the original bit error rate value collected (range [0, 0.1]). =0 (minimum effective value of bit error rate); =0.1 (maximum effective value of bit error rate; if exceeded, the channel cannot transmit normally).

[0151] Transmission delay ( Negative correlation parameter: the smaller the value, the better the channel quality. Normalization formula:

[0152] ;

[0153] in, This represents the normalized transmission delay (range [0,1]). This represents the raw value of the transmission delay collected (unit: ms, range: 0ms~1000ms). =0ms (minimum effective value of transmission delay); =1000ms (maximum effective value of transmission delay; exceeding this value will not meet the low latency requirement).

[0154] Then, a weighted composite value is calculated:

[0155] Weights are assigned based on the degree of influence of each parameter on channel quality (signal-to-noise ratio has the greatest impact, weight 0.4; bit error rate and transmission delay have comparable impacts, each weighted 0.3). The formula for the comprehensive value is as follows:

[0156] ;

[0157] in, This represents the overall channel quality value (range [0,1], the closer the value is to 1, the better the channel quality).

[0158] Indicates the signal-to-noise ratio weight; Indicates the bit error rate weight; Represents the transmission delay weight; the weight constraint is: This ensures that the calculation of the composite value conforms to the weighted logic.

[0159] S221. Channel quality composite value based on continuous acquisition period The channel quality change rate was calculated. ;

[0160] Specifically, the rate of change is calculated using linear regression to characterize the trend of channel quality changes: assuming continuous... The channel quality composite values ​​for each cycle are as follows: (Corresponding period number) ), rate of change For linear regression equations The slope is calculated using the following formula:

[0161] ;

[0162] in, This represents the rate of change in channel quality (unit: 1 / period, positive value indicates channel quality improvement, negative value indicates deterioration). Indicates the number of consecutive data acquisition cycles; Represents the period number (positive integer, ); Indicates the first The overall channel quality value for each cycle;

[0163] For example, if 5 cycles Given values ​​of 0.8, 0.78, 0.75, 0.72, and 0.7 respectively, substituting these values ​​into the formula yields the following calculation: Characterizes the degradation of channel quality per cycle .

[0164] S222, Preset negative rate of change trigger threshold When the channel quality change rate When the channel quality is determined to be deteriorating, only the basic layer coded content is selected as the transmitted data; when If the channel quality remains stable, the jointly coded content of the base layer and enhancement layer is selected as the transmission data.

[0165] Specifically, a preset negative rate of change trigger threshold is set. (Based on historical data of channel fluctuations in aquaculture waters, when each cycle) When the drop exceeds 0.05, the channel deteriorates significantly, and the transmitted content needs to be simplified. The judgment logic is as follows:

[0166] ;

[0167] in:

[0168] This indicates the preset negative rate of change trigger threshold (unit: 1 / cycle). This represents the calculated rate of change in channel quality;

[0169] when At this time, the channel quality deteriorates rapidly, and only the core instruction information of the basic layer is transmitted to reduce the amount of data in order to reduce transmission latency and bit error rate;

[0170] when At any time, the channel quality is stable or changes slowly, transmitting joint content to ensure that the shore-based control center obtains complete decision-making information.

[0171] In this step, in S2, the encoded instructions are encapsulated according to a preset frame structure, including the following steps:

[0172] S223. The frame structure of the preset instruction encapsulation is "header identifier - node information - hierarchical instruction segment - timestamp - check code". Each field is seamlessly spliced ​​in order and stored in the edge computing unit of the unmanned feeding vessel.

[0173] Specifically, to ensure the recognizability and integrity of instructions during transmission, the preset frame structure is "header identifier - node information - hierarchical instruction segment - timestamp - checksum". These fields are seamlessly concatenated in sequence and stored in the buffer of the edge computing unit (capacity ≥ 1MB). The specific length is as follows:

[0174] Header identifier: 8 characters;

[0175] Node information: 32 bits;

[0176] Layered instruction segment: 19-bit (base layer only) or 83-bit (joint transmission);

[0177] Timestamp: 32 bits;

[0178] Check digit: 128 bits;

[0179] When only the base layer is transmitted, the total frame length is Bit, used during joint transmission Bit.

[0180] S224. The header identifier is set to a preset binary fixed code, which is used by the shore-based control center to identify the command type; the node information includes the identification code of the sensing node unit and a summary of GPS positioning coordinates.

[0181] Specifically, the header identifier and node information are set as follows:

[0182] Header identifier: Set to an 8-bit binary fixed code 10101100. This code is a unique identifier for aquaculture communication commands. After receiving the command, the shore-based control center first parses the header identifier. If it matches the preset code, it continues parsing. Otherwise, it is determined to be an invalid command and discarded to avoid confusion with commands from other devices.

[0183] Node information: includes a 16-bit sensing node unit identification code (completely consistent with the code assigned by S11, ensuring data traceability) and a 16-bit GPS positioning coordinate summary;

[0184] Furthermore, the above-mentioned GPS positioning coordinate summary generation steps are as follows: convert the latitude and longitude coordinates of the sensing node unit (accurate to 6 decimal places, such as latitude 30.123456°, longitude 120.654321°) into decimal integers (latitude × 10^6 = 30123456, longitude × 10^6 = 120654321), perform CRC-16 operation (polynomial 0x8005) on the two integers, and generate a 16-bit CRC-16 value as the GPS positioning coordinate summary, which simplifies the coordinate data and can uniquely represent the node position.

[0185] S225. Based on the transmission mode determined by S222, fill in the basic layer content after S218 encoding, or the joint encoding content of the basic layer and the enhancement layer. The timestamp adopts a preset format that is compatible with the data acquisition cycle of S1 and is accurate to a preset time unit.

[0186] Specifically, based on the transmission mode determined by S222, the base layer data or combined data after little-endian encoding is filled in. When filling, the original binary format after encoding is preserved, and no extra redundant fields are added to ensure data compactness.

[0187] Meanwhile, the timestamp uses UTC time format, accurate to the second, with "2020-01-01 00:00:00" as the starting time point. The difference in seconds between the current collection time and the starting time point is calculated and converted into a 32-bit binary number. For example, the difference in seconds between 2024-05-20 14:30:25 and the starting time point is 1398263425, which corresponds to a 32-bit binary number as the timestamp, ensuring that the time correlation between instructions and collected data is traceable.

[0188] S226. Generate a checksum using a combined CRC-32 and MD5 algorithm. First, perform a CRC-32 operation on the header identifier, node information, hierarchical instruction segment, and timestamp fields. Then, perform an MD5 operation on the CRC-32 operation result together with the header identifier, node information, hierarchical instruction segment, and timestamp fields. Finally, use the combined operation result as the checksum and append it to the end of the frame structure to complete the instruction encapsulation.

[0189] Specifically, a combined CRC-32 and MD5 verification mechanism is adopted to balance the requirements of transmission error detection and data anti-tampering. The specific implementation is as follows:

[0190] First, perform the CRC-32 calculation:

[0191] The standard CRC-32 / MPEG-2 checksum algorithm is adopted, and the polynomial is: (represented in hexadecimal as 0x04C11DB7), the data to be processed is a binary data string formed by concatenating four fields: header identifier, node information, hierarchical instruction segment, and timestamp. The calculation formula is:

[0192] ;

[0193] in, This represents a 32-bit CRC-32 checksum.

[0194] This represents a binary data string composed of header identifier, node information, hierarchical instruction segment, and timestamp in sequence. This represents the standard CRC-32 / MPEG-2 cyclic redundancy check function.

[0195] Then perform the MD5 calculation:

[0196] 32-bit CRC-32 check value With data string according to" in front, The binary strings are concatenated in the order of "last" to form a new data string. ( (representing binary concatenation), for The standard MD5-128 message digest algorithm is used, and the formula is:

[0197] ;

[0198] in, This represents a 128-bit MD5 checksum (i.e., the final checksum of this scheme). This represents the standard MD5-128 message digest algorithm function.

[0199] Finally, the process involves encapsulation: a 128-bit checksum is then created. Seamlessly spliced ​​to the end of the frame structure to form a complete instruction frame, which is stored in the edge computing unit of the unmanned feeding vessel for subsequent network transmission.

[0200] S3. Multi-node anti-interference networking: Using the unmanned feeding boat as a mobile communication relay, a dynamic network is built with each sensing node unit to select effective communication links and autonomously switch to the optimal channel.

[0201] In this step, S3, the multi-node anti-interference networking includes the following steps:

[0202] S31. The unmanned feeding vessel periodically sends link detection signals through the communication unit. Each sensing node unit responds with a response signal based on its own unique identifier. The unmanned feeding vessel records the signal arrival time, signal strength and link connectivity status of each sensing node unit and establishes a node link information table.

[0203] Specifically, the unmanned feeding vessel is equipped with a LoRa+4G dual-mode communication unit (supporting switching between the 433MHz LoRa band and the 4GCat-1 band, adapting to the long-distance and high-bandwidth communication requirements of aquaculture waters), and sends link detection signals to all monitoring points in the aquaculture waters at a cycle of 10 seconds. The detection signal adopts a standardized frame structure, with the format "feeding vessel identification (8-bit binary) - detection sequence number (16-bit binary) - detection timestamp (32-bit binary)", where the detection sequence number increments by a natural number (starting from 1, and resetting to 1 after overflow), ensuring that each detection signal is unique and traceable;

[0204] Furthermore, after receiving the detection signal, each sensing node unit feeds back a response signal within 100ms. The response signal format is "node unique identifier code (16-bit binary, consistent with S11) - GPS positioning coordinate summary (16-bit binary, consistent with S224) - signal reception strength (8-bit binary)". After receiving the response signal, the communication unit of the unmanned feeding vessel records key information and establishes a node link information table, which is stored in the cache of the edge computing unit. After each detection and response reception is completed, the unmanned feeding vessel automatically updates the node link information table. For nodes that do not feed back a response signal, their link connectivity status is set to "disconnected", and the detection success count is reset to zero. For nodes that successfully feed back a response, the signal arrival time, received signal strength, and last detection time are updated, and the detection success count is incremented by 1.

[0205] In addition, for nodes that do not respond, a "periodic retry probe mechanism" is activated—the link probe is re-initiated every 2 data collection cycles (i.e., 2×1-5 minutes) for 5 cycles; if there is still no response, it is marked as an "offline node" and uploaded to the shore-based control center, while the subsequent periodic probes every 30 minutes are retained until the connection is restored.

[0206] S32. Using link stability, transmission delay, and packet loss rate as the core screening indicators, quantitatively evaluate the detected communication links, eliminate invalid links with signal strength below the preset threshold or packet loss rate exceeding the preset range, retain valid links that meet data transmission requirements, and update the node link information table.

[0207] Specifically, using "link stability, transmission latency, and packet loss rate" as the core screening indicators, all detected communication links are quantitatively evaluated. The specific operation is as follows:

[0208] First, the indicators are quantified and calculated, including:

[0209] Link stability The calculation is based on the results of the most recent preset number of detections (e.g., 3-10 times), and the formula is as follows:

[0210] ;

[0211] in:

[0212] Indicates link stability (range [0,1]); This indicates the number of times a response signal was successfully received in the most recent preset number of probes; This indicates the total number of times the system has been detected recently (based on a preset count).

[0213] Transmission delay The formula is based on the signal round-trip time:

[0214] ;

[0215] in:

[0216] Indicates transmission delay (unit: ms);

[0217] Indicates the arrival time of the response signal (taken from the node link information table);

[0218] Indicates the time the detection signal was sent (consistent with the timestamp in the detection signal);

[0219] Packet loss rate The calculation is based on the results of the most recent preset number of detections (e.g., 5-20 times), and the formula is as follows:

[0220] ;

[0221] in, Indicates the packet loss rate (range [0,1]); This indicates the number of times a response signal was not received in the most recent preset number of probes; This indicates the total number of times the system has been detected recently (based on a preset count).

[0222] Next, the filtering thresholds and rules are determined, where:

[0223] Signal strength threshold: preset based on the communication environment and characteristics of the communication unit in the aquaculture water area (e.g., not lower than -90dBm). When it is lower than this threshold, the signal attenuation is severe and the transmission reliability is significantly reduced.

[0224] Packet loss rate threshold: Preset upper limit for packet loss rate (e.g., not exceeding 10%), if the packet loss rate is exceeded, continuous data transmission cannot be guaranteed;

[0225] Quantitative evaluation score: The comprehensive evaluation score of the link is calculated using a weighted summation method, and the formula is as follows:

[0226] ;

[0227] in, This represents the overall link evaluation score (range [0,1]).

[0228] The weights for link stability, transmission latency, and packet loss rate are respectively (satisfying) (e.g., set to 0.4, 0.3, 0.3 respectively).

[0229] This indicates a reference value for transmission delay (e.g., 100ms). For the transmission delay normalization result ( The larger the value, the lower the score for this item.

[0230] Filtering logic: If the received signal strength of the link... Below the preset threshold, or packet loss rate Exceeding the preset limit, or the comprehensive evaluation score Links below a preset scoring threshold (e.g., 0.6) are deemed invalid and removed from the node link information table; the remaining links are deemed valid and their corresponding stability, transmission latency, and packet loss rate data are retained and updated in the information table.

[0231] S33. For the candidate channels corresponding to the effective links, combine the core channel parameters collected in S2 to calculate the comprehensive communication quality score of each candidate channel, and determine the channel with the highest score and that meets the low latency transmission standard as the optimal channel.

[0232] Specifically, the candidate channel communication quality score and optimal channel determination are as follows:

[0233] Candidate channel range definition: Select multiple LoRa candidate channels commonly used in aquaculture waters (e.g., 4-10), with frequency bands that meet local electromagnetic environment compatibility requirements, avoid frequency band conflicts with other IoT devices, and ensure channel usage compliance and anti-interference foundation.

[0234] Communication quality comprehensive score calculation: Combining the core channel parameters (signal-to-noise ratio, bit error rate, transmission delay) collected by S219 with the link indicators (link stability) of S32, a weighted algorithm is used to calculate the comprehensive communication quality score of each candidate channel. The formula is as follows:

[0235] ;

[0236] in,

[0237] This represents the overall communication quality score of the candidate channel (range [0,1], the higher the score, the better the channel quality).

[0238] This represents the normalized signal-to-noise ratio (calculated using the same method as S220, ranging from [0,1]).

[0239] This represents the channel bit error rate (raw value, normal range [0, 0.1]).

[0240] Indicates channel transmission delay (raw value, unit: ms);

[0241] Indicates a reference value for transmission delay (e.g., 1000ms);

[0242] This indicates the stability of the corresponding link (consistent with the S32 calculation method, range [0,1]).

[0243] The weights of each indicator are respectively (satisfying) (e.g., set to 0.3, 0.2, 0.2, 0.3 respectively).

[0244] Optimal channel determination rule: First, select channels that meet the "low latency transmission requirement" (transmission delay). Candidate channels with a latency ≤ preset upper limit (e.g., 200-500ms) are selected; among the channels that meet the latency requirements, a comprehensive communication quality score is chosen. The channel with the highest score is selected as the optimal channel; if multiple channels have a score difference ≤ a preset threshold (e.g., 0.01), the channel with the highest signal-to-noise ratio is selected first. A higher channel is selected; if the signal-to-noise ratio remains the same, one of them is randomly selected as the optimal channel, and the optimal channel identifier is stored in the node link information table.

[0245] S34. Monitor the quality status of the current communication link in real time. When the comprehensive link quality score is lower than the preset ratio of the optimal channel score, or when an abnormal situation occurs, trigger the autonomous switching mechanism and quickly switch to the optimal channel based on the node link information table. At the same time, repeat steps S31-S33 periodically to update the link and channel information and maintain the anti-interference stability of the dynamic network.

[0246] Specifically, the link quality monitoring, dynamic switching, and information updates in this step include:

[0247] Real-time monitoring mechanism: The unmanned feeding vessel monitors the quality status of the current communication link in real time according to a preset monitoring cycle (e.g., 3-10 seconds / time). The monitoring indicators include the comprehensive communication quality score. Packet loss rate Transmission delay The monitoring data is taken directly from the communication unit and node link information table.

[0248] The specific trigger conditions for switching include:

[0249] Scoring Trigger: Overall communication quality score of the current link. Below the optimal channel score The preset ratio (e.g., 70%-90%).

[0250] Abnormal Trigger: Any of the following abnormal conditions occurs: the link connectivity status is "disconnected" for a preset number of consecutive times (e.g., 2-5 times); packet loss rate. The latency suddenly rises above a preset threshold (e.g., 0.2-0.4); transmission delay The signal strength RSSI is lower than the preset low threshold (e.g., -95dBm to -100dBm) for a preset period of time (e.g., 1-3 times) and continues for a preset period of time.

[0251] Autonomous switching mechanism:

[0252] When the switching conditions are triggered, the communication unit of the unmanned feeding vessel immediately executes the switching procedure, which includes:

[0253] Stop sending data to the current link and buffer the instruction frames to be transmitted (the buffer capacity is adapted to short-term data storage needs to avoid data loss during the handover).

[0254] Based on the optimal channel identifier in the node link information table, quickly switch to the optimal channel, and complete the switching process within a preset time limit (e.g., 300-1000ms).

[0255] After the handover is completed, a "link handover notification" (containing the original channel identifier, the new channel identifier, and the handover timestamp) is sent to the shore-based control center, and data transmission is resumed.

[0256] If the optimal channel quality still does not meet the requirements after switching (overall score < preset qualified threshold, such as 0.5-0.6), then other candidate channels corresponding to valid links will be tried in descending order of score until a qualified channel is switched to (maximum number of attempts is 3; exceeding the number of attempts will trigger a channel anomaly alarm). If no qualified channel is found after triggering the channel anomaly alarm, the default public channel of the 4G communication unit (such as the 1800MHz band) will be automatically enabled for forced transmission; at the same time, the "insufficient channel resources" information will be encapsulated as an emergency command, and the currently available bandwidth will be used to complete the transmission first.

[0257] Furthermore, the information update specifically includes: repeating steps S31-S33 at a preset update cycle (e.g., 30-120 seconds / time), re-detecting links, filtering valid links, calculating channel scores and updating the optimal channel, and synchronously updating all fields in the node link information table to ensure that the link and channel information always match the communication environment of the current aquaculture area, maintaining the anti-interference stability of the dynamic network. If the unmanned feeding vessel moves to a new aquaculture area (judging from its own GPS location changes by more than a preset distance threshold, such as 30-100 meters), a full-process update of S31-S33 is immediately triggered to quickly adapt to the communication link environment of the new area.

[0258] S4. Data transmission scheduling: The encoded instruction content is transmitted in time-sharing according to the priority of the decision instruction, and forwarded by the unmanned feeding vessel to the shore-based control center through a dual-mode redundant link;

[0259] In this step, in S4, the data transmission scheduling includes the following steps:

[0260] S41. Based on the aquaculture event types generated by S2, set multiple priorities according to the urgency of the events. The decision instructions corresponding to events involving the survival safety of the aquaculture objects are classified as the highest priority, and the decision instructions corresponding to regular feeding events are classified as the basic priority. Establish a mapping relationship between priority and instruction type and store it in the edge computing unit of the unmanned feeding vessel.

[0261] Specifically, based on the six types of aquaculture events defined in S215, three levels of priority are established according to the degree of impact of the events on the survival of the aquaculture objects and the aquaculture benefits. A fixed mapping relationship between priority and instruction type is established, as follows:

[0262] The highest priority (P1) corresponds to the feeding suspension event due to insufficient dissolved oxygen (code 110) and the emergency feeding adjustment event due to abnormal water quality (code 010). These events directly threaten the survival of the aquaculture species and require immediate response.

[0263] Medium priority (P2) corresponds to water temperature change and feed adjustment events (code 101) and overfeeding and feed cessation events (code 100). These events may lead to aquaculture losses and require rapid handling.

[0264] Basic priority (P3) corresponds to regular feeding events (code 001) and supplemental feeding events due to insufficient feed intake (code 011). These events are routine operations in aquaculture and can be transmitted in sequence.

[0265] Furthermore, the aforementioned mapping relationship is stored in a structured data format in the non-volatile storage area (such as a Flash chip) of the edge computing unit of the unmanned feeding vessel, supporting remote updates via a shore-based control center to adapt to the needs of new aquaculture event types. After the edge computing unit generates decision instructions, it automatically extracts the event type code from the instructions and matches the corresponding priority level by querying the mapping relationship, providing a core basis for subsequent time slot allocation.

[0266] S42. According to the priority mapping relationship, independent transmission time slots are allocated to instructions of different priorities. The highest priority instructions occupy the priority transmission time slot, and the basic priority instructions occupy the remaining time slots in a time-sharing manner according to a preset order. The time slot length is adapted to the amount of instruction content data after S2 encoding.

[0267] Specifically, considering the data acquisition cycle (1-5 minutes) of S1 and the requirements for command transmission efficiency, the data transmission cycle is set to 10 seconds, with several independent time slots within each cycle to ensure that all priority commands can be transmitted within one data acquisition cycle, thus avoiding data backlog.

[0268] Furthermore, the time slot length is related to the total length of the instruction after S2 encoding. and communication link transmission rate Adaptation ensures complete transmission of commands; the calculation formula is as follows:

[0269] ;

[0270] in, Indicates the length of a single time slot (unit: ms); Indicates the total length of the instruction (unit: bits, with a value of 19 bits or 83 bits, taken from S218). This indicates the transmission rate of the communication link (unit: kbps, such as the typical transmission rate of LoRa links, which is 12.5-250 kbps). The conversion factor from bit to byte;

[0271] Indicates the time slot protection interval (unit: ms, preset to 5-20ms, used to avoid signal interference between adjacent time slots);

[0272] This indicates the floor function, ensuring that the complete instruction can be accommodated even with fluctuations in the transmission rate.

[0273] In addition, two priority transmission time slots are reserved in each transmission cycle and specifically allocated to the highest priority (P1) instructions. The time slot length is calculated based on the longest instruction (83 bits) to ensure that the highest priority instructions are transmitted without delay. Three intermediate transmission time slots are reserved for intermediate priority (P2) instructions, with the time slot length consistent with the priority transmission time slots. The remaining time slots are all basic transmission time slots and allocated to basic priority (P3) instructions, which are transmitted in a first-in-first-out (FIFO) order. If a certain priority has no instructions to be transmitted, its corresponding time slot is automatically allocated to basic priority instructions, improving time slot utilization.

[0274] Understandably, if there are insufficient time slots in a single transmission cycle, a "priority fallback scheduling" is executed—time slots are allocated first to P1 (urgent instructions) and P2 (feeding instructions), while P3 (regular data) is delayed until the next transmission cycle; if there are still insufficient time slots in the next cycle, the P3 data is compressed (using a conventional lightweight compression algorithm in this field) before transmission.

[0275] S43. The unmanned feeding vessel activates the preset dual-mode communication link, binds the main link to the optimal channel determined by S3, associates the backup link with the redundant channel, and configures the link switching trigger conditions.

[0276] Specifically, the unmanned feeding vessel uses a LoRa+4G dual-mode communication link. The main link is bound to the optimal channel determined by S33 (preferably LoRa band, which is suitable for long-distance, low-power transmission requirements in aquaculture waters). The backup link is associated with a redundant channel. The redundant channel is selected from the candidate channels of S3 with the second highest comprehensive communication quality score and meets the low latency requirements. If the optimal channel is LoRa band, the redundant channel can be selected from 4G band, forming cross-band redundancy and improving link reliability.

[0277] Specifically, this embodiment presets four types of triggering conditions. The link switching will be initiated when any one of the conditions is met, as follows:

[0278] Main link packet loss rate (≥2 failures in 10 consecutive transmissions);

[0279] Main link transmission delay (Three consecutive instruction transmissions exceed this threshold);

[0280] Main link signal strength (Continues for 2 transmission cycles);

[0281] The main link connectivity status is "disconnected" (no response after two consecutive link probes).

[0282] In addition, the channel parameters, switching thresholds and other configuration information of the dual-mode link are stored in the link configuration file of the edge computing unit, which allows staff to dynamically adjust them according to the actual communication environment of the aquaculture waters.

[0283] S44. The unmanned feeding vessel forwards the coded instructions to the shore-based control center through the main link according to the time-sharing scheduling rules; it monitors the transmission status of the main link in real time, and immediately switches to the backup link to continue transmission when the main link status triggers the switching condition.

[0284] Specifically, the communication unit of the unmanned feeding vessel initiates scheduling according to a set 10-second transmission cycle. It first checks for the existence of a highest priority (P1) instruction; if present, it immediately occupies the priority transmission slot for forwarding. After the highest priority instruction is transmitted, it sequentially checks and processes medium priority (P2) and basic priority (P3) instructions, transmitting them one by one according to the allocated time slots. During transmission, after each instruction is forwarded, the "instruction identifier - transmission time slot - link type - transmission status" information is recorded in real time and stored in the transmission log for easy subsequent traceability.

[0285] Specifically, the communication unit continuously monitors the status of the main link every 2 seconds, collects key indicators such as packet loss rate, transmission latency, and signal strength in real time, and compares them with preset switching conditions.

[0286] Furthermore, when the main link status triggers the switching condition, the communication unit immediately suspends main link data transmission and caches the instruction frames to be transmitted in the edge computing unit's temporary storage area (cache capacity ≥ 20 frames to avoid data loss during switching); it quickly switches to the backup link, with the switching process completed within 1 second. After the switch is completed, a "link switching notification" is sent to the shore-based control center, including the switching reason, new link parameters, and switching timestamp; the uncompleted transmission instructions are retrieved from the temporary storage area and forwarded through the backup link in the original priority order. After the switch, the status of the backup link is continuously monitored. If the main link status returns to normal (meeting "packet loss rate ≤ 0.05, transmission latency ≤ 200ms, signal strength ≥ -85dBm" for 3 consecutive transmission cycles), it automatically switches back to the main link to ensure transmission efficiency.

[0287] S5. Data reception and status feedback: After receiving data, the shore-based control center verifies the integrity of the command and initiates retransmission as needed; each node provides real-time feedback on its own status, forming a communication closed loop.

[0288] In this step, S5, data reception and status feedback includes the following steps:

[0289] S51. The shore-based control center receives the coded instructions forwarded by the unmanned feeding vessel, extracts the checksum from the instructions, recalculates the received header identifier, node information, hierarchical instruction segment, and timestamp field using the same algorithm, and compares it with the extracted checksum.

[0290] Specifically, the shore-based control center is equipped with a dual-mode communication receiving module and a high-performance data processing unit that match the model of the unmanned feeding vessel. The command receiving and verification process is as follows:

[0291] First, the command reception and checksum extraction are performed: After receiving the command frame forwarded by the unmanned feeding vessel, the communication receiving module splits each field according to the frame structure defined in S223 and extracts the 128-bit MD5 checksum at the end of the frame. Simultaneously, the four core fields—header identifier, node information, hierarchical instruction segment, and timestamp—are separated and combined to form the data string to be verified. .

[0292] Then, verification calculations and comparisons are performed, as follows:

[0293] The data processing unit performs double verification according to the same algorithm specified in S226: first, the data string... Perform CRC-32 / MPEG-2 operations to generate a 32-bit checksum. ; then With data string according to" in front, The data strings are concatenated in the order of "last" to form a new data string. ,right Perform an MD5 operation to generate a 128-bit checksum. Finally, a comparison was made. With extraction If the two are completely consistent, the instruction is determined to be complete and unaltered; if they are inconsistent, the instruction transmission is determined to be abnormal.

[0294] S52. If the verification results are consistent, parse the instruction content; if the verification results are inconsistent, the shore-based control center initiates a retransmission request to the unmanned feeding vessel through a dual-mode redundant link, clarifying the identifier and transmission link of the retransmission instruction.

[0295] Specifically, if the verification results are inconsistent, the data processing unit immediately generates a retransmission request frame. The frame structure is defined as "control center identifier (16-bit binary) - retransmission instruction identifier (32-bit binary) - specified transmission link (2-bit binary: 01 = primary link, 10 = backup link) - retransmission time limit (16-bit binary, unit: seconds)". The "retransmission instruction identifier" is consistent with the unique identifier of the abnormal instruction, ensuring that the unmanned feeding vessel can accurately identify the instruction that needs to be retransmitted. The retransmission instruction identifier is generated by concatenating "original timestamp of instruction (16-bit binary) + unique identifier of node / device (16-bit binary) + number of retransmissions (2-bit binary)", for a total of 34 bits of binary encoding; the number of retransmissions takes values ​​of 01 (1st retransmission), 10 (2nd retransmission), and 11 (3rd retransmission).

[0296] Specifically, retransmission requests are sent to the unmanned feeding vessel via a dual-mode redundant link. If the anomaly is determined to be caused by the primary link, the backup link is selected first to send the request, thereby improving the request delivery rate.

[0297] Furthermore, the shore-based control center sets a retransmission waiting time limit (preset to 30 seconds). If a retransmission instruction is received from the unmanned feeding vessel within the time limit, the verification process of S51 is repeated. If no retransmission instruction is received after the time limit or the verification is still abnormal after retransmission, an "instruction transmission abnormality alarm" is triggered, and the abnormal information (including instruction identifier, transmission link, and number of abnormalities) is recorded in detail to facilitate staff to troubleshoot the problem.

[0298] S53. Each sensing node unit and the unmanned feeding vessel shall provide real-time feedback on their own operating status information according to a preset cycle, including power supply status, communication link quality, data acquisition and processing status, and the feedback information shall be associated with their own unique identity identifier.

[0299] Specifically, each sensing node unit and the unmanned feeding vessel report their operational status to the shore-based control center at a preset cycle of 30 seconds per report (adapted to the S1 data acquisition cycle). The feedback information adopts a standardized frame structure, with the format "node unique identifier (16-bit binary, consistent with S11) - status type code (8-bit binary) - status data (64-bit binary) - feedback timestamp (32-bit binary)". The status data includes power supply status (00 = normal, 01 = low power, 10 = power outage) and communication link quality (compared to S32). Consistent), data acquisition status (00=normal, 01=abnormal, 10=not collected), data processing status (only for unmanned feeding boats, 00=normal, 01=delayed, 10=failure); the sensing node unit directly sends feedback frames through its own communication structure, and the unmanned feeding boat synchronously sends them during the idle period of the instruction transmission time slot, avoiding the occupation of additional transmission resources.

[0300] S54. The shore-based control center receives status feedback information from all sensing node units, establishes and updates the sensing node unit status table in real time, triggers an alarm mechanism for abnormal states, and stores the command parsing results in association with the sensing node unit status information.

[0301] Specifically, after receiving node status feedback, the shore-based control center establishes a status table for sensing node units and updates it in real time. The table fields include "node ID, node type, power supply status, link quality score, acquisition status, processing status, last feedback time, and anomaly marker".

[0302] The anomaly detection rules are as follows: if the power supply status is "power outage", the link quality score is <0.5, the data collection / processing status is "abnormal / failed", or no status information is received for two consecutive feedback cycles, it is marked as abnormal. At the same time, alarms are divided into two levels: minor alarms trigger interface prompts and log records, while serious alarms simultaneously notify staff via SMS and APP push notifications. The command parsing results and node status information are stored in association with timestamps, and queries can be performed by node ID, time range, and event type, providing data support for aquaculture management and fault diagnosis.

[0303] This embodiment also provides a low-latency, anti-interference communication system for multi-source aquaculture data from unmanned feeding vessels, used to execute the aforementioned low-latency, anti-interference communication method for multi-source aquaculture data from unmanned feeding vessels, including:

[0304] The sensing node unit is deployed at different monitoring points in the aquaculture water area at preset intervals. Each sensing node unit is equipped with a unique identification code and positioning function, and has a built-in local cache module for collecting aquaculture water area environmental data in real time at preset time periods and storing it in the local cache module. It can also provide feedback on its own operating status information at preset periods, and the feedback information is associated with its own unique identification.

[0305] The sensing unit is mounted on the unmanned feeding vessel and has a built-in local cache module. It is used to synchronously start collecting feeding status data with the sensing node unit at a preset time period, store the collected data in the local cache module, and can provide feedback on its own data collection and processing status.

[0306] The edge computing unit, deployed on the unmanned feeding vessel, is used for real-time fusion processing of multi-source aquaculture data, generating standard decision instructions, executing hierarchical coding and instruction encapsulation, and completing channel quality assessment and related data storage. It includes a lightweight neural network module, an encoding module, and a data processing module, among which:

[0307] The lightweight neural network module is configured with an inference frame rate and input resolution adapted to real-time data processing. It is used to retrieve aquatic environmental data collected by the perception node unit and feeding status data collected by the sensor unit. It performs normalization processing on the environmental data and standardization processing on the feeding status data. It extracts single-modal features through the DepthwiseConv layer and completes cross-modal feature fusion through the Bottleneck layer. Then, it outputs preliminary decision information containing aquaculture event category code and feeding amount numerical code through the classifier. After dual validity verification of the single feeding amount of the unmanned feeding boat and the upper limit of the daily feeding amount of the aquatic object, it generates standard decision instructions that conform to the preset format.

[0308] The encoding module adopts a hierarchical encoding method driven by aquaculture events and a preset end-sequence encoding rule to encode the content of the basic layer and the enhancement layer separately. It also completes the instruction encapsulation according to the preset frame structure of "header identifier - node information - hierarchical instruction segment - timestamp - check code". The check code is generated by the CRC-32 and MD5 joint algorithm.

[0309] The data processing module is used to store the node link information table and the mapping relationship between priority and instruction type, and to calculate the overall channel quality value and rate of change.

[0310] The communication unit, mounted on the unmanned feeding vessel, acts as a mobile communication relay. It periodically sends link detection signals, records the signal arrival time, signal strength, and link connectivity status of each sensing node, and establishes a node link information table. It quantitatively evaluates communication links using link stability, transmission delay, and packet loss rate as core screening indicators, eliminates invalid links, and updates the node link information table. It calculates a comprehensive communication quality score for each candidate channel based on core channel parameters, selects the optimal channel, monitors link quality in real time, and triggers an autonomous switching mechanism to maintain network anti-interference stability. Simultaneously, it enables dual-mode communication links, binding the primary link to the optimal channel and associating the backup link with redundant channels. It forwards encoded instructions according to time-sharing scheduling rules, monitors link transmission status in real time, and triggers switching.

[0311] The shore-based control center receives coded instructions forwarded by the unmanned feeding vessel through its communication unit, extracts the checksum from the instructions, and recalculates the received header identifier, node information, hierarchical instruction segments, and timestamp fields using a combined CRC-32 and MD5 algorithm, comparing the recalculated values ​​with the checksum. If the checksums match, the control center parses the instruction content; otherwise, it initiates a retransmission request to the unmanned feeding vessel via a dual-mode redundant link, specifying the identifier and transmission link for the retransmission instruction. Simultaneously, the control center receives status feedback information from the sensing node units and the unmanned feeding vessel, establishes and updates the sensing node unit status table in real time, triggers an alarm mechanism for abnormal states, and associates and stores the instruction parsing results with the sensing node unit status information.

[0312] Those skilled in the art will understand that the process of implementing all or part of the steps of the above embodiments can be carried out by hardware or by a program instructing the relevant hardware.

[0313] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-source aquaculture data low-latency anti-interference communication method for unmanned feeding ships, characterized in that, Includes the following steps: S1. Multi-source aquaculture data collection: Real-time collection of aquatic environmental data and feeding status data is achieved through sensing node units at different monitoring points in the water area and sensing units carried by unmanned feeding boats; each sensing node unit is equipped with a unique identification and positioning function. S2. Edge Data Fusion and Layered Encoding and Encapsulation of Instructions: Based on the environmental data and feeding status data collected in S1, a lightweight neural network is used for real-time fusion processing at the unmanned feeding vessel to generate decision instructions for corresponding aquaculture events. The decision instructions adopt a layered encoding method driven by aquaculture events: the basic layer contains information on the type of aquaculture event and the amount of feed, and the enhancement layer contains the event basis of the decision instructions. Communication channel parameters are continuously collected for a preset number of periods, and the channel quality change rate is calculated. If the channel quality change rate exceeds a preset trigger threshold, it is determined that the channel quality is deteriorating, and only the content of the basic layer is transmitted. If the channel quality change rate does not exceed the preset trigger threshold, it is determined that the channel quality is stable, and the joint content of the basic layer and the enhancement layer is transmitted. The encoded instructions are then encapsulated according to a preset frame structure. In step S2, a lightweight neural network is used for real-time fusion processing to generate decision instructions for the corresponding aquaculture events, including the following steps: S211. Select MobileNet-SSD as a lightweight neural network and deploy it on the edge computing unit of the unmanned feeding boat, and configure the network inference frame rate and input resolution to adapt to real-time data processing. S212. Retrieve the aquatic environment data and feeding status data collected in S1, perform normalization processing on the environmental data, and perform standardization processing on the feeding status data. S213. Input the preprocessed data into the neural network according to the dimension of "environmental feature group - feeding feature group", extract single-modal features through the DepthwiseConv layer, complete cross-modal feature fusion through the Bottleneck layer, and then output preliminary decision information containing the category code of breeding event and the numerical code of feeding amount through the classifier. S214. Based on the single feeding amount of the unmanned feeding vessel and the upper limit of the average daily feed intake of the aquaculture species, a dual validity check is performed. After removing abnormal data, a standard decision instruction conforming to the preset format is generated. In step S2, the decision instruction adopts a hierarchical coding method driven by aquaculture events, including the following steps: S215. Divide aquaculture events into multiple core types, assign a unique binary code to each type, and form a fixed mapping between the code and the event type; S216. Extract the aquaculture event type and feeding amount information from the decision instruction, use the binary code of S215 as the event type field, set the feeding amount quantification unit based on the accuracy of the unmanned feeding boat feeding equipment, generate the feeding amount code field, and splice them in the order of "event type code - feeding amount code" to form the basic layer. S217. Taking the event basis of the decision-making instruction as the core, extract the normalized aquaculture water environment data and the quantitative feeding status data, generate summaries for the normalized aquaculture water environment data and the quantitative feeding status data respectively, and splice them in the order of "environmental data summary - feeding status data summary" to form an enhancement layer. S218. Using a preset end-order encoding rule, the encoding module of the edge computing unit encodes the content of the base layer and the enhancement layer separately. The total length of the encoded instruction is as follows: Characterization, when transmitting the base layer alone During joint transmission ,in The length of the encoded base layer data. The length of the encoded enhancement layer data; S3. Multi-node anti-interference networking: Using the unmanned feeding boat as a mobile communication relay, a dynamic network is built with each sensing node unit to select effective communication links and autonomously switch to the optimal channel. S4. Data transmission scheduling: The encoded instruction content is transmitted in time-sharing according to the priority of the decision instruction, and forwarded by the unmanned feeding vessel to the shore-based control center through a dual-mode redundant link; S5. Data reception and status feedback: After receiving data, the shore-based control center verifies the integrity of the command and initiates retransmission as needed; each node provides real-time feedback on its own status, forming a communication closed loop.

2. The low-latency, anti-interference communication method for multi-source aquaculture data from unmanned feeding vessels according to claim 1, characterized in that, In step S1, the multi-source aquaculture data collection includes the following steps: S11. Deploy the sensing node units at different monitoring points in the aquaculture water area at preset intervals, assign a unique identification code to each sensing node unit, and enable the positioning function of the sensing node unit to obtain the real-time location information of the sensing node unit. S12, the sensing node unit controlling the water monitoring points and the sensing unit carried by the unmanned feeding boat synchronously start the data acquisition operation according to the preset time cycle; S13. Store the aquatic environment data collected by the sensing node unit and the feeding status data collected by the unmanned feeding boat sensing unit to the local cache module of the corresponding acquisition terminal.

3. The low-latency, anti-interference communication method for multi-source aquaculture data from unmanned feeding vessels according to claim 1, characterized in that, In step S2, communication channel parameters are continuously collected for a preset number of periods and the channel quality change rate is calculated to adapt to the transmitted content, including the following steps: S219. Through the communication unit carried by the unmanned feeding boat, the channel core parameters, including signal-to-noise ratio, bit error rate, and transmission delay, are continuously collected for a preset number of cycles, and the collection cycle is synchronized with the data collection cycle of S1. S220. Based on the collected core channel parameters, a weighted algorithm is used to calculate the overall channel quality value for each cycle. , This is the normalized result after weighting the parameters; S221. Channel quality composite value based on continuous acquisition period The channel quality change rate was calculated. ; S222, Preset negative rate of change trigger threshold When the channel quality change rate When the channel quality is determined to be deteriorating, only the basic layer coded content is selected as the transmitted data; when If the channel quality remains stable, the jointly coded content of the base layer and enhancement layer is selected as the transmission data.

4. The low-latency, anti-interference communication method for multi-source aquaculture data from unmanned feeding vessels according to claim 3, characterized in that, In step S2, the encoded instructions are encapsulated according to a preset frame structure, including the following steps: S223. The frame structure of the preset instruction encapsulation is "header identifier - node information - hierarchical instruction segment - timestamp - check code". Each field is seamlessly spliced ​​in order and stored in the edge computing unit of the unmanned feeding vessel. S224. The header identifier is set to a preset binary fixed code, which is used by the shore-based control center to identify the command type; the node information includes the identification code of the sensing node unit and a summary of GPS positioning coordinates. S225. Based on the transmission mode determined by S222, fill in the basic layer content after S218 encoding, or the joint encoding content of the basic layer and the enhancement layer. The timestamp adopts a preset format that is compatible with the data acquisition cycle of S1 and is accurate to a preset time unit. S226. Generate a checksum using a combined CRC-32 and MD5 algorithm. First, perform a CRC-32 operation on the header identifier, node information, hierarchical instruction segment, and timestamp fields. Then, perform an MD5 operation on the CRC-32 operation result together with the header identifier, node information, hierarchical instruction segment, and timestamp fields. Finally, use the combined operation result as the checksum and append it to the end of the frame structure to complete the instruction encapsulation.

5. The low-latency, anti-interference communication method for multi-source aquaculture data from unmanned feeding vessels according to claim 4, characterized in that, In step S3, the multi-node anti-interference networking includes the following steps: S31. The unmanned feeding vessel periodically sends link detection signals through the communication unit. Each sensing node unit responds with a response signal based on its own unique identifier. The unmanned feeding vessel records the signal arrival time, signal strength and link connectivity status of each sensing node unit and establishes a node link information table. S32. Using link stability, transmission delay, and packet loss rate as the core screening indicators, quantitatively evaluate the detected communication links, eliminate invalid links with signal strength below the preset threshold or packet loss rate exceeding the preset range, retain valid links that meet data transmission requirements, and update the node link information table. S33. For the candidate channels corresponding to the effective links, combine the core channel parameters collected in S2 to calculate the comprehensive communication quality score of each candidate channel, and determine the channel with the highest score and that meets the low latency transmission standard as the optimal channel. S34. Monitor the quality status of the current communication link in real time. When the comprehensive link quality score is lower than the preset ratio of the optimal channel score, or when an abnormal situation occurs, trigger the autonomous switching mechanism and quickly switch to the optimal channel based on the node link information table. At the same time, repeat steps S31-S33 periodically to update the link and channel information and maintain the anti-interference stability of the dynamic network.

6. The low-latency, anti-interference communication method for multi-source aquaculture data from unmanned feeding vessels according to claim 5, is characterized in that... In step S4, data transmission scheduling includes the following steps: S41. Based on the aquaculture event types generated by S2, set multiple priorities according to the urgency of the events. The decision instructions corresponding to events involving the survival safety of the aquaculture objects are classified as the highest priority, and the decision instructions corresponding to regular feeding events are classified as the basic priority. Establish a mapping relationship between priority and instruction type and store it in the edge computing unit of the unmanned feeding vessel. S42. According to the priority mapping relationship, independent transmission time slots are allocated to instructions of different priorities. The highest priority instructions occupy the priority transmission time slot, and the basic priority instructions occupy the remaining time slots in a time-sharing manner according to a preset order. The time slot length is adapted to the amount of instruction content data after S2 encoding. S43. The unmanned feeding vessel activates the preset dual-mode communication link, binds the main link to the optimal channel determined by S3, associates the backup link with the redundant channel, and configures the link switching trigger conditions. S44. The unmanned feeding vessel forwards the coded instructions to the shore-based control center through the main link according to the time-sharing scheduling rules; it monitors the transmission status of the main link in real time, and immediately switches to the backup link to continue transmission when the main link status triggers the switching condition.

7. The low-latency, anti-interference communication method for multi-source aquaculture data from unmanned feeding vessels according to claim 6, characterized in that, In step S5, data reception and status feedback include the following steps: S51. The shore-based control center receives the coded instructions forwarded by the unmanned feeding vessel, extracts the checksum from the instructions, recalculates the received header identifier, node information, hierarchical instruction segment, and timestamp field using the same algorithm, and compares it with the extracted checksum. S52. If the verification results are consistent, parse the instruction content; If the verification results are inconsistent, the shore-based control center will send a retransmission request to the unmanned feeding vessel through a dual-mode redundant link, clearly identifying the retransmission command and the transmission link. S53. Each sensing node unit and the unmanned feeding vessel shall provide real-time feedback on their own operating status information according to a preset cycle, including power supply status, communication link quality, data acquisition and processing status, and the feedback information shall be associated with their own unique identity identifier. S54. The shore-based control center receives status feedback information from all sensing node units, establishes and updates the sensing node unit status table in real time, triggers an alarm mechanism for abnormal states, and stores the command parsing results in association with the sensing node unit status information.

8. A low-latency, anti-interference communication system for multi-source aquaculture data from unmanned feeding vessels, used to execute the low-latency, anti-interference communication method for multi-source aquaculture data from unmanned feeding vessels as described in any one of claims 1-7, characterized in that, include: The sensing node unit is deployed at different monitoring points in the aquaculture water area at a preset interval. Each sensing node unit is equipped with a unique identification code and positioning function, and has a built-in local cache module for collecting aquaculture water area environmental data in real time at a preset time period and storing it in the local cache module. It can also feed back its own operating status information at a preset period, and the feedback information is associated with its own unique identification. The sensing unit is mounted on the unmanned feeding vessel and has a built-in local cache module. It is used to synchronously start the feeding status data collection with the sensing node unit at a preset time period, store the collected data in the local cache module, and can provide feedback on its own data collection and processing status. An edge computing unit, deployed on an unmanned feeding vessel, is used to perform real-time fusion processing of multi-source aquaculture data, generate standard decision instructions, execute hierarchical coding and instruction encapsulation, and complete channel quality assessment and related data storage. The communication unit is mounted on the unmanned feeding vessel and serves as a mobile communication relay. It is used to periodically send link detection signals, record the signal arrival time, signal strength, and link connectivity status of each sensing node unit, and establish a node link information table. A shore-based control center is used to receive coded instructions forwarded by the unmanned feeding vessel through a communication unit.

Images