Fish keep-alive transportation process-oriented activity detection method and system

Abstract

The invention provides an activity detection method and system for a fish keep-alive transportation process, and relates to the technical field of information processing. A multi-modal fusion classification model is constructed by collecting fish activity standard parameters, prediction parameters and influence parameters, an activity index and an environment accumulation parameter are calculated in combination with transportation time, then an activity threshold value is determined, closed-loop adjustment is carried out on temperature, humidity and oxygen parameters through a self-adaptive feedback module, and the fish activity detection accuracy is improved. Real-time monitoring and dynamic management of fish activity are achieved, and the survival rate and stability of keep-alive transportation are improved.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to a method and system for detecting fish activity during the live transport process. Background Technology

[0002] The transportation and preservation of fish are crucial for ensuring their quality and survival. Waterless preservation is a clean method for transporting and preserving live fish. However, traditional waterless preservation methods often struggle to determine fish activity levels due to insufficient oxygen, variations in the transportation environment and fish behavior, and the uncertainty of transport time. These factors can lead to increased stress responses in fish and lower accuracy in activity assessment, thus affecting their survival rate and activity.

[0003] Therefore, it is particularly important to develop an effective activity detection method that can monitor and evaluate the activity status of fish in real time during waterless transportation. Summary of the Invention

[0004] In order to overcome the shortcomings of the prior art, the purpose of this invention is to provide a method and system for detecting the activity of live fish during the live transport process, which can realize real-time monitoring and visualization of the activity of live fish.

[0005] To achieve the above objectives, the present invention provides the following solution: A method for detecting fish activity during live transport, comprising: Fish are transported alive according to the preservation and transport environment, and fish activity standard parameters, activity prediction parameters, and activity influencing parameters are collected simultaneously during the transport process. Among them, the activity standard parameters include respiratory rate standard parameters, respiratory intensity standard parameters, blood glucose parameters, cortisol parameters, color standard parameters, and mucus standard parameters; the activity prediction parameters include infrared parameters, RGB parameters, and impedance parameters; the activity influencing parameters include transport time, as well as adjustable temperature parameters, humidity parameters, and oxygen parameters. Using the datasets corresponding to the activity standard parameters, activity prediction parameters, and activity impact parameters obtained within a single synchronous acquisition cycle as samples, fish activity is classified according to the activity standard parameters in the samples to obtain an activity knowledge base, and each sample is labeled with an activity tag. Based on the classification results of the activity knowledge base, the activity prediction parameters and the activity tags are fused and associated to form a multimodal fusion association classification model. The classification results of the multimodal fusion association classification model are combined with the transportation time to calculate the fish activity index. The cumulative temperature parameter, cumulative humidity parameter, and cumulative oxygen parameter are calculated based on the temperature parameter, the humidity parameter, and the oxygen parameter. The fish activity threshold is determined in combination with the transportation time. The activity index and the activity threshold are input into the adaptive feedback adjustment module to implement dynamic closed-loop adjustment of the temperature, humidity and oxygen parameters in the survival transport environment, and to provide real-time feedback of the adjustment results.

[0006] Preferably, the classification results of the multimodal fusion association classification model are combined with the transportation time to calculate the fish activity index, and the cumulative temperature parameter, cumulative humidity parameter, and cumulative oxygen parameter are calculated based on the temperature parameter, the humidity parameter, and the oxygen parameter, and the fish activity threshold is determined in combination with the transportation time, including: The classification function for fish activity is set as follows: ; where f cl (S) represents the classification function for fish activity indicators. It determines the activity level based on the range of fish activity values. Specifically, function values ​​from 1 to 5 correspond to different activity level classifications: 1: If the fish activity value S is within the range [S0, S1], it indicates that the fish activity is low.

[0007] 2: If the fish activity value S is within the range of [S1, S2], it indicates that the fish activity is low to moderate.

[0008] 3: If the fish activity value S is within the range of [S2, S3], it indicates that the fish activity is moderate.

[0009] 4: If the fish activity value S is within the range of [S3, S4], it indicates that the fish activity is relatively high.

[0010] 5: If the fish activity value S is greater than S4, it indicates that the fish activity is very high.

[0011] S0, S1, S2, S3, S4: These are the key points for classifying fish activity values, used to define thresholds for low, medium, and high levels of fish activity. They represent different ranges of fish activity values.

[0012] The activity index of a group of fish is predicted based on the activity S of individual fish during transportation and the transportation time t. The formula for calculating the activity index is as follows: Where t represents the transportation time, i.e., the time elapsed during the transportation of the fish; A t This represents the initial activity value of the fish at the start of transportation. C is a constant for health-related indicators. β and ξ represent preset weighting coefficients used to adjust the influence weight of different parameters. n represents the dimension of the activity parameter evolution.

[0013] Through formula Calculate the cumulative temperature parameter, cumulative humidity parameter, and cumulative oxygen parameter; among which, It is an intermediate variable in the transportation time. , and They are time points Temperature, humidity, and oxygen concentration at that time; These are accumulated parameters for temperature, humidity, or oxygen. Through formula Calculate the fish activity threshold; wherein, , and These are the weighting coefficients for environmental parameters, used to represent the influence of temperature, humidity, and oxygen concentration on the activity threshold. E represents the fish activity threshold. T (t) represents the temperature-related influencing factor, indicating the effect of temperature on fish activity value; it is a function that changes over time. H (t) represents the humidity-related influencing factor, indicating the effect of humidity on fish activity values, which changes over time; E o (t) represents the influencing factor related to oxygen concentration, indicating the effect of oxygen concentration on fish activity value, which changes over time; E(t) is the result of a weighted combination, which integrates the effects of temperature, humidity and oxygen concentration on fish activity value.

[0014] Preferably, the adaptive feedback adjustment module is constructed as follows: Based on the current population's activity index At and activity threshold Calculate the activity bias; the formula for calculating the activity bias is: ;in, This refers to the activity deviation; According to the formula Calculate the temperature regulation amount; where, , and These are the proportional, integral, and derivative control coefficients for temperature regulation. The temperature adjustment amount; According to the formula Calculate the humidity adjustment amount; where, , and These are the proportional, integral, and derivative control coefficients for humidity regulation. The humidity adjustment amount; According to the formula Calculate oxygen regulation ;in, , and These are the proportional, integral, and derivative control coefficients for humidity regulation. The oxygen regulation amount; These represent the coefficients of the humidity control, temperature control, and differential control systems, respectively, and represent the weights of the influence of different control processes on oxygen regulation. According to the formula Calculate the overall adjustment amount; where, This refers to the overall adjustment amount; These represent the degree of influence of temperature, humidity, and oxygen on the overall regulation amount, respectively. After adjusting according to the aforementioned comprehensive adjustment amount, the formula is used. Update temperature, humidity, and oxygen concentration; The current temperature adjustment amount is indicated by calculating the current temperature. and temperature change The obtained temperature value; The current humidity adjustment amount is indicated by calculating the current humidity. and humidity change The obtained humidity value; This indicates the current oxygen adjustment amount, based on the current oxygen level. and changes in oxygen Calculated; The adjustment interval is adjusted according to the rate of change in fish activity; the formula for the adjustment interval is: ;in, The adjustment interval is... ϑ 1 represents the adjustment sensitivity coefficient, which characterizes the degree of influence of changes in fish activity on the adjustment interval. The larger the value, the more significant the effect of changes in activity on shortening the interval. ϑ 2 represents the baseline adjustment weight, indicating the basic adjustment frequency that needs to be maintained even when the fish activity changes little or even close to zero, in order to prevent the system from losing feedback regulation.

[0015] An activity detection system for fish during live transport includes: A flexible multi-scale impedance sensor is attached to the surface of a fish to collect real-time data on changes in the fish's surface impedance and output impedance parameters. The RGB image acquisition module is used to acquire color images of the fish's surface and output RGB parameters. The infrared image acquisition module is used to acquire images of the temperature distribution on the fish body and output infrared parameters. The environmental parameter acquisition module is used to collect temperature, humidity, and oxygen parameters in the live transport environment. The grading system module is used to preprocess and extract features from the raw data from the flexible multi-scale impedance sensor, the RGB image acquisition module, and the infrared image acquisition module. Based on the extracted standard parameters of respiratory frequency, respiratory intensity, blood glucose, cortisol, color, and mucus, it constructs a set of activity standard parameters. The impedance parameters, RGB parameters, and infrared parameters are used as a set of activity prediction parameters. The temperature, humidity, and oxygen parameters collected by the environmental parameter acquisition module, as well as the transportation time recorded by the host computer module, are used to form a set of activity influencing parameters. A multimodal fusion network model is used to process, analyze, and classify the activity influencing parameters and the activity standard parameters to classify the fish into different activity states. The host computer module is used to centrally control and manage the operation of the flexible multi-scale impedance sensor, the RGB image acquisition module, the infrared image acquisition module, the environmental parameter acquisition module, and the classification system module. It records transportation time, displays sensor and classification result data in real time, provides data storage and historical query functions, and connects to a remote monitoring system via wireless transmission for remote control and early warning.

[0016] Preferably, the fabrication process of the flexible multi-scale impedance sensor includes: Import the pre-designed electrode pattern into the laser direct writing device; Laser-induced graphene was prepared on a flexible substrate using the laser direct writing device to form a circular LIG electrode structure with four arrays. A mixture of PDMS solution and curing agent is uniformly coated on the surface of the LIG electrode structure by spin coating. The substrate coated with the solution is then placed on a heating table at 60°C for curing for 2 hours, thereby forming a uniform PDMS protective layer. The cured substrate film and the LIG electrode structure are transferred to a flexible substrate through a transfer process. The transferred film is then cut into four identical flexible units to obtain the front-end unit of the sensor. The pre-designed circuit pattern is imported into the laser direct writing device, and laser etching is performed on the PI / Cu composite film to form electrical connection lines on the flexible circuit board, thus obtaining a flexible copper circuit. After the circuit is fabricated, the main control module, voltage conversion module, impedance acquisition module, I2C multiplexer module, and Bluetooth wireless transmission module are fixed to the surface of the flexible copper circuit using a soldering process. The main control module is responsible for coordinating the work of each part and processing data. The voltage conversion module provides a stable power supply to the system. The impedance acquisition module collects key data from the sensors in real time. The I2C multiplexer module expands the communication capability with external devices. The Bluetooth wireless transmission module enables wireless data transmission. In this system, the main control module coordinates the work of each module and processes data to ensure normal system operation. The voltage conversion module provides a stable power supply to ensure that each module operates at the appropriate voltage. The impedance acquisition module collects key data from sensors in real time, providing necessary monitoring information. The I2C multiplexing module expands the communication capabilities with external devices, enabling the system to connect to multiple external devices. The Bluetooth wireless transmission module transmits data to other devices wirelessly, facilitating remote monitoring and data storage. These five modules work together to complete the various functions of the system.

[0017] A mixture of PDMS solution and curing agent is applied evenly to the surface of the flexible copper circuit again, and the flexible copper circuit is placed on a heating table to cure for 2 hours. After completion, the back-end unit of the sensor is formed. The front-end unit and the back-end unit of the sensor are connected using flexible ribbon cables to complete the assembly of the flexible multi-scale impedance sensor.

[0018] Preferably, the detection process of the activity detection system includes: The RGB acquisition module and the infrared acquisition module are fixed at the center of the black box of the grading line, the flexible multi-scale impedance sensor is fixed on the robotic arm module, and the fish to be tested are placed on the grading line conveyor belt in sequence. The host computer module triggers the RGB acquisition module, infrared acquisition module and flexible multi-scale impedance sensor to synchronously acquire RGB parameters, infrared parameters and impedance parameters at preset fixed time intervals. The acquired RGB parameters, infrared parameters, and impedance parameters are input into the adaptive mode module, and the RGB mode, infrared mode, impedance mode, RGB-infrared fusion mode, RGB-impedance fusion mode, infrared-impedance fusion mode, or total fusion mode are selected according to the signal quality of the parameters. The features of the selected modality are input into the multimodal fusion association classification model to obtain the fish activity classification results.

[0019] Preferably, the method for constructing the multimodal fusion association classification model includes: Multiple experimental samples were obtained by randomly grouping live fish purchased from the same batch, subjecting them to waterless low-temperature dormancy and live transport treatment; The flexible multi-scale impedance sensor, the RGB image acquisition module, and the infrared image acquisition module are used to acquire raw impedance datasets, raw RGB datasets, and raw infrared datasets at fixed time intervals, and store them in the knowledge base module. For each set of collected data, the standard parameters of fish respiratory rate, respiratory intensity, blood glucose, cortisol, color, and mucus were measured simultaneously to form an activity standard dataset. The corresponding RGB, infrared, and impedance parameters were then used to form an activity prediction dataset. A fish activity grading fusion network is trained based on the activity prediction dataset and the activity standard dataset. The network includes, in sequence, an RGB-infrared dual-modal information fusion network layer, an impedance information separation network layer, and an RGB-infrared-impedance tri-modal information fusion network layer. After determining the parameters of each network layer using cross-validation, the trained multimodal fusion association classification model is output.

[0020] Preferably, the RGB-infrared dual-modal information fusion network layer includes: Dimensionality reduction feature fusion layer, used to perform dimensionality reduction functions on RGB and infrared features. Extract dimensionality reduction features; The dimension-upgrading feature fusion layer is used to perform dimension-upgrading functions on RGB and infrared features. Spatial alignment is then performed to obtain upgraded features; The weighted feature fusion layer is used to linearly combine the dimensionality-reduced and dimensionality-upgraded features according to weighting coefficients α and β to obtain dual-modal fusion features. ; The impedance information separation network layer includes: The time-series feature extraction layer is used to extract time-series features from impedance signals. ; Spatial feature extraction layer, used to extract spatial distribution features from impedance signals. ; Linear interpolation fusion layer, used to interpolate functions Will and Integration into a unified representation ; The RGB-infrared-impedance three-mode information fusion network layer includes: The feature fusion layer is used to process the dual-modal fused features and the unified representation through a fusion network function once. and secondary fusion network function Processed and weighted by factors , fusion results obtained ; An activity prediction layer is used to perform an activation function on the fusion result. And output activity prediction value ;in, For activity prediction results, For activation function, As a weighting factor, For bias terms; An adaptive control mechanism layer is used to dynamically adjust the error according to the adaptive mechanism. Activity prediction results Dynamic correction is performed to obtain the corrected predicted value. ; An error feedback layer is used to dynamically adjust the error of the adaptive mechanism. Feedback is sent to each fusion network function to continuously update the model parameters.

[0021] Preferably, the activity prediction dataset is defined as follows: ; in, Represents the RGB image features of the i-th sample Infrared image features and impedance signal characteristics ; The active reference dataset is defined as follows: ; in, It is the fish activity level label corresponding to sample i, with activity levels ranging from 1 to 5, each representing a different activity state.

[0022] Preferably, the processing flow of the activity prediction dataset is as follows: Each modality in the activity prediction dataset is standardized using the following formula: ; in, The RGB values ​​represent the results of standardization processing based on the RGB data collected by each sensor. The mean of RGB values. The standard deviation of RGB; This represents the measured value from the infrared sensor, obtained through standardization. This is the average value of the infrared measurements. The standard deviation of infrared radiation. This represents the impedance measurement value, obtained through standardization. The average impedance. The standard deviation of impedance, The coefficients used to remove outliers are calculated using standardized coefficients and are used to measure the impact of outlier data on the model. This represents the new value after standardization, which is the data value obtained after removing outliers and standardization. Use a mapping function to map the model's predicted output. The mapping function is mapped to discrete activity levels. Used to convert predicted values ​​into corresponding activity levels of 1-5; the expression of the mapping function is: .

[0023] The present invention discloses the following technical effects: This invention overcomes the damage caused by conventional detection techniques by employing a multimodal sensing system, achieving non-destructive and accurate detection of live fish activity. Through data sampling using a flexible multi-scale modal sensor, an RGB image acquisition module, and an infrared acquisition module, it solves the problems of data source acquisition and the accuracy of acquiring different signals under different environments, resulting in richer and more accurate classification data. The multimodal information fusion network model addresses the inadequacy of single-modal methods in assessing live fish activity, achieving comprehensive and accurate perception of live fish activity. Finally, by utilizing a Bluetooth wireless transmission module to connect to a host computer, it solves the problem of real-time monitoring of live fish activity, enabling real-time monitoring and visualization of live fish activity. Attached Figure Description

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

[0025] Figure 1 A flowchart of the method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the fish activity detection system provided in an embodiment of the present invention; Figure 3 A schematic diagram of a flexible multi-scale sensor provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the principle of flexible multi-scale impedance measurement provided in an embodiment of the present invention. Figure 5 This is a schematic diagram of the system module structure provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the non-destructive testing method for fish activity provided in an embodiment of the present invention; Figure 7 A schematic diagram illustrating the method for constructing a fish activity detection model provided in an embodiment of the present invention; Figure 8 A schematic diagram illustrating the construction process of the fish activity hierarchical fusion network provided in an embodiment of the present invention; Figure 9A flowchart for fish activity prediction and grading provided in an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a method for detecting fish activity during live transport, comprising: Step 100: Conduct live transport of fish according to the live transport environment, and simultaneously collect fish activity standard parameters, activity prediction parameters, and activity influencing parameters during the transport process; among them, the activity standard parameters include respiratory rate standard parameters, respiratory intensity standard parameters, blood glucose parameters, cortisol parameters, color standard parameters, and mucus standard parameters; the activity prediction parameters include infrared parameters, RGB parameters, and impedance parameters; the activity influencing parameters include transport time, as well as adjustable temperature parameters, humidity parameters, and oxygen parameters; Step 200: Using the datasets corresponding to the activity standard parameters, activity prediction parameters, and activity impact parameters obtained within a single synchronous acquisition cycle as samples, classify fish activity based on the activity standard parameters in the samples, obtain an activity knowledge base, and label each sample with an activity tag. Step 300: Based on the classification results of the activity knowledge base, the activity prediction parameters and activity labels are fused and associated to form a multimodal fusion association classification model. Step 400: Combine the classification results of the multimodal fusion association classification model with the transportation time to calculate the fish activity index, and calculate the cumulative temperature parameter, cumulative humidity parameter, and cumulative oxygen parameter based on the temperature parameter, humidity parameter, and oxygen parameter, and determine the fish activity threshold based on the transportation time. Step 500: Input the activity index and activity threshold into the adaptive feedback adjustment module to implement dynamic closed-loop adjustment of temperature, humidity and oxygen parameters in the survival transport environment, and provide real-time feedback of the adjustment results.

[0029] Figure 2 This is a schematic diagram of the fish activity detection system provided in an embodiment of the present invention; as shown below. Figure 2As shown, an activity detection system for fish survival processes is disclosed. The system includes a flexible multi-scale impedance sensor, an RGB image acquisition module, an infrared image acquisition module, a grading system module, and an operating host computer module.

[0030] The flexible multi-scale impedance sensor is attached to the surface of the fish to acquire real-time data on the fish's physiological state. The RGB image acquisition module is used to acquire images of the fish's body surface to assess the fish's external health status. The infrared image acquisition module is used to collect temperature changes in the fish to assess its physiological activity. The grading system module is used to process, analyze, and classify the data acquired by the flexible multi-scale impedance sensor, RGB image acquisition module, and infrared image acquisition module, and to classify fish into different active states. The host computer module is used to centrally control and manage the operation of all sub-modules, display sensor data in real time, provide data storage and historical query functions, and connect to the remote monitoring system via wireless transmission for remote control and early warning. Among them, the flexible multi-scale impedance sensor (reference) Figure 3 The device includes a sensing array with four electrodes, each containing a base layer (PDMS) and a sensitive layer. The sensitive layer of each electrode has a LIG sensing medium. All four electrodes are connected to an integrated circuit board. An impedance conversion chip on the circuit board excites the fish body through a specific electrical signal, and collects the impedance change data of the target fish body surface in real time to reflect its internal physiological changes. Specifically, the PDMS thickness of the flexible unit at the sensor front end is 2mm-4mm, and can be 3mm. The diameter of the LIG circular electrode is 5cm, and the thickness is 0.05mm.

[0031] Optionally, in the integrated circuit board, the thickness of the PI flexible substrate is 0.10 mm to 0.20 mm, specifically, it can be 0.15 mm; The thickness of the flexible Cu circuit is one of 0.015mm, 0.25mm or 0.35mm, for example, it can be 0.25mm; The thickness of PDMS can cover the height of all electronic components.

[0032] The sensor back-end unit measures 35mm x 30mm.

[0033] Impedance data measurement principle reference Figure 4Impedance measurements, including amplitude and phase angle, were taken at frequencies of 50 kHz, 30 kHz, 20 kHz, 10 kHz, 5 kHz, 3 kHz, 2 kHz, 1 kHz, 500 Hz, 300 Hz, 200 Hz, and 100 Hz to ensure comprehensive coverage of both high and low frequencies. A total of 12 impedance measurements were performed on the four electrodes, and the final impedance value was calculated as the average of the 12 measurements. During each measurement, two electrodes served as both the working electrode and the counter electrode for each other.

[0034] The fabrication process of the flexible multi-scale impedance sensor includes: Import the pre-designed electrode pattern into the laser direct writing device; Laser-induced graphene (LIG) was fabricated on a flexible substrate using a laser direct writing device, forming a circular LIG electrode structure with four arrays. A mixture of PDMS solution and curing agent was uniformly coated on the electrode surface by spin coating. The substrate coated with the solution was then placed on a heating table at 60°C for curing for 2 hours, thereby forming a uniform PDMS protective layer. The cured substrate film and LIG circular electrode are transferred to a flexible substrate through a transfer process. The transferred film is then cut into four identical flexible units to obtain the front-end unit of the sensor.

[0035] The pre-designed circuit pattern is imported into a laser direct writing device, and laser etching is performed on the PI / Cu composite film to form electrical connection lines on the flexible circuit board, thus obtaining a flexible copper circuit.

[0036] After the circuit fabrication is completed, components such as the main control module, voltage conversion module, impedance acquisition module, I2C multiplexing module, and Bluetooth wireless transmission module are fixed to the surface of the flexible Cu circuit using a soldering process. Next, a mixture of PDMS solution and curing agent is applied evenly to the surface of the flexible circuit again, and it is placed on a heating stage to cure for 2 hours, thus forming the back-end unit of the sensor.

[0037] The front-end and back-end units of the sensor are connected using flexible ribbon cables to complete the assembly of the flexible multi-scale impedance sensor.

[0038] The RGB image acquisition module uses a high-resolution camera and image processing algorithms to acquire information such as the color and shape of the fish's surface and analyze the fish's external health status; the infrared image acquisition module uses an infrared thermal imager to monitor the temperature distribution of the fish's body and reflect the fish's thermal response under different environmental conditions in real time. refer to Figure 5The system's workflow includes: inputting signals through front-end electrodes and processing the data through impedance conversion circuits and impedance sensing modules; acquiring images with the assistance of supplementary LEDs using RGB and infrared cameras, and storing the data in a flash memory module; subsequently, uploading the data to a data knowledge base and classifying and evaluating it through a hierarchical system module; real-time monitoring and issuing alarms by an early warning module, displaying the data on the host computer, and wireless data transmission via a Bluetooth module.

[0039] Preferably, the hierarchical system module includes a communication module, allowing RGB and infrared image data to be transmitted to the host computer via different communication modules. RGB image data is acquired by the camera module through a high-bandwidth interface, such as USB or Ethernet, then compressed or formatted by the data processing unit before being transmitted to the host computer via USB or Ethernet. Infrared image data is acquired by the infrared sensor module, undergoes signal processing and format conversion, and is finally transmitted to the host computer via the wireless communication module. During transmission, the image data is encoded and compressed according to the transmission protocol to improve transmission efficiency and reduce bandwidth consumption. The host computer receives and processes this data through a decoding module, ultimately presenting it as an image or video.

[0040] Preferably, to ensure the stability and accuracy of data transmission, a verification and error correction mechanism is added to the image data transmission, while ensuring the high efficiency and stability of the communication interface: Preferably, the main control module uses an STM32 microcontroller to coordinate the operation of each module and control the data acquisition and processing of the sensors. A crystal oscillator provides the clock signal to the microcontroller, while resistors and capacitors are used for power supply decoupling, signal filtering, and circuit stabilization. The voltage conversion module uses a DC-DC buck-boost converter to convert the external power supply voltage into a stable voltage suitable for each module, and a voltage regulator ensures stable system operation. The impedance acquisition module amplifies the signal from the sensor front end using an operational amplifier, and an ADC converts the analog signal into a digital signal for processing by the main control module. A low-noise amplifier further reduces signal noise to ensure accuracy. The I2C multiplexing module expands the I2C bus, ensuring that multiple devices can communicate via the bus. The I2C expander helps achieve this function and regulates the signal stability of the bus using resistors and capacitors. Finally, the Bluetooth wireless transmission module transmits the processed data to external devices, such as smartphones or computers, via Bluetooth, and the Bluetooth antenna ensures stable signal transmission. The front end of the entire system consists of LIG electrodes and a PDMS protective layer, responsible for acquiring external signals, while the back end processes, converts, and transmits data through multiple modules. The various modules of the system are connected by flexible cables to ensure smooth signal transmission and coordinated operation between modules, thereby enabling the sensor to operate efficiently and stably.

[0041] Preferably, the grading system module automatically assesses the activity of the fish based on sensor data, and uses a multimodal fusion network model to classify the data, thereby achieving accurate identification of the health status of different fish species. The host computer module displays real-time data of fish activity detection to the user through a human-machine interface, and can adjust operating parameters based on system feedback to ensure the maximum survival rate of fish during the survival process.

[0042] refer to Figure 6 A non-destructive testing method for fish activity, the testing method comprising: The RGB acquisition module and infrared acquisition module are fixed in the center of the black box of the grading line, the flexible multi-scale impedance sensor is fixed on the robotic arm module, and the fish to be tested are placed one by one on the grading line conveyor belt. The host computer module, RGB acquisition module, infrared acquisition module, and flexible multi-scale impedance sensor acquire RGB signals, infrared signals, and impedance signals at fixed time intervals. The working mode is selected through an adaptive modal module; the working modes include: RGB mode, infrared mode, impedance mode, RGB infrared fusion module, RGB impedance fusion module, infrared impedance fusion module, and overall fusion module; The RGB signal, infrared signal, and impedance signal are input into the fish activity fusion and correlation classification model to detect the activity of the fish and obtain the activity classification result of the fish to be tested.

[0043] A method for constructing a fish activity detection model, referring to... Figure 7 The construction method includes: Several experimental samples were obtained by randomly grouping live fish purchased from the same batch, subjecting them to waterless low-temperature dormancy and live transport treatment; The flexible multi-scale impedance sensor, RGB acquisition module and IR acquisition module are used to acquire data at fixed time intervals to obtain the original RGB dataset, IR dataset and impedance dataset of the experiment, and the original RGB dataset, IR dataset and impedance dataset of the experiment are saved to the preset knowledge base module. The specific dataset is , , and These represent RGB image data, infrared image data, and impedance data acquired at different times, respectively.

[0044] For each group of experimental samples, respiratory rate, respiratory intensity, blood glucose, cortisol, color grade, and mucus content were measured at fixed intervals to obtain fish activity characterization datasets and fish activity classification datasets, which were then saved to the knowledge base module; the specific datasets are as follows:

[0045]

[0046] in, , , , , and These are the specific values ​​for respiratory rate, respiratory intensity, blood glucose, cortisol, color grade, and mucus level.

[0047] For the data collected from each group within the RGB raw dataset, IR raw dataset, and impedance raw dataset of the experiment, fish activity standard parameters were simultaneously collected to obtain a fish activity prediction dataset and a fish activity reference dataset; the activity prediction dataset is as follows: in, Represents the RGB image features of the i-th sample Infrared image features and impedance signal characteristics .

[0048] The active reference dataset is defined as follows: in, It is the fish activity level label corresponding to sample i, with activity levels ranging from 1 to 5, each representing a different activity state.

[0049] Standardize each mode in the input data, assuming Let RGB, infrared, and impedance eigenvectors be the eigenvectors, respectively. Then the normalization formula is:

[0050] in, and Here, represents the mean and standard deviation, respectively. W is a function for detecting and removing outliers. To replace the median of outliers. Based on the model's predicted output. This maps the predicted results to the fish's activity level. Assuming the model's output is continuous, a threshold is used to map it to discrete activity levels. A mapping function is defined. This function converts the predicted values ​​into corresponding activity levels of 1-5:

[0051] A method for determining a fish activity index and an activity threshold, comprising: determining the fish activity index during live transport based on the classification results obtained from the fusion association classification model and the transport time; determining the fish activity threshold based on the environmental cumulative parameters and the transport time according to claim 1; and setting the classification function for fish activity as follows:

[0052] S0, S1, S2, S3, and S4 represent the thresholds for very low, low, medium, high, and very high activity of the activity parameter, respectively.

[0053] The activity index At of a group of fish is predicted based on the activity S of individual fish during transportation and the transportation time t.

[0054] Where At is the initial activity index at the start of transportation, and t is the transportation time. , and These are the weighting coefficients.

[0055] During transportation, environmental factors can affect the activity of fish. An environmental cumulative parameter is defined to represent the combined impact of these environmental factors during transportation.

[0056] The cumulative parameters of temperature, humidity, and oxygen concentration are calculated using the following formula:

[0057] in, It is an intermediate variable in the transportation time. , and They are time points Temperature, humidity, and oxygen concentration at that time.

[0058] Activity threshold The activity index of fish during transportation Cumulative parameters with environment The balance point between these two factors is calculated using the following formula to determine the activity threshold:

[0059] in , and These are the weighting coefficients of environmental parameters, used to represent the effects of temperature, humidity, and oxygen concentration on the activity threshold.

[0060] The adaptive regulation mechanism layer includes automatically adjusting environmental conditions during transportation based on changes in the fish activity index A(t) and environmental parameters T(t), H(t), and O(t), and the difference between the activity index and the activity threshold. It is a key input for feedback regulation, representing the impact of the current environment on fish activity:

[0061] Temperature regulation Calculations based on activity differences are used to ensure fish are within their optimal temperature range:

[0062] in, , and These are the proportional, integral, and derivative control coefficients for temperature regulation.

[0063] Humidity regulation Calculated based on activity differences to ensure fish are within the optimal humidity range:

[0064] in, , and These are the proportional, integral, and derivative control coefficients for humidity regulation.

[0065] oxygen regulation Calculations based on activity differences are used to ensure fish are within the optimal oxygen concentration range:

[0066] in, , and These are the proportional, integral, and derivative control coefficients for oxygen concentration adjustment.

[0067] Comprehensive adjustment amount Taking into account the combined effects of temperature, humidity, and oxygen concentration, the overall effect of adjusting environmental parameters is:

[0068] in, , and These are the weighting coefficients for temperature, humidity, and oxygen concentration regulation.

[0069] After each adjustment, the new environmental parameters are updated using the following formula:

[0070] Adjust the regulation interval according to the rate of change in fish activity. To make the system more sensitive or more stable:

[0071] The error feedback layer provides feedback on the prediction anomalies from the active prediction layer and makes dynamic adjustments to continuously improve model performance.

[0072]

[0073] in, This represents the total loss of the error message feedback layer; This is the prediction error of the activity prediction layer; It is the overfitting loss function; It is the variance loss function; and These are the weighting coefficients for overfitting and the variance loss function, balancing the effects of various losses; It is the learning rate, which controls the degree of influence of the feedback layer on the model; It is the gradient of the feedback weight coefficients, representing the process by which the model adjusts based on feedback information.

[0074] For the data collected from each group within the RGB raw dataset, IR raw dataset, and impedance raw dataset of the experiment, fish activity standard parameters were simultaneously collected to obtain a fish activity prediction dataset and a fish activity reference dataset. The fish activity standard parameters include the activity reference dataset; the fish activity standard parameters include respiratory standard parameters, biochemical standard parameters, and appearance standard parameters; the respiratory standard parameters include respiratory rate standard parameters and respiratory intensity standard parameters; the biochemical standard parameters include blood glucose parameters and cortisol parameters; the appearance standard parameters include color standard parameters and mucus standard parameters; the activity prediction parameters include the fish's infrared parameters, RGB parameters, and impedance parameters. A fish activity hierarchical fusion network is constructed based on the fish activity prediction dataset and the activity reference dataset; the fish activity hierarchical fusion network includes: an RGB infrared dual-mode information fusion network layer, an impedance information separation network layer, and an RGB infrared impedance tri-mode information fusion network layer; The RGB infrared dual-modal information fusion network layer performs fish activity detection on the RGB infrared signals collected in the RGB infrared mode; The impedance information separation network layer performs fish activity detection on the impedance signal under the impedance mode. The RGB infrared impedance three-mode information fusion network layer performs fish activity detection on the RGB signal, infrared signal and impedance signal collected under the fusion mode; The RGB and infrared dual-modal information fusion network layer includes a dimensionality reduction feature fusion layer, a dimensionality increase feature fusion layer, and a weighted feature fusion layer. The dimensionality reduction feature fusion layer performs linear matrix feature extraction and fusion on RGB features and infrared features. The dimensionality increase feature fusion layer performs aligned pixel extraction and fusion on RGB features and infrared features. The weighted feature fusion layer performs equal-ratio weighted fusion on the extracted features of the dimensionality reduction feature fusion layer and the dimensionality increase feature fusion layer. The impedance information separation network layer includes a time-series feature extraction layer, a spatial feature extraction layer, and a linear interpolation fusion layer. The time-series feature extraction layer includes performing time-series feature extraction on the impedance signal. The spatial feature extraction layer includes performing spatial distribution feature extraction on the impedance signal. The linear interpolation fusion layer includes fusing the spatial and time features of the impedance signal by linear interpolation. The RGB, infrared, and impedance triple-modal information fusion network layer consists of a feature extraction layer, a feature fusion layer, an activity prediction layer, an adaptive regulation mechanism layer, and an error information feedback layer. The feature extraction layer is responsible for further extracting features from RGB, infrared, and impedance feature data respectively. The feature association layer is used for weighted fusion of the feature information extracted by the feature extraction layer. The activity prediction layer is used for predicting the activity state of fish based on the fused result features of the feature fusion layer. The adaptive regulation mechanism layer is used for dynamically adjusting the environmental conditions during the transportation process. The error information feedback layer is used for feeding back the prediction abnormal information of the activity prediction layer and making dynamic adjustments to continuously improve the model performance.

[0075] Preferably, the calculation formula for the adaptive modal pattern adjustment is as follows:

[0076] where , and are the signal quality ratios of each modality, and their calculation formula is:

[0077] QRGB, QIR, and QZ are the signal qualities of each modality, representing the signal qualities of the RGB, infrared, and impedance modalities.

[0078] When QRGB < TRGB, the signal of the RGB modality is considered interference, and the infrared image modality and the impedance modality are selected as the model inputs.

[0079] When QIR < TIR, the signal of the IR modality is considered interference, and the RGB image modality and the impedance modality are selected as the model inputs.

[0080] When QIMP < TIMP, the signal of the IMP mode is considered as interference, and the RGB image mode and the infrared image mode are selected as the model inputs. T is the interference threshold for each mode.

[0081] Taking the classification of sturgeon activity as an example, a specific multi-modal network architecture is given. Refer to Figure 8 , for the data of the RGB and infrared channels, an early fusion method is adopted for processing. The RGB-infrared dual-modal information fusion network layer linearly extracts and fuses the RGB and infrared features through a dimensionality reduction feature fusion layer, a dimensionality increase feature fusion layer, and a weighted feature fusion layer, so as to integrate them into an input format of 6×224×224, aiming to reduce the model parameters and improve the calculation speed.

[0082] The impedance data adopts a mid-term and late-stage parallel processing method. First, the impedance data is subjected to feature extraction through a single-layer fully connected network to capture the key information in the data. Next, the feature maps of the amplitude and phase angle are obtained through matrix multiplication, and this step is crucial for understanding the multi-dimensional characteristics of the data. To optimize the calculation performance, only 20 neurons are retained in the fully connected layer, effectively reducing the calculation complexity, and finally generating a 1×20×20 feature map. On this basis, the bilinear interpolation method is used to expand the size of the feature map to 1×224×224. This expansion ensures that the feature map can be seamlessly docked with the feature map after the RGB and IR fusion (8×224×224), so as to obtain an integrated feature map of 9×224×224. This integrated feature map can effectively combine the information carried by the multi-modal data.

[0083] The impedance data (amplitude and phase angle) first passes through a fully connected layer to output the prediction result, and then is combined with the output result of the backbone network module. The purpose of this process is to improve the accuracy of the model output and ensure the full exchange of information. Finally, these results are input into a fully connected layer to obtain the final classification result. After adjusting the number of channels, the features of fish pressure are extracted from the fused data to form a 1×5 decision vector. In this way, after the amplitude and phase angle features are processed by independent fully connected layers, they are integrated, and the final output is a 1×5 decision vector. To ensure the accuracy of classification, a classifier is used to obtain the final classification result and clarify the probability distribution of each category. To effectively prevent the occurrence of overfitting, a fitting reduction layer is added at the end of the fully connected layer, and its probability coefficient is set to 0.7 to randomly discard some neurons, thereby enhancing the generalization ability of the model. This network design can successfully achieve the efficient fusion of multi-modal data and the extraction of features, significantly improving the accuracy and reliability of the classification task.

[0084] Refer to Figure 9 , this embodiment also provides a more specific method for classifying fish activity. The specific method includes: 1) Number the live fish to be tested, and start the multimodal sensing system 1 corresponding to the number by operating the host computer 2; 2) Multimodal sensing system 1 System initialization: Determine if the sensing system is working properly; if not, restart the system for adjustment; if it still does not work properly after adjustment, issue an alarm to remind manual inspection. 3) The multimodal sensing system 1 collects RGB signals, infrared signals, and impedance signals from live fish, and collects multiple times within the same period of time to ensure data stability; it compares the RGB signals, infrared signals, and impedance signals with the knowledge base to determine whether the collected signals are abnormal data; if they are unusable, it re-collects; if the re-collected signals are still abnormal data, it alarms and conducts manual inspection. Prioritize the simultaneous application of the three modalities to more accurately determine the activity of the live fish; in single-modal mode, a basic activity judgment can be made on the live fish, which is suitable for working environments that cannot simultaneously meet the requirements of the three modalities and where rapid screening of the activity of live fish is required. In the fusion mode, the activity of live fish can be accurately predicted, making it suitable for high-end restaurants and scientific research.

[0085] The database includes: RGB signals, infrared signals, and impedance signals of live fish collected during the experiment. , The activity standards and activity levels of the live fish collected in the experiment are as follows:

[0086]

[0087] 4) Determine if the RGB, infrared, and impedance signals are affected by environmental interference. If none are affected, enable the fusion mode; if the impedance signal is affected, enable the RGB and infrared modes; if the RGB and infrared signals are affected, enable the impedance mode; if the infrared signal is affected, enable the RGB and impedance modes; if the RGB signal is affected, enable the impedance and infrared modes; if the RGB and impedance signals are affected, enable the infrared mode. Specifically, environmental interference includes: when the sensor is in an environment with high light intensity, the data from the RGB acquisition module and the infrared acquisition module will be affected, causing the sensor to malfunction; when the sensor is in an environment with poor temperature and humidity, the multi-scale impedance sensor will be affected, causing abnormal data. The specific formula for judging signal interference is as follows: if If the signal is disturbed, it means that the signal has been interfered with.

[0088] 5) After selecting the chosen modality, import the data into the fish activity classification model for data preprocessing.

[0089] 6) If it is an impedance mode, the data is input to the impedance information separation network layer and the information fusion network layer; If it is RGB mode, IR mode, or RGB and IR mode, then the data is input to the RGB infrared dual-mode information fusion network layer; If the data is in RGB and impedance modes, IR and impedance modes, or RGB, IR and impedance modes, then the data is input into the RGB infrared dual-mode information fusion network, the impedance information separation network layer, and the information fusion network layer. The specific calculation formula includes:

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097] An adaptive control mechanism layer is introduced into the network, with specific formulas including:

[0098] An error feedback layer is introduced into the network, and the specific formula includes:

[0099] 7) Weight the output of the information fusion network layer to output the activity level of the live fish. The specific formula is as follows: ,

[0100] in, This represents the total activity score. As for the activity level, These are specific activity scores for respiratory rate, respiratory intensity, blood glucose, cortisol, color grade, and mucus level. The value can be one of (0, 0.25, 0.5, 0.75, 1), and all scoring criteria rely on the knowledge base module for scoring.

[0101] As an optional implementation, this embodiment also provides an activity detection system for the live transport process of fish, characterized in that it includes: A flexible multi-scale impedance sensor is attached to the surface of a fish to collect real-time data on changes in the fish's surface impedance and output impedance parameters. The RGB image acquisition module is used to acquire color images of the fish's surface and output RGB parameters. The infrared image acquisition module is used to acquire images of the temperature distribution on the fish body and output infrared parameters. The environmental parameter acquisition module is used to collect temperature, humidity, and oxygen parameters in the live transport environment. The grading system module is used to preprocess and extract features from the raw data from the flexible multi-scale impedance sensor, the RGB image acquisition module, and the infrared image acquisition module. Based on the extracted standard parameters of respiratory frequency, respiratory intensity, blood glucose, cortisol, color, and mucus, it constructs a set of activity standard parameters. The impedance parameters, RGB parameters, and infrared parameters are used as a set of activity prediction parameters. The temperature, humidity, and oxygen parameters collected by the environmental parameter acquisition module, as well as the transportation time recorded by the host computer module, are used to form a set of activity influencing parameters. A multimodal fusion network model is used to process, analyze, and classify the activity influencing parameters and the activity standard parameters to classify the fish into different activity states. The host computer module is used to centrally control and manage the operation of the flexible multi-scale impedance sensor, the RGB image acquisition module, the infrared image acquisition module, the environmental parameter acquisition module, and the classification system module. It records transportation time, displays sensor and classification result data in real time, provides data storage and historical query functions, and connects to a remote monitoring system via wireless transmission for remote control and early warning.

[0102] The beneficial effects of this invention are as follows: This invention achieves non-destructive and accurate detection of fish activity through a multimodal sensing system; it achieves rich and accurate classification data through data sampling using a flexible impedance-RGB-IR sensing system; it achieves comprehensive and accurate perception of live fish activity under different working environments and scenarios through a live fish activity classification model; and it achieves real-time monitoring and visualization of live fish activity by connecting to an operating host computer using a Bluetooth wireless transmission module.

[0103] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0104] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for detecting fish activity during live transport, characterized in that, include: Fish are transported alive according to the preservation and transport environment, and fish activity standard parameters, activity prediction parameters, and activity influencing parameters are collected simultaneously during the transport process. Among them, the activity standard parameters include respiratory rate standard parameters, respiratory intensity standard parameters, blood glucose parameters, cortisol parameters, color standard parameters, and mucus standard parameters; the activity prediction parameters include infrared parameters, RGB parameters, and impedance parameters; the activity influencing parameters include transport time, as well as adjustable temperature parameters, humidity parameters, and oxygen parameters. Using the datasets corresponding to the activity standard parameters, activity prediction parameters, and activity impact parameters obtained within a single synchronous acquisition cycle as samples, fish activity is classified according to the activity standard parameters in the samples to obtain an activity knowledge base, and each sample is labeled with an activity tag. Based on the classification results of the activity knowledge base, the activity prediction parameters and the activity tags are fused and associated to form a multimodal fusion association classification model. The classification results of the multimodal fusion association classification model are combined with the transportation time to calculate the fish activity index. The cumulative temperature parameter, cumulative humidity parameter, and cumulative oxygen parameter are calculated based on the temperature parameter, the humidity parameter, and the oxygen parameter. The fish activity threshold is determined in combination with the transportation time. The activity index and the activity threshold are input into the adaptive feedback adjustment module to implement dynamic closed-loop adjustment of the temperature, humidity and oxygen parameters in the survival transport environment, and to provide real-time feedback of the adjustment results.

2. The method for detecting fish activity during live transport according to claim 1, characterized in that, The classification results of the multimodal fusion association classification model are combined with the transportation time to calculate the fish activity index. Furthermore, cumulative temperature, cumulative humidity, and cumulative oxygen parameters are calculated based on the temperature, humidity, and oxygen parameters. Finally, the fish activity threshold is determined in conjunction with the transportation time, including: The classification function for fish activity is set as follows: Wherein, S0, S1, S2, S3 and S4 represent the thresholds of very low, low, medium, high and very high activity of the activity parameter, respectively. The activity index of a group of fish is predicted based on the activity S of individual fish during transportation and the transportation time t. The formula for calculating the activity index is as follows: ;in, A t It is the initial activity index at the start of transportation, and t is the transportation time. , and All are weighting coefficients, where n is the dimension involved in the activity parameter, and C is a constant of the health-related indicator; Through formula Calculate the cumulative temperature parameter, cumulative humidity parameter, and cumulative oxygen parameter; among which, It is an intermediate variable in the transportation time. , and They are time points Temperature, humidity, and oxygen concentration at that time; These are accumulated parameters for temperature, humidity, or oxygen. Through formula Calculate the fish activity threshold; wherein, , and These are the weighting coefficients for environmental parameters, used to represent the influence of temperature, humidity, and oxygen concentration on the activity threshold. E represents the fish activity threshold. T (t) represents the temperature-related influencing factor, indicating the effect of temperature on fish activity value; it is a function that changes over time. H (t) represents the humidity-related influencing factor, indicating the effect of humidity on fish activity values, which changes over time; E o (t) represents the influencing factor related to oxygen concentration, indicating the effect of oxygen concentration on fish activity value, which changes over time; E(t) is the result of a weighted combination, which integrates the effects of temperature, humidity and oxygen concentration on fish activity value.

3. The method for detecting fish activity during live transport according to claim 2, characterized in that, The method for constructing the adaptive feedback adjustment module is as follows: Based on the current population's activity index At and activity threshold Calculate the activity bias; The formula for calculating activity bias is: ;in, This refers to the activity deviation; According to the formula Calculate the temperature regulation amount; where, , and These are the proportional, integral, and derivative control coefficients for temperature regulation. The temperature adjustment amount; According to the formula Calculate the humidity adjustment amount; where, , and These are the proportional, integral, and derivative control coefficients for humidity regulation. The humidity adjustment amount; According to the formula Calculate oxygen regulation ;in, , and These are the proportional, integral, and derivative control coefficients for humidity regulation. The oxygen regulation amount; These represent the coefficients of the humidity control, temperature control, and differential control systems, respectively, and represent the weights of the influence of different control processes on oxygen regulation. According to the formula Calculate the overall adjustment amount; where, This refers to the overall adjustment amount; These represent the degree of influence of temperature, humidity, and oxygen on the overall regulation amount, respectively. After adjusting according to the aforementioned comprehensive adjustment amount, the formula is used. Update temperature, humidity, and oxygen concentration; The current temperature adjustment amount is indicated by calculating the current temperature. and temperature change The obtained temperature value; The current humidity adjustment amount is indicated by calculating the current humidity. and humidity change The obtained humidity value; This indicates the current oxygen adjustment amount, based on the current oxygen level. and changes in oxygen Calculated; The adjustment interval is adjusted according to the rate of change in fish activity; the formula for the adjustment interval is: ;in, The adjustment interval is... ϑ 1 represents the adjustment sensitivity coefficient, which characterizes the degree of influence of changes in fish activity on the adjustment interval. The larger the value, the more significant the effect of changes in activity on shortening the interval. ϑ 2 represents the baseline adjustment weight, indicating the basic adjustment frequency that needs to be maintained even when the fish activity changes little or even close to zero, in order to prevent the system from losing feedback regulation.

4. A fish activity detection system for the live transport process, characterized in that, include: A flexible multi-scale impedance sensor is attached to the surface of a fish to collect real-time data on changes in the fish's surface impedance and output impedance parameters. The RGB image acquisition module is used to acquire color images of the fish's surface and output RGB parameters. The infrared image acquisition module is used to acquire images of the temperature distribution on the fish body and output infrared parameters. The environmental parameter acquisition module is used to collect temperature, humidity, and oxygen parameters in the live transport environment. The grading system module is used to preprocess and extract features from the raw data from the flexible multi-scale impedance sensor, the RGB image acquisition module, and the infrared image acquisition module. Based on the extracted standard parameters of respiratory frequency, respiratory intensity, blood glucose, cortisol, color, and mucus, it constructs a set of activity standard parameters. The impedance parameters, RGB parameters, and infrared parameters are used as a set of activity prediction parameters. The temperature, humidity, and oxygen parameters collected by the environmental parameter acquisition module, as well as the transportation time recorded by the host computer module, are used to form a set of activity influencing parameters. A multimodal fusion network model is used to process, analyze, and classify the activity influencing parameters and the activity standard parameters to classify the fish into different activity states. The host computer module is used to centrally control and manage the operation of the flexible multi-scale impedance sensor, the RGB image acquisition module, the infrared image acquisition module, the environmental parameter acquisition module, and the classification system module. It records transportation time, displays sensor and classification result data in real time, provides data storage and historical query functions, and connects to a remote monitoring system via wireless transmission for remote control and early warning.

5. The fish activity detection system for the live transport process according to claim 4, characterized in that, The fabrication process of the flexible multi-scale impedance sensor includes: Import the pre-designed electrode pattern into the laser direct writing device; Laser-induced graphene was prepared on a flexible substrate using the laser direct writing device to form a circular LIG electrode structure with four arrays. A mixture of PDMS solution and curing agent is uniformly coated on the surface of the LIG electrode structure by spin coating. The substrate coated with the solution is then placed on a heating table at 60°C for curing for 2 hours, thereby forming a uniform PDMS protective layer. The cured substrate film and the LIG electrode structure are transferred to a flexible substrate through a transfer process. The transferred film is then cut into four identical flexible units to obtain the front-end unit of the sensor. The pre-designed circuit pattern is imported into the laser direct writing device, and laser etching is performed on the PI / Cu composite film to form electrical connection lines on the flexible circuit board, thus obtaining a flexible copper circuit. After the circuit is fabricated, the main control module, voltage conversion module, impedance acquisition module, I2C multiplexer module, and Bluetooth wireless transmission module are fixed to the surface of the flexible copper circuit using a soldering process. The main control module is responsible for coordinating the work of each part and processing data. The voltage conversion module provides a stable power supply to the system. The impedance acquisition module collects key data from the sensors in real time. The I2C multiplexer module expands the communication capability with external devices. The Bluetooth wireless transmission module enables wireless data transmission. A mixture of PDMS solution and curing agent is applied evenly to the surface of the flexible copper circuit again, and the flexible copper circuit is placed on a heating table to cure for 2 hours. After completion, the back-end unit of the sensor is formed. The front-end unit and the back-end unit of the sensor are connected using flexible ribbon cables to complete the assembly of the flexible multi-scale impedance sensor.

6. The fish activity detection system for the live transport process according to claim 4, characterized in that, The detection process of the activity detection system includes: The RGB acquisition module and the infrared acquisition module are fixed at the center of the black box of the grading line, the flexible multi-scale impedance sensor is fixed on the robotic arm module, and the fish to be tested are placed on the grading line conveyor belt in sequence. The host computer module triggers the RGB acquisition module, infrared acquisition module and flexible multi-scale impedance sensor to synchronously acquire RGB parameters, infrared parameters and impedance parameters at preset fixed time intervals. The acquired RGB parameters, infrared parameters, and impedance parameters are input into the adaptive mode module, and the RGB mode, infrared mode, impedance mode, RGB-infrared fusion mode, RGB-impedance fusion mode, infrared-impedance fusion mode, or total fusion mode are selected according to the signal quality of the parameters. The features of the selected modality are input into the multimodal fusion association classification model to obtain the fish activity classification results.

7. The fish activity detection system for the live transport process according to claim 4, characterized in that, The method for constructing the multimodal fusion association classification model includes: Multiple experimental samples were obtained by randomly grouping live fish purchased from the same batch, subjecting them to waterless low-temperature dormancy and live transport treatment; The flexible multi-scale impedance sensor, the RGB image acquisition module, and the infrared image acquisition module are used to acquire raw impedance datasets, raw RGB datasets, and raw infrared datasets at fixed time intervals, and store them in the knowledge base module. For each set of collected data, the standard parameters of fish respiratory rate, respiratory intensity, blood glucose, cortisol, color, and mucus were measured simultaneously to form an activity standard dataset. The corresponding RGB, infrared, and impedance parameters were then used to form an activity prediction dataset. A fish activity grading fusion network is trained based on the activity prediction dataset and the activity standard dataset. The network includes, in sequence, an RGB-infrared dual-modal information fusion network layer, an impedance information separation network layer, and an RGB-infrared-impedance tri-modal information fusion network layer. After determining the parameters of each network layer using cross-validation, the trained multimodal fusion association classification model is output.

8. The fish activity detection system for the live transport process according to claim 7, characterized in that, The RGB-infrared dual-modal information fusion network layer includes: Dimensionality reduction feature fusion layer, used to perform dimensionality reduction functions on RGB and infrared features. Extract dimensionality reduction features; The dimension-upgrading feature fusion layer is used to perform dimension-upgrading functions on RGB and infrared features. Spatial alignment is then performed to obtain upgraded features; The weighted feature fusion layer is used to linearly combine the dimensionality-reduced and dimensionality-upgraded features according to weighting coefficients α and β to obtain dual-modal fusion features. ; The impedance information separation network layer includes: The time-series feature extraction layer is used to extract time-series features from impedance signals. ; Spatial feature extraction layer, used to extract spatial distribution features from impedance signals. ; Linear interpolation fusion layer, used to interpolate functions Will and Integration into a unified representation ; The RGB-infrared-impedance three-mode information fusion network layer includes: The feature fusion layer is used to process the dual-modal fused features and the unified representation through a fusion network function once. and secondary fusion network function Processed and weighted by factors , fusion results obtained ; An activity prediction layer is used to perform an activation function on the fusion result. And output activity prediction value ;in, For activity prediction results, For activation function, As a weighting factor, For bias terms; An adaptive control mechanism layer is used to dynamically adjust the error according to the adaptive mechanism. Activity prediction results Dynamic correction is performed to obtain the corrected predicted value. ; An error feedback layer is used to dynamically adjust the error of the adaptive mechanism. Feedback is sent to each fusion network function to continuously update the model parameters.

9. The fish activity detection system for the live transport process according to claim 7, characterized in that, The definition of the activity prediction dataset is: ； in, Represents the RGB image features of the i-th sample Infrared image features and impedance signal characteristics ; The active reference dataset is defined as follows: ； in, It is the fish activity level label corresponding to sample i, with activity levels ranging from 1 to 5, each representing a different activity state.

10. The fish activity detection system for the live transport process according to claim 9, characterized in that, The processing flow of the activity prediction dataset is as follows: Each modality in the activity prediction dataset is standardized using the following formula: ； in, The RGB values ​​represent the results of standardization processing based on the RGB data collected by each sensor. The mean of RGB values. The standard deviation of RGB; This represents the measured value from the infrared sensor, obtained through standardization. This is the average value of the infrared measurements. The standard deviation of infrared radiation. This represents the impedance measurement value, obtained through standardization. The average impedance. The standard deviation of impedance, The coefficients used to remove outliers are calculated using standardized coefficients and are used to measure the impact of outlier data on the model. This represents the new value after standardization, which is the data value obtained after removing outliers and standardization. Use a mapping function to map the model's predicted output. The mapping function is mapped to discrete activity levels. Used to convert predicted values ​​into corresponding activity levels of 1-5; the expression of the mapping function is: .

Images