A ship ballast water quality intelligent detection system based on multi-sensor fusion

Abstract

The application discloses a ship ballast water quality intelligent detection system based on multi-sensor fusion, and particularly relates to the technical field of ship environmental protection monitoring, and comprises a hierarchical heterogeneous multi-sensor array module, three types of sensor groups are arranged at four key points of a ballast tank; a data acquisition and preprocessing module, original signals are filtered, normalized and time and space aligned; a multi-level information fusion and intelligent analysis module, biological and optical features are fused through a CNN model to distinguish living and non-living particles, and then an adaptive weighted D-S evidence fusion model is used to output comprehensive quality and biological classification indexes; a decision output and early warning module, based on three-level threshold value determination state and graded early warning pushing. The hierarchical heterogeneous multi-sensor array module, the data acquisition and preprocessing module, the multi-level information fusion and intelligent analysis module and the decision output and early warning module are used to construct the ship ballast water quality intelligent detection system.

Description

Technical Field

[0001] This invention relates to the field of ship environmental monitoring technology, and more specifically, to an intelligent detection system for ship ballast water quality based on multi-sensor fusion. Background Technology

[0002] With the rapid development of the global shipping industry, ballast water, as a core medium for maintaining ship navigation stability and adjusting draft, carries invasive alien species during its transoceanic transport, becoming one of the main causes of marine ecological invasion worldwide. To curb this ecological risk, the International Maritime Organization (IMO) has issued the Ballast Water Management Convention, explicitly requiring ships to be equipped with ballast water quality testing systems to ensure that the biological safety and physicochemical indicators of discharged water meet international and regional environmental standards. Therefore, ballast water quality testing technology has become a key research direction in the field of ship environmental monitoring. While various testing schemes have been developed in existing technologies, many technical shortcomings still exist in practical applications, making it difficult to meet the needs for accurate, real-time, and comprehensive testing.

[0003] In actual use, there are still some shortcomings, such as insufficient monitoring representativeness: since the sensors are usually arranged in a single or a few locations, it is difficult to fully reflect the spatial distribution differences of water quality in different areas (dead corners of the compartment) within the ballast tank, which can easily lead to monitoring blind spots and affect the overall accuracy of the assessment.

[0004] Limited accuracy in bioactivity identification: Existing systems mostly rely on single biological signals such as chlorophyll fluorescence, which makes it difficult to effectively distinguish between live microorganisms with similar fluorescence characteristics and inactive organic particles or inorganic suspended matter. This can easily lead to false alarms or missed alarms, resulting in biases in the estimation of live biological density.

[0005] The information fusion level is shallow and the adaptive capability is weak: existing solutions mostly use simple data fusion with fixed weights or independent parameter comparison, and fail to dynamically adjust the weight of different evidence (such as biological signals and physicochemical parameters) in the comprehensive evaluation according to water quality conditions (such as sudden changes in physicochemical parameters);

[0006] Real-time processing and early warning granularity are coarse: Existing systems typically only provide a binary judgment of "qualified / exceeding the standard", lacking fine distinction of the level of exceeding the standard. Moreover, early warning information is mostly limited to local audible and visual alarms and fails to achieve hierarchical linkage and disposal suggestion push with the ship's central control system and remote management platform, which is not conducive to making rapid and accurate response decisions.

[0007] Therefore, there is an urgent need for a detection system that can achieve comprehensive, high-precision, adaptive intelligent assessment of ballast water quality and provide refined decision support, in order to overcome the above-mentioned deficiencies of existing technologies and meet increasingly stringent requirements for ship ballast water management and environmental protection. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of the prior art, the present invention provides an intelligent detection system for ship ballast water quality based on multi-sensor fusion, which solves the problems mentioned in the background art through the following solutions.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an intelligent detection system for ship ballast water quality based on multi-sensor fusion, comprising:

[0010] Layered heterogeneous multi-sensor array module: The module adopts a compartment-partitioned distributed layout, with a set of sensor sub-arrays deployed at four locations: the inlet and outlet of the ballast tank, the middle of the compartment, and the dead corner of the compartment. The sensor sub-arrays include a physicochemical parameter sensor group for measuring physicochemical parameters, a bioactivity feature sensor group for detecting bioactivity features, and an optical feature sensor group for acquiring particulate matter image information.

[0011] Data acquisition and preprocessing module: Connects to the hierarchical heterogeneous multi-sensor array module, acquires the raw signals of each sensor, and performs filtering, normalization, and spatiotemporal alignment preprocessing on the raw signals, outputting spatiotemporally aligned fusion preparation data;

[0012] Multi-level information fusion and intelligent analysis module: Connected to the data acquisition and preprocessing module, used to perform intelligent analysis on the fused pre-processed data to evaluate ballast water quality. The intelligent analysis includes:

[0013] C101: First-level fusion unit: It fuses the bioactivity signals of the bioactivity feature sensor group with the image morphology features of the optical feature sensor group, and distinguishes between living organisms and inactive particles through a pre-trained neural network model, and outputs an estimated value of the density of living organisms in ballast water.

[0014] C102: Secondary fusion unit, which uses the estimated density of living organisms, physicochemical parameter data, and chlorophyll fluorescence intensity as input evidence to construct an adaptive weighted evidence fusion model. Based on the measured physicochemical parameter data, the evidence weight of the adaptive weighted evidence fusion model is dynamically adjusted to output the comprehensive quality index and biological classification index of ballast water.

[0015] Decision output and early warning module: Connected to the multi-level information fusion and intelligent analysis module, it is used to receive the comprehensive quality indicators and key biological classification indicators of the ballast water, compare them with the preset quality standard indicators, and generate qualified status information and early warning information.

[0016] The technical effects and advantages of this invention are as follows:

[0017] 1. In view of the problem that the sensor deployment points in the existing technology are single and easy to form monitoring blind spots, the present invention adopts a hierarchical heterogeneous multi-sensor array module, and deploys sensor sub-arrays at four key locations: the water inlet, water outlet, the middle of the chamber, and the dead corner of the chamber. Through the partitioned distributed arrangement, the water quality data of different areas in the chamber can be comprehensively acquired.

[0018] 2. To address the problem that existing technologies rely on a single biological signal and have difficulty distinguishing between living and inactive particles, this invention uses a first-level fusion unit to fuse the bioactive signal of the bioactive feature sensor group with the image morphological features of the optical feature sensor group, and uses a pre-trained convolutional neural network model for intelligent recognition.

[0019] 3. To address the issues of shallow information fusion levels and fixed weights in existing technologies, this invention designs a two-level fusion unit and constructs an adaptive weighted evidence fusion model based on improved DS evidence theory. This model can dynamically adjust the weights of each piece of evidence according to real-time measured physicochemical parameter data. When physicochemical parameters are abnormal, the weights are automatically reduced and the decision weight of biological characteristics is increased, thereby enhancing the system's adaptability and the reliability of the fusion results under complex working conditions such as sudden changes in water quality.

[0020] 4. To address the issues of coarse early warning granularity and lack of tiered handling guidance in existing systems, this invention establishes a three-level judgment mechanism through decision output and early warning modules. Based on the judgment results, differentiated audible and visual alarms and handling prompts are triggered, and the early warning information is pushed to the ship's central control system in real time. This achieves multi-level information linkage between local, shipboard, and remote systems, enhancing the timeliness and systematic nature of ballast water management. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the system module structure of the present invention.

[0022] Figure 2 This is a schematic diagram of the hierarchical heterogeneous multi-sensor array module structure of the present invention.

[0023] Figure 3 This is a schematic diagram of the data acquisition and preprocessing module of the present invention.

[0024] Figure 4 This is a schematic diagram of the multi-level information fusion and intelligent analysis module structure of the present invention.

[0025] Figure 5 This is a schematic diagram of the decision output and early warning module 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] Please see Figures 1-5 As shown, this embodiment of the invention provides an intelligent detection system for ship ballast water quality based on multi-sensor fusion, comprising:

[0028] Layered heterogeneous multi-sensor array module: The module adopts a compartment-partitioned distributed layout, with a set of sensor sub-arrays deployed at four locations: the inlet and outlet of the ballast tank, the middle of the compartment, and the dead corner of the compartment. The sensor sub-arrays include a physicochemical parameter sensor group for measuring physicochemical parameters, a bioactivity feature sensor group for detecting bioactivity features, and an optical feature sensor group for acquiring particulate matter image information.

[0029] Data acquisition and preprocessing module: Connects to the hierarchical heterogeneous multi-sensor array module, acquires the raw signals of each sensor, and performs filtering, normalization, and spatiotemporal alignment preprocessing on the raw signals, outputting spatiotemporally aligned fusion preparation data;

[0030] Multi-level information fusion and intelligent analysis module: Connected to the data acquisition and preprocessing module, used to perform intelligent analysis on the fused pre-processed data to evaluate ballast water quality. The intelligent analysis includes:

[0031] C101: First-level fusion unit: It fuses the bioactivity signals of the bioactivity feature sensor group with the image morphology features of the optical feature sensor group, and distinguishes between living organisms and inactive particles through a pre-trained neural network model, and outputs an estimated value of the density of living organisms in ballast water.

[0032] C102: Secondary fusion unit, which uses the estimated density of living organisms, physicochemical parameter data, and chlorophyll fluorescence intensity as input evidence to construct an adaptive weighted evidence fusion model. Based on the measured physicochemical parameter data, the evidence weight of the adaptive weighted evidence fusion model is dynamically adjusted to output the comprehensive quality index and biological classification index of ballast water.

[0033] Decision output and early warning module: Connected to the multi-level information fusion and intelligent analysis module, it is used to receive the comprehensive quality indicators and key biological classification indicators of the ballast water, compare them with the preset quality standard indicators, and generate qualified status information and early warning information.

[0034] The hierarchical heterogeneous multi-sensor array module, in this embodiment, follows the process of surveying and positioning, sub-array deployment, sensor group configuration, wiring and debugging, and calibration and verification. It is adapted to the enclosed, high-humidity, and vibration-prone conditions of ship ballast tanks, specifically including:

[0035] A101: Ballast Tank Operating Condition Survey and Layout Point Marking

[0036] Three-dimensional spatial mapping of the target ship's ballast tanks was carried out to determine the specific coordinates and spatial dimensions of the tank's inlet pipe interface, outlet drain valve, unobstructed area in the middle of the tank, and dead corners (areas prone to water accumulation).

[0037] Based on the vibration frequency of the ship during navigation and the direction of ballast water injection / discharge, mark the sensor installation reference surface at the four deployment points, ensuring that the reference surface is flat and far away from the turbulent area of ​​the pipeline to avoid water flow impacting the sensor probe.

[0038] Record the degree of seawater corrosion and the range of temperature changes at each location to provide a basis for subsequent sensor selection.

[0039] A102: Fixed installation of partitioned sensor subarrays

[0040] For the four deployment points, waterproof and corrosion-resistant sensor mounting brackets were installed. The height of the brackets was 10 to 15 cm higher than the lowest water level of the ballast tank to prevent the silt at the bottom of the tank from covering the sensor probes.

[0041] Further explanation is needed regarding the four deployment points. Inlet point: Fix the sensor subarray bracket 50cm directly below the outlet of the pipeline, facing the direction of water flow, to ensure that the ballast water can flow directly through the sensor probe when injected.

[0042] Outlet location: Fix the bracket 30cm in front of the drain valve and place it close to the cabin wall to avoid the negative pressure during drainage from damaging the sensor;

[0043] Mid-section of the compartment: The sub-array is mounted using a suspended bracket, placing the sensor probe in the middle layer of the ballast water to collect average parameters of the water inside the compartment;

[0044] Dead-end points in the compartment: The sub-array is installed at an angle using a wall-mounted bracket, with the probe facing the dead-end siltation area to ensure that parameters of the stagnant water can be collected.

[0045] A103: Configuration and Connection of Three Types of Sensor Groups within the Subarray

[0046] Physicochemical parameter sensor group configuration: Each sub-array integrates a pH sensor, a turbidity sensor, a salinity sensor, and a dissolved oxygen sensor. The sensor probes are vertically immersed in water, and the connection between the probes and the bracket is sealed with a double-layer waterproof sealing ring.

[0047] Bioactive characteristic sensor group configuration: integrates fluorescent microbial sensor and bioelectrochemical sensor. The fluorescent sensor probe is directly facing the weak natural light area. The electrochemical sensor and the physicochemical parameter sensor are kept 10cm apart to prevent signal crosstalk.

[0048] Optical feature sensor group configuration: integrates a high-speed microscopic imager and a laser particle size analyzer. The imager lens faces the direction of water flow and is coated with an anti-fouling coating. The laser particle size analyzer's transmitter and receiver are arranged in parallel with a spacing of 20cm to capture the shape and trajectory of particles.

[0049] The three types of sensor groups in each subarray are connected through an integrated data acquisition board. The acquisition board has a built-in signal amplification module and outputs a standard analog signal.

[0050] A104: Module Wiring and Interference Suppression

[0051] Electromagnetic interference resistant shielded cables are used to connect the acquisition boards of each subarray to the data acquisition and preprocessing module in the ship's central control room. The cables are fixed along the bulkhead, avoiding the high-voltage power lines of the ship's power system, with a spacing of more than 50cm. Waterproof and pressure-resistant cable seals are used at the cable penetration points to prevent seawater leakage.

[0052] Surge protectors are installed on each subarray acquisition board to resist instantaneous voltage fluctuations during ship navigation and protect sensor components.

[0053] A105: Module Debugging and Calibration

[0054] Single-point calibration: Standard samples were used to calibrate the sensor groups at the four types of points respectively. Physicochemical sensors were calibrated using standard water quality solutions, biological sensors were calibrated using microbial suspensions of known concentrations, and optical sensors were calibrated using particulate matter samples of standard particle size. The calibration coefficients were recorded.

[0055] Zonal consistency calibration: Inject standard water solution into the ballast tank, collect sensor data at four points, adjust the magnification of the acquisition plate, and ensure that the detection deviation of the same parameter at different points is less than 2%;

[0056] Operating condition simulation calibration: Simulate ship navigation vibration with a vibration frequency of 5 to 10 Hz and ballast water salinity abrupt change with a salinity range of 5 to 35‰ to verify the stability of sensor data and eliminate abnormal data caused by vibration or salinity changes.

[0057] The ballast water injection / discharge procedure was initiated, and physical, chemical, biological, and optical data from four points were collected in real time to verify the continuity and integrity of data transmission. For the sensors in the dead corners of the compartment, an automatic cleaning program was set up every 24 hours to remove surface deposits through the micro scraper built into the probe, ensuring long-term monitoring accuracy.

[0058] The data acquisition and preprocessing module, in this embodiment, works in conjunction with the hierarchical heterogeneous multi-sensor array module to establish a stable and reliable data acquisition link, perform standardized preprocessing, and provide high-quality fusion preparation data for the multi-level information fusion and intelligent analysis module. Specifically, this includes:

[0059] B101: Pre-dock preparation and operational condition compatibility check

[0060] Hardware check and debugging: Confirm the core components of the data acquisition and preprocessing module, which include a multi-channel data acquisition card, signal conversion module, edge preprocessing unit, and anti-interference power supply. Power on and test the function of each component to ensure that each channel is fault-free and the signal interface is free from oxidation.

[0061] Sensor array interface matching confirmation: Check the interface type of the integrated data acquisition board of the four sub-arrays in the hierarchical heterogeneous multi-sensor array module one by one (analog interface 4-20mA current / 0-5V voltage; digital interface RS485 / CAN), and record the sensor signal type of each sub-array: physical and chemical sensors are mostly analog interfaces, and biological / optical sensors are mostly digital interfaces.

[0062] Anti-interference hardware deployment: Install electromagnetic shielding boxes in the ship's central control room in advance, prepare anti-electromagnetic interference shielded cables and waterproof sealed joints for the corresponding sensor array wiring through the cabin, and avoid strong electrical interference to the transmission of signals from the ship's power system.

[0063] B102: Physical hardware connection to the hierarchical heterogeneous multi-sensor array module

[0064] B1021: Subarray docking: Connect the output terminals of the acquisition boards of the four sensor subarrays to the multi-channel data acquisition card one-to-one via anti-interference shielded cables;

[0065] Analog sensor signals are connected to the analog channel of the acquisition card, and digital sensor signals are connected to the digital channel of the acquisition card. Each subarray corresponds to an independent channel group, and the channels are labeled with inlet / outlet / middle of the compartment / dead corner of the compartment to avoid confusion.

[0066] B1022: Cable fixing and sealing: The cable is fixed along the ship's bulkhead, avoiding the power system's high-voltage lines with a spacing of ≥50cm. The part passing through the bulkhead is sealed with a waterproof and pressure-resistant sealing joint to prevent seawater leakage. The joints at both ends of the cable are wrapped with insulating and waterproof tape, and anti-pull-out buckles are installed to prevent the joints from loosening due to ship navigation vibration.

[0067] The grounding and power supply connections are made, and the grounding terminals of the data acquisition and preprocessing module and the sensor array acquisition board are all connected to the ship's dedicated grounding pile. The grounding resistance is controlled within [specific range]. Within this range, potential difference interference is eliminated.

[0068] B103: Communication Protocol Adaptation and Data Acquisition Configuration

[0069] For digital sensors such as fluorescent microbial sensors and high-speed microscopic imagers, the Modbus RTU communication protocol is configured in the edge unit of the preprocessing module to be consistent with the communication protocol of the sensor array acquisition board; commands are sent through a serial port debugging tool to verify whether the sensor signals of each subarray are read normally.

[0070] The acquisition parameters are set according to the sensor type: the acquisition frequency of the physicochemical sensor (pH value, turbidity) is set to 1Hz, the acquisition frequency of the biosensor (fluorescence sensor) is set to 2Hz, and the acquisition frequency of the optical sensor (high-speed microscopic imager) is set to 5 frames / second; at the same time, timestamps and location identifiers are bound to the sensor signals, corresponding to the 4 deployment points, to provide data tags for subsequent spatiotemporal alignment.

[0071] The preprocessing module has a built-in solid-state drive with a cache capacity of ≥100GB. When the sensor array data is transmitted in a burst or the communication is interrupted briefly, the original data is automatically cached to avoid data loss and is automatically retransmitted after the connection is restored.

[0072] B104: Raw Signal Acquisition and Real-time Transmission

[0073] Multi-channel synchronous acquisition: When the data acquisition program is started, the preprocessing module synchronously receives the raw signals from all sensors of the four sub-arrays through the multi-channel acquisition card, and analyzes the physical quantities, biomass, and optical image data in the signals in real time, storing them in a temporary database according to the format of location, sensor type, and timestamp.

[0074] Data Acquisition Status Monitoring: Real-time monitoring of signal strength and transmission status of each channel. When a channel signal is interrupted or the signal-to-noise ratio is lower than the threshold ≤30dB, a local alarm is immediately triggered, and the corresponding point and sensor are marked. The system then switches to a backup channel to ensure continuous data acquisition.

[0075] B105: Raw signal preprocessing (filtering, normalization, spatiotemporal alignment)

[0076] B1051: Filtering: Differentiated filtering algorithms are used for different types of interference from sensor signals: For vibration interference signals from physicochemical sensors, wavelet transform filtering algorithm (3 decomposition layers, threshold 0.02) is used to remove high-frequency noise;

[0077] For the fluorescence signal of the biosensor, a moving average filter (window size 5) is used to suppress random interference;

[0078] For the image signal from the optical sensor, median filtering is used to remove salt-and-pepper noise while preserving the morphological characteristics of the particles.

[0079] B1052: Normalization Processing: The filtered sensor data of different magnitudes are uniformly normalized to the [0,1] interval using the min-max normalization algorithm, with the following formula: ,in For single sensor raw data, This represents the historical extreme values ​​of the sensor, ensuring that the magnitudes of physicochemical, biological, and optical data are consistent, thus laying the foundation for subsequent fusion analysis.

[0080] B1053: Spatiotemporal alignment processing: Based on the timestamp and location identifier bound during data acquisition, a two-dimensional alignment is performed. The two-dimensional alignment includes time alignment: using the system clock as a reference, the data acquired by different sensors are aligned to the same time node to eliminate the time difference caused by the difference in sensor response speed.

[0081] Spatial alignment: Based on the coordinates of the four points, a spatial grid model is established, and the data of the same type of sensor at each point are associated with the corresponding grid nodes to ensure the consistency of the data in the spatial dimension. Finally, the spatiotemporally aligned fusion preparation data is output.

[0082] The multi-level information fusion and intelligent analysis module is connected to the data acquisition and preprocessing module, and performs intelligent analysis on the fused pre-data to evaluate the ballast water quality. The intelligent analysis includes:

[0083] C1: Establish a bidirectional communication connection with the output of the data acquisition and preprocessing module via the anti-interference industrial Ethernet TCP / IP protocol; configure the IP address to be on the same local area network segment as the preprocessing module, set the data receiving port to 8080, and ensure that the data transmission bandwidth is ≥100Mbps to meet the high-speed transmission requirements of optical image data; read the spatiotemporal aligned fusion preparatory data output by the data acquisition and preprocessing module, confirm that the data format is a structured format of point identification, sensor type, feature data, and timestamp, configure data parsing rules in the multi-level information fusion and intelligent analysis module, and extract three types of core data:

[0084] Bioactivity characteristic data: fluorescence intensity values, bioelectrochemical response signals;

[0085] Optical image feature data: particle morphology features (area, perimeter, roundness), texture features (grayscale mean, contrast), motion trajectory features;

[0086] Physicochemical parameters: pH, turbidity, salinity, dissolved oxygen, and chlorophyll fluorescence intensity.

[0087] C2: Load the pre-trained convolutional neural network (CNN) model (used to distinguish between living organisms and inactive particles) into the multi-level information fusion and intelligent analysis module;

[0088] The model training set consists of a pre-labeled dataset of live microorganisms, inactive particulate matter images, and biological signals coupled together. The training parameters are: learning rate 0.001, number of iterations 500, and batch size 32. The model weights are initialized, the inference threshold is set to 0.8, and an output probability ≥ 0.8 is used to determine a living organism.

[0089] Further explanation is needed regarding the setting of the inference threshold. In liveness biometrics, the inference threshold is not calculated using a single formula, but rather is a hyperparameter selected based on model performance evaluation and a balance of requirements. Specifically, it includes:

[0090]

[0091] in This represents the probability of a living organism category output by the neural network model, which couples the input image with biological signals. This value is a floating-point number between 0 and 1, representing the confidence level that the input sample belongs to a living organism. This indicates the preset inference threshold, which is set to 0.8 based on historical data;

[0092] threshold The setting principle, threshold The selection of [aspect name] is not arbitrary, but based on the model's performance evaluation on the validation set. It is determined by balancing key metrics, and the selection criteria typically involve the following steps:

[0093] Model output: For each sample in the validation set, the model outputs a probability. ;

[0094] Try different thresholds: Iterate through a series of possible thresholds, such as from 0.5 to 0.95, with a step size of 0.05, for each candidate threshold. :

[0095] Computational performance metrics: accuracy The proportion of samples actually predicted to be alive: ;

[0096] Recall rate The proportion of samples that were actually alive that were successfully predicted: ;

[0097] The harmonic mean of precision and recall is a commonly used metric for measuring the overall accuracy of a binary classification model. TP represents true positive, FP represents false positive, and FN represents false negative;

[0098] choose The reason: This is usually a bias towards high accuracy. Choice: When the model determines a particle as a living organism with a probability of P≥0.8, this determination result is very reliable, with a very low probability of false positives and misclassifying inactive substances as living organisms;

[0099] The trade-off is that some recall rate may be sacrificed. This means that some living organisms with weak activity or atypical characteristics may be missed due to a probability P<0.8 and are therefore classified as inactive.

[0100] Final decision: In the application scenario of ballast water quality monitoring, avoiding false alarms is usually more important than avoiding false alarms, as false alarms will lead to unnecessary alarms; therefore, setting a high threshold of 0.8 to ensure that the output live organism density estimate is highly reliable is a reasonable and rigorous engineering choice.

[0101] C3: Preprocessing module data access and verification, real-time data reception and caching. Start the data reception program and receive the fusion preparation data at 100ms intervals.

[0102] Data validity verification involves two levels of verification for the received data: format verification: data without timestamps or missing location identifiers are removed; value range verification: based on the standard parameter range, abnormal data that exceeds the reasonable value range is removed, such as data with pH < 4 or pH > 10.

[0103] Data that passes the verification process enters the fusion analysis process, while data that fails the verification is marked with abnormal points and sensor types and fed back to the preprocessing module.

[0104] C101: First-level fusion unit: It fuses the bioactivity signals of the bioactivity feature sensor group with the image morphology features of the optical feature sensor group, and distinguishes between living organisms and inactive particles through a pre-trained neural network model, and outputs an estimated value of the density of living organisms in ballast water.

[0105] Further explanation is needed regarding the C101 primary fusion unit, which involves the extraction and fusion of biological and optical features.

[0106] Feature extraction from synchronization data at a single location:

[0107] Bioactivity characteristics: peak intensity and decay time constant of extracted fluorescence signals, and response amplitude of bioelectrochemical signals;

[0108] Optical image features: High-speed microscopic imaging images are binarized to extract the morphology, texture, and motion trajectory features of particles; a feature-level stitching and fusion method is used to stitch the bioactivity feature vectors with the optical image feature vectors at the same time stamp into a dimensional [missing information]. The fused feature vector, The total number of features;

[0109] CNN Model Inference and Classification: The fused feature vector is input into the pre-trained CNN model, and the model outputs the class probability of the particle: the probability of a living organism P1 and the probability of an inactive particle P2. If P1 ≥ 0.8, it is determined to be a living organism; if P1 < 0.8, it is determined to be an inactive particle.

[0110] Calculation of live organism density estimate: The number of particles identified as live organisms per unit volume of ballast water is statistically analyzed. Combined with the detection volume of the optical sensor, the estimated live organism density is calculated using the imaging field of view and sampling flow rate, according to the following formula: ,in Indicates the density of living organisms, The number of living organisms detected per unit time, It measures the volume of water detected per unit time; it outputs estimated values ​​of live organism density at 4 locations, associates the location identifiers with timestamps, and enters the C102 secondary fusion unit.

[0111] C102: Secondary fusion unit, which uses the estimated density of living organisms, physicochemical parameter data, and chlorophyll fluorescence intensity as input evidence to construct an adaptive weighted evidence fusion model. Based on the measured physicochemical parameter data, the evidence weight of the adaptive weighted evidence fusion model is dynamically adjusted to output the comprehensive quality index and biological classification index of ballast water.

[0112] C1021: Input Evidence Set Construction and Weight Initialization: Constructing a three-level input evidence set.

[0113] Evidence A: Estimated live organism density from the first-level fusion output ;

[0114] Evidence B: Physicochemical parameter data (pH, turbidity, salinity, dissolved oxygen) output by the preprocessing module.

[0115] Evidence C: Chlorophyll fluorescence intensity values ​​output by the preprocessing module;

[0116] Initialize evidence weights: The total weight is 1.

[0117] C1022: Dynamic weight adjustment based on physicochemical parameters

[0118] Calculate the deviation between the physicochemical parameters and the preset reference values. ,in For the number of physicochemical parameters, Measured values As the baseline value;

[0119] like (Stable physicochemical parameters): Keep the initial evidence weights unchanged;

[0120] like (Slight deviation of physicochemical parameters): Reduce the weight of evidence B to 0.1, increase the weight of evidence A to 0.7, and keep the weight of evidence C at 0.2;

[0121] like (Severe deviation of physicochemical parameters): The weight of evidence B is reset to 0, the weight of evidence A is adjusted to 0.8, and the weight of evidence C is adjusted to 0.2;

[0122] The core logic of weight adjustment is that the greater the deviation of physicochemical parameters, the lower their reference value for comprehensive quality assessment, and priority should be given to biometric data.

[0123] C1023: Adaptive Weighted Evidence Fusion Model Reasoning

[0124] A fusion model is constructed using an improved DS evidence theory:

[0125] Assign a basic probability mass function to each piece of evidence, and calculate the degree of support for each piece of evidence for qualified / excessive and harmful / harmless based on the data credibility.

[0126] Introducing the conflict coefficient When the degree of conflict between evidence At the same time, a weighted average method is used to correct the mass function to avoid distortion of the fusion result;

[0127] The conflict coefficient needs further explanation. The formula used to measure the degree of conflict between pieces of evidence is: Where A and B are mutually exclusive propositions in the identification framework, and a conflict coefficient threshold is set. :

[0128] like With minimal conflict of evidence, the DS synthesis rules were directly applied for fusion.

[0129] like Due to significant conflict in the evidence, a weighted average correction method is adopted, which involves weighting the mass functions of each piece of evidence according to their respective weights before synthesizing them to avoid distortion of the fusion result.

[0130] All evidence is synthesized according to the adjusted weights, and a comprehensive quality index (qualified / slightly exceeding the standard / severely exceeding the standard) and a biological classification index (plankton / harmful microorganisms / inactive particles) are output.

[0131] The data output and early warning module is connected to encapsulate the comprehensive quality indicators, biological classification indicators, and live organism density data from four locations in the format of timestamp, location, and indicator results, and transmits them to the decision output and early warning module via the industrial bus.

[0132] The decision output and early warning module is connected to the multi-level information fusion and intelligent analysis module. It is used to receive the comprehensive quality indicators and key biological classification indicators of the ballast water, compare them with the preset quality standard indicators, and generate qualified status information and early warning information.

[0133] D101: Docking Preparation and Multi-level Information Fusion and Intelligent Analysis Module Protocol Adaptation

[0134] D1011: Hardware and Communication Connection: The main control unit of the decision output and early warning module establishes a two-way communication connection with the output of the multi-level information fusion and intelligent analysis module through the anti-interference industrial bus RS485;

[0135] The module is installed inside the central control panel in the ship's bridge and is equipped with a backup power module to ensure uninterrupted operation when the ship's power grid fluctuates.

[0136] Configure communication parameters: set baud rate to 115200bps, data bits to 8 bits, stop bits to 1 bit, parity bits to none, set data reception timeout to 500ms, trigger a communication error alarm if timeout occurs.

[0137] D1012: Data Format and Tag Adaptation. Configure data parsing rules in the main control unit. The data received from the multi-level information fusion and intelligent analysis module is a structured data packet containing timestamps, location identifiers, comprehensive quality indicators, biological classification indicators, and live organism density values. Define core data fields:

[0138] Overall quality indicators: Qualified / Slightly exceeded / Severely exceeded;

[0139] Biological classification index: plankton / harmful microorganisms / non-active particles;

[0140] Location markings: Inlet / Outlet / Middle of compartment / Dead corner of compartment;

[0141] Live organism density value Units are per L, rounded to two decimal places.

[0142] D1013: Preset quality standard loading

[0143] Based on the Ballast Water Management Convention and the specific environmental standards for ship navigation areas, preset quality standard indicators are entered into the database to establish a three-level judgment threshold table:

[0144] Pass: Threshold for live organism density Quantity / L, threshold deviation of harmful microorganisms The overall assessment concluded that no indicators exceeded the standard.

[0145] Mild exceedance: Live organism density threshold Quantity / L, threshold deviation of harmful microorganisms The overall assessment indicates that a single indicator exceeds the standard;

[0146] Severe exceedance: Threshold for live organism density Quantity / L, threshold deviation of harmful microorganisms The overall assessment indicates that multiple indicators exceed the standards.

[0147] D102: Data Access and Validity Verification for Multi-Level Information Fusion and Intelligent Analysis Module

[0148] Real-time data reception and caching: The data reception program is started and synchronously receives data packets output by the multi-level information fusion and intelligent analysis module at a frequency of 1 second / time; a circular buffer (capacity ≥ 24h of data volume) is established in the module's built-in Flash memory, and the data is stored by timestamp + location to avoid data loss due to communication interruption.

[0149] Data validity is verified in two ways: format verification, removal of data packets without timestamps, missing location identifiers, or incomplete fields, marking the reception time of abnormal data packets and recording it in the log;

[0150] Logical verification: Based on a preset threshold table, data that clearly exceeds the reasonable value range (such as negative live organism density or harmful microorganisms accounting for >100%) is removed, judged as invalid data, and fed back to the multi-level information fusion and intelligent analysis module to trigger a data retransmission instruction.

[0151] D103: Quality Indicator Comparison and Status Determination

[0152] Single-point index comparison: For data packets that have passed verification, the data is split by point, and the live organism density value and the proportion of harmful microorganisms of each point are compared with the preset threshold table one by one to preliminarily determine the quality level of the point.

[0153] Overall ballast tank condition assessment: Summarize the preliminary assessment results from the four locations and output the overall mass condition of the ballast tank according to the following rules:

[0154] If all points are qualified, the overall cabin condition is deemed qualified; if 1 to 2 points are slightly exceeding the standard and there are no points with severely exceeding the standard, the overall cabin condition is deemed slightly exceeding the standard; if there is ≥1 point with severely exceeding the standard or ≥3 points with slightly exceeding the standard, the overall cabin condition is deemed severely exceeding the standard.

[0155] D104: Tiered Early Warning Generation and Multi-Terminal Information Push

[0156] Based on the overall status assessment of cabin D103, a corresponding level of early warning was triggered, and the alarm format was linked to the corresponding handling prompts:

[0157] Qualified: Alarm mode: Green indicator light stays on; Handling prompt: Ballast water quality meets standards and can be discharged normally;

[0158] Slight exceedance: Alarm mode: yellow indicator light flashes + buzzer sounds intermittently (every 5 seconds); Action suggestion: it is recommended to change the water in the area exceeding the standard, monitor the water quality, and then discharge the water.

[0159] Severe exceedance: Alarm mode: red indicator light stays on and buzzer sounds continuously; Action prompt: Do not discharge! Immediately start the ballast water purification device, replace all water in the cabin, and then retest.

[0160] Local display: The display screen of the early warning terminal in the driver's cab shows the quality status of the entire cabin, the locations exceeding the standard, the density curve of live organisms, and the handling prompts in real time;

[0161] Central control push: pushes early warning information and exceedance reports to the ship's central control system and stores them in the ship's navigation log;

[0162] Cloud upload: The severe exceedance warning information is uploaded to the shipping company's cloud management platform in real time via satellite communication module, which facilitates remote monitoring and dispatch.

[0163] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0164] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart detection system for ship ballast water quality based on multi-sensor fusion, characterized in that, include: Layered heterogeneous multi-sensor array module: The module adopts a compartment-partitioned distributed layout, with a set of sensor sub-arrays deployed at four locations: the inlet and outlet of the ballast tank, the middle of the compartment, and the dead corner of the compartment. The sensor sub-arrays include a physicochemical parameter sensor group for measuring physicochemical parameters, a bioactivity feature sensor group for detecting bioactivity features, and an optical feature sensor group for acquiring particulate matter image information. Data acquisition and preprocessing module: Connects to the hierarchical heterogeneous multi-sensor array module, acquires the raw signals of each sensor, and performs filtering, normalization, and spatiotemporal alignment preprocessing on the raw signals, outputting spatiotemporally aligned fusion preparation data; Multi-level information fusion and intelligent analysis module: Connected to the data acquisition and preprocessing module, used to perform intelligent analysis on the fused pre-processed data to evaluate ballast water quality. The intelligent analysis includes: C101: First-level fusion unit: Extracts peak fluorescence intensity, decay time constant and bioelectrochemical signal response amplitude from single-site synchronous data; extracts particulate morphology, texture and motion trajectory features from high-speed microscopic imaging images after binarization. A feature-level splicing and fusion method is adopted to splice the bioactivity feature vectors and optical image feature vectors at the same time stamp into a fused feature vector; The fused feature vectors are input into a pre-trained CNN model, which outputs class probabilities. A value of P1 ≥ 0.8 indicates a living organism, while P1 < 0.8 indicates an inactive particle. The formula calculates the estimated density of living organisms, where Indicates the density of living organisms, The number of living organisms detected per unit time, Output the estimated density of living organisms at four locations for the volume of water detected per unit time. C102: Secondary fusion unit, including C1021: Constructing a three-level input evidence set, where evidence A is the estimated density of living organisms, evidence B is the physicochemical parameter data, and evidence C is the chlorophyll fluorescence intensity value. The initial evidence weights are: evidence A weight 0.5, evidence B weight 0.3, and evidence C weight 0.2, with a total weight of 1. C1022: Calculate the deviation between physicochemical parameters and preset benchmark values. The calculation formula is as follows: ,middle For the number of physicochemical parameters, Measured values As the baseline value, if Keep the initial evidence weights unchanged; like Decrease the weight of evidence B to 0.1, increase the weight of evidence A to 0.7, and keep the weight of evidence C at 0.2; like The weight of evidence B is reset to 0, the weight of evidence A is adjusted to 0.8, and the weight of evidence C is adjusted to 0.

2. C1023: A fusion model is constructed using an improved DS evidence theory, assigning a basic probability mass function to each piece of evidence and introducing a conflict coefficient. The calculation formula is: Where A and B are mutually exclusive propositions in the identification framework, and a conflict coefficient threshold is set. : like With minimal conflict of evidence, the DS synthesis rules were directly applied for fusion. like The evidence is highly conflicting. A weighted average correction method is adopted to weight the mass functions of each piece of evidence according to their respective weights before synthesizing them. This avoids distortion of the fusion result. The DS synthesis rule and weighted average method are used to correct the mass function and output the comprehensive quality index and biological classification index. Decision output and early warning module: Connected to the multi-level information fusion and intelligent analysis module, it is used to receive the comprehensive quality indicators and biological classification indicators of the ballast water, compare them with the preset quality standard indicators, and generate qualified status information and early warning information.

2. The intelligent detection system for ship ballast water quality based on multi-sensor fusion according to claim 1, characterized in that, The hierarchical heterogeneous multi-sensor array module includes: A101: Conduct three-dimensional spatial mapping of the target ship's ballast tanks, determine the specific coordinates and spatial dimensions of the four deployment points, mark the sensor installation reference surface, and record the operating data of each point. A102: Install waterproof and corrosion-resistant sensor mounting brackets at four deployment points. The bracket at the inlet point is fixed 50cm directly below the outlet of the pipeline, and the bracket at the outlet point is fixed 30cm in front of the drain valve and attached to the tank wall. A floating bracket is used in the middle of the tank, and a wall-mounted inclined bracket is used in the dead corner of the tank. The height of the bracket is 10 to 15cm higher than the lowest water level of the ballast tank. A103: In each subarray, the physicochemical parameter sensor group integrates a pH sensor, a turbidity sensor, a salinity sensor, and a dissolved oxygen sensor; the bioactivity characteristic sensor group integrates a fluorescent microbial sensor and a bioelectrochemical sensor; and the optical characteristic sensor group integrates a high-speed microscopic imager and a laser particle size analyzer. The three types of sensor groups are connected through an integrated data acquisition board. A104: Electromagnetic interference resistant shielded cables are used to connect each subarray to the data acquisition and preprocessing module. The cables are fixed along the cabin wall and the distance between them and the power lines is greater than 50cm. Waterproof and pressure-resistant sealed joints are used at the parts that pass through the cabin. Surge protectors are installed at each subarray acquisition board. A105: Perform single-point calibration, zone consistency calibration, and working condition simulation calibration in sequence, and set an automatic cleaning program for the cabin dead-angle point sensors every 24 hours.

3. The intelligent detection system for ship ballast water quality based on multi-sensor fusion according to claim 2, characterized in that, In the A103, the fluorescence sensor probe faces the area of ​​weak natural light, and the electrochemical sensor and the physicochemical parameter sensor are kept 10cm apart; in the optical feature sensor group, the imager lens faces the direction of water flow and is coated with an anti-fouling coating, and the laser particle size analyzer transmitter and receiver are arranged in parallel with a spacing of 20cm.

4. The intelligent detection system for ship ballast water quality based on multi-sensor fusion according to claim 1, characterized in that, The data acquisition and preprocessing module includes: B101: The component includes a multi-channel data acquisition card, a signal conversion module, an edge preprocessing unit, an anti-interference power supply, an electromagnetic shielding box, an anti-electromagnetic interference shielded cable, and a waterproof sealing connector. B102: Connect the output terminals of the acquisition boards of the four sensor sub-arrays to the multi-channel data acquisition card one-to-one through anti-interference shielded cables. Analog signals are connected to the analog channels, and digital signals are connected to the digital channels. Wrap the connectors at both ends of the cable with insulating and waterproof tape and install anti-pull-out buckles. The grounding terminals of the module and the sensor array acquisition boards are connected to the ship-specific grounding pile. B103: Configures the ModbusRTU communication protocol for digital sensors, sets the acquisition frequency of physicochemical sensors to 1Hz, the acquisition frequency of biological sensors to 2Hz, and the acquisition frequency of optical sensors to 5 frames / second, binds timestamps and location identifiers to sensor signals, and has a built-in solid-state drive with a cache capacity of ≥100GB. B104: Synchronously receives raw sensor signals through a multi-channel acquisition card, stores them in a formatted temporary database, monitors channel signal status in real time, and triggers a local alarm and switches to a backup channel when the signal is interrupted or the signal-to-noise ratio is ≤30dB. B105: The original signal is processed using a differential filtering algorithm. The physicochemical sensor signal is filtered by wavelet transform with a threshold of 0.02 and a 3-level decomposition. The biosensor fluorescence signal is filtered by moving average with a window size of 5. The optical sensor image signal is filtered by median filtering. The data is unified to the [0,1] interval by the min-max normalization algorithm. Spatiotemporal alignment is performed based on timestamps and location identifiers.

5. The intelligent detection system for ship ballast water quality based on multi-sensor fusion according to claim 1, characterized in that, The multi-level information fusion and intelligent analysis module includes: The data acquisition and preprocessing module is connected via an anti-interference industrial Ethernet TCP / IP protocol. The IP addresses are in the same local area network segment, the data receiving port is 8080, the transmission bandwidth is ≥100Mbps, and biological activity characteristic data, optical image characteristic data, physicochemical parameter data and chlorophyll fluorescence intensity value are extracted. Load a pre-trained convolutional neural network (CNN) model. The model training set is a coupled dataset of labeled images of live microorganisms and inactive particulate matter and biological signals. The training parameters are a learning rate of 0.001, 500 iterations, a batch size of 32, and an inference threshold of 0.

8. Data for fusion preparation is received at 100ms intervals, and format and value range verification is performed. Data that passes the verification enters the fusion analysis process.

6. The intelligent detection system for ship ballast water quality based on multi-sensor fusion according to claim 1, characterized in that, The decision output and early warning module includes: D101: Establishes bidirectional communication with the multi-level information fusion and intelligent analysis module via the anti-interference industrial bus RS485, and establishes a three-level judgment threshold table: Pass: Threshold for live organism density Quantity / L, threshold deviation of harmful microorganisms The overall assessment concluded that no indicators exceeded the standard. Mild exceedance: Live organism density threshold Quantity / L, threshold deviation of harmful microorganisms The overall assessment indicates that a single indicator exceeds the standard; Severe exceedance: Threshold for live organism density Quantity / L, threshold deviation of harmful microorganisms The overall assessment indicates that multiple indicators exceed the standards. D102: Receives structured data packets, has a built-in Flash circular buffer with a capacity of ≥24h of data, performs format and logical checks on the data, and feeds back invalid data to the multi-level information fusion and intelligent analysis module and triggers retransmission; D103: Based on the three-level judgment threshold table established by D101, the indicators of each point are compared, and the results of the four points are summarized to determine the overall cabin status. If all sampling points are qualified, the entire cabin is qualified; if 1 to 2 sampling points are slightly exceeding the standard but no severely exceeding the standard, the entire cabin is slightly exceeding the standard; if ≥1 severely exceeding the standard or ≥3 slightly exceeding the standard, the entire cabin is severely exceeding the standard. D104: A green indicator light remains on for a qualified status; a yellow indicator light flashes and a buzzer sounds intermittently for a slight exceedance; a red indicator light remains on and a buzzer sounds continuously for a severe exceedance. Relevant information is displayed on the bridge terminal and pushed to the ship's central control system. Severe exceedance information is uploaded to the cloud management platform via satellite communication module.

7. The intelligent detection system for ship ballast water quality based on multi-sensor fusion according to claim 6, characterized in that, The decision output and early warning module generates qualified status information and early warning information, including: Qualified: Alarm mode: Green indicator light stays on; Handling prompt: Ballast water quality meets standards and can be discharged normally; Slight exceedance: Alarm mode: yellow indicator light flashing + buzzer sounding intermittently; Action suggestion: it is recommended to change the water in the area exceeding the standard, monitor the water level, and then discharge the water. Severe exceedance: Alarm mode: red indicator light stays on and buzzer sounds continuously; Action prompt: Do not discharge, immediately start ballast water purification, replace all water in the cabin and then retest.

Images