Shallow sea monitoring system and method based on multi-modal optical sensing and deep learning

Abstract

This invention relates to the field of optical water quality monitoring technology, specifically to a shallow sea monitoring system and method based on multimodal optical sensing and deep learning. The system includes an electric field excitation module for generating an alternating electric field with adjustable frequency and amplitude; an optical sensing module for generating characteristic luminescent signals under alternating electric field excitation; a visual acquisition module for capturing the characteristic luminescent signals and generating multispectral image data; an edge processing module for real-time processing of the multispectral image data and water quality parameter inversion; a wireless communication module for transmitting the processing results to a terminal device; a terminal presentation module for receiving data and providing visualization and early warning management; and an energy management module for dynamically scheduling power supply strategies based on environmental energy availability. This invention achieves the detection of multiple water quality parameters through alternating current electroluminescence technology, solving the problem of low integration in multi-parameter detection systems and reducing system complexity and maintenance costs.

Description

Technical Field

[0001] This invention relates to the field of optical water quality monitoring technology, specifically to a shallow sea monitoring system and method based on multimodal optical sensing and deep learning. Background Technology

[0002] Shallow water quality monitoring is a crucial technological support for marine environmental protection and the safety of mariculture. Existing monitoring technologies mainly fall into two categories: electrochemical sensors and optical sensors. Electrochemical sensors detect parameters through ion-selective electrodes, but suffer from problems such as electrode polarization, biofouling, and long-term drift, resulting in short maintenance cycles. Optical sensors utilize fluorescence or light absorption properties for detection, but are mostly single-parameter detection modes, rely on external light sources, and exhibit low system integration and high power consumption. In recent years, edge computing and deep learning technologies have been introduced into the field of water quality monitoring, but existing solutions mostly employ general-purpose computing platforms with power consumption exceeding watts, making them unsuitable for the energy supply conditions of the marine environment.

[0003] To address the aforementioned issues, those skilled in the art have made various attempts. For example, using interdigitated electrode structures to achieve conductivity detection has limited functionality; deploying convolutional neural networks on FPGA platforms for image classification only provides qualitative analysis and consumes significant power; and utilizing triboelectric nanogenerators to harvest wave energy to drive buoys lacks energy storage units and solar energy supplementation, making continuous power supply impossible. Existing technologies have not yet solved the system integration problems of multi-parameter synchronous detection, edge intelligent processing, and adaptive energy management.

[0004] The purpose of this invention is to provide a shallow sea monitoring system and method based on multimodal optical sensing and deep learning, which can realize the simultaneous detection of multiple water quality parameters, millisecond-level edge intelligent response and wind-solar collaborative adaptive power supply, and meet the needs of long-term in-situ monitoring in shallow seas. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a shallow sea monitoring system and method based on multimodal optical sensing and deep learning, which can effectively solve the problems mentioned in the existing technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides a shallow sea monitoring system based on multimodal optical sensing and deep learning, comprising:

[0008] An electric field excitation module is used to generate an alternating electric field with adjustable frequency and amplitude. The electric field excitation module includes a signal generation unit, a power amplification unit, an electrode driving unit, and an excitation control unit.

[0009] An optical sensing module is electrically connected to the electric field excitation module via an electrode interface. It is used to generate characteristic light emission signals under alternating electric field excitation and simultaneously sense liquid level, conductivity and temperature. The optical sensing module includes an electrode substrate unit, a zoned light emission unit, a temperature sensing unit and an optical window unit.

[0010] The visual acquisition module, optically coupled to the optical sensing module through an optical window, is used to capture the characteristic emission signal and generate multispectral image data. The visual acquisition module includes an optical imaging unit, a spectral resolution unit, an exposure control unit, and a synchronization triggering unit.

[0011] An edge processing module is connected to the visual acquisition module via a data bus and is used to perform real-time processing of the multispectral image data and water quality parameter inversion. The edge processing module includes a preprocessing unit, a feature extraction unit, a ratio calculation unit, and a parameter inversion unit.

[0012] A wireless communication module is connected to the edge processing module via a communication interface and is used to transmit the processing results to the terminal device;

[0013] The terminal presentation module is connected to the wireless communication module via a wireless protocol and is used to receive data and perform visualization and early warning management.

[0014] The energy management module is connected to the electric field excitation module, visual acquisition module, edge processing module and wireless communication module via a power bus, and is used to dynamically schedule power supply strategies based on environmental energy availability.

[0015] Furthermore, in the electric field excitation module:

[0016] The signal generating unit is used to generate an alternating voltage signal with adjustable frequency and amplitude;

[0017] The power amplification unit is connected to the signal generation unit and is used to amplify the current and match the impedance of the alternating voltage signal.

[0018] The electrode driving unit is connected to the power amplification unit and the optical sensing module, and is used to apply the amplified alternating electric field to the electrode substrate unit.

[0019] The excitation control unit is connected to the signal generation unit and is used to switch between conventional monitoring mode, high-sensitivity detection mode or rapid screening mode according to the monitoring scenario requirements.

[0020] Furthermore, in the optical sensing module:

[0021] The electrode substrate unit includes an insulating substrate and an interdigitated electrode array disposed thereon. The interdigitated electrode array includes a plurality of alternating electrode fingers and electrode gaps. The electrode gaps are used to simultaneously sense liquid level and conductivity through liquid conductivity bridging effect.

[0022] The partitioned light-emitting unit covers the electrode substrate unit and includes a reference light-emitting component located in the middle and a first sensing component and a second sensing component distributed on both sides of the reference light-emitting component. The reference light-emitting component is loaded with a chemically stable fluorescent material, and the first sensing component and the second sensing component are respectively loaded with functional fluorescent materials sensitive to different water quality parameters.

[0023] The temperature sensing unit is disposed on the edge region or the back region of the electrode substrate unit, and is made of conductive polymer material. It is used to detect temperature by resistance temperature characteristics or thermal radiation characteristics.

[0024] The optical window unit is disposed above the partitioned light-emitting unit and is made of transparent hydrophobic material. It is used to transmit characteristic fluorescence and isolate seawater.

[0025] Furthermore, the workflow of the optical sensing module includes:

[0026] Electric field coupling step: The alternating electric field is coupled to the partitioned light-emitting unit through the electrode substrate unit, thereby exciting each sensing component to generate characteristic light emission;

[0027] Liquid level sensing step: Utilizing the liquid conductive bridging effect formed by the contact between the electrode gap and seawater, the liquid level height is calculated based on the number and position of the conductive electrode pairs;

[0028] Conductivity sensing steps: Measure the current-voltage characteristics of the interdigitated electrode array, and calculate the water conductivity by combining the electrode geometric parameters;

[0029] Temperature sensing step: The temperature characteristics of the resistance are monitored by the temperature sensing unit, and the salinity interference is eliminated by combining the conductivity compensation algorithm.

[0030] Furthermore, in the partitioned light-emitting unit:

[0031] The reference light-emitting component is loaded with ZnS:Cu fluorescent material to provide an optical reference signal;

[0032] The first sensing component is loaded with platinum metal porphyrin fluorescent material and is used to detect dissolved oxygen concentration through fluorescence quenching effect;

[0033] The second sensing component is loaded with Rhodamine B fluorescent material and is used to detect pH value by means of changes in fluorescence intensity;

[0034] The system also includes a turbidity detection function, which uses the stable emission of the partitioned light-emitting unit as the background light source and inverts the turbidity value by analyzing the scattered light distribution characteristics.

[0035] Furthermore, in the visual acquisition module:

[0036] The optical imaging unit includes a fixed-focus lens and a CMOS image sensor, used for optical imaging of the characteristic luminescent signal;

[0037] The spectral resolution unit is disposed in the incident light path of the optical imaging unit and includes a switchable narrowband filter group for separating the characteristic wavelengths of different sensing components.

[0038] The exposure control unit is connected to the optical imaging unit and is used to dynamically adjust the exposure parameters according to the ambient light intensity.

[0039] The synchronous triggering unit is connected to the electric field excitation module and the optical imaging unit to achieve phase locking between the electric field excitation period and the image acquisition frame rate.

[0040] Furthermore, in the edge processing module:

[0041] The preprocessing unit is used to perform denoising, enhancement, and region segmentation processing on the multispectral image data;

[0042] The feature extraction unit is connected to the preprocessing unit and includes a lightweight convolutional neural network model for extracting deep semantic features and spectral features of the image.

[0043] The ratio calculation unit is connected to the feature extraction unit and is used to calculate the normalized ratio between the fluorescence intensity of each sensing component and the reference light intensity of the reference light-emitting component.

[0044] The parameter inversion unit is connected to the ratio calculation unit and is used to map the normalized ratio to a quantitative value of water quality parameters based on a preset calibration curve.

[0045] Furthermore, the workflow of the edge processing module includes:

[0046] Image preprocessing steps: Adaptive denoising, contrast enhancement, and color space segmentation are performed on the original image to extract the effective luminous areas of each sensing component;

[0047] Feature extraction step: Extract convolutional features and spectral features of the image using the lightweight convolutional neural network model;

[0048] Ratio calculation steps: Calculate the normalized ratio of the fluorescence intensity of each sensing component to the reference light intensity of the reference component, and perform temperature and salinity compensation.

[0049] Parameter inversion steps: Based on the calibration curve, the compensated ratio is converted into quantitative values ​​of water quality parameters, and an uncertainty estimate is output.

[0050] Furthermore, the energy management module includes:

[0051] The wind energy harvesting unit includes a rotary triboelectric nanogenerator for converting ambient wind energy into electrical energy;

[0052] A light energy harvesting unit, including a solar panel, is used to convert ambient light energy into electrical energy;

[0053] An energy storage unit, including a supercapacitor bank, is connected to the wind energy harvesting unit and the solar energy harvesting unit for storing electrical energy.

[0054] The power management unit is connected to the energy storage unit and each power module, and is used to switch between active power supply mode and energy storage discharge mode according to the availability of ambient energy.

[0055] A shallow sea monitoring method based on multimodal optical sensing and deep learning includes the following steps:

[0056] S100 generates an alternating electric field with adjustable frequency and amplitude;

[0057] S200: Under the excitation of the alternating electric field, a characteristic light emission signal is generated, and the liquid level, conductivity and temperature are sensed simultaneously.

[0058] S300, Capture the characteristic emission signal and generate multispectral image data;

[0059] S400, Perform real-time processing and water quality parameter inversion on the multispectral image data;

[0060] S500: Transmit the processing results to the terminal device;

[0061] S600: Receive the processing results and perform visualization and early warning management;

[0062] S700: Dynamic power supply scheduling strategy based on environmental energy availability.

[0063] The technical solution provided by this invention has the following advantages compared with the known prior art:

[0064] This invention achieves deep coupling of electric field excitation, optical response, and electrical sensing through the innovative integration of AC electroluminescence technology and interdigitated electrode structure. It can simultaneously detect six water quality parameters, namely pH value, dissolved oxygen concentration, temperature, turbidity, liquid level, and conductivity, within a single sensing unit. This solves the problems of multiple discrete sensors and low system integration in existing technologies, and significantly reduces system complexity and maintenance costs.

[0065] This invention utilizes a lightweight deep learning model for edge deployment, compressing the convolutional neural network to less than 500KB and running it on an embedded inference framework. This enables edge intelligent processing with a single-frame inference time of 10ms-50ms and power consumption of less than 200mW. Compared to existing general-purpose computing platforms with watt-level power consumption, energy consumption is reduced by an order of magnitude, and continuous operation without external power for more than 72 hours can be achieved, meeting the long-term in-situ monitoring needs of shallow offshore environments. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0067] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0068] Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0070] The present invention will be further described below with reference to embodiments.

[0071] Example:

[0072] Reference Figure 1 The shallow sea monitoring system based on multimodal optical sensing and deep learning comprises seven functional modules: an electric field excitation module, an optical sensing module, a visual acquisition module, an edge processing module, a wireless communication module, a terminal presentation module, and an energy management module. These modules are connected via electrical interfaces, optical windows, data buses, communication interfaces, or power buses to form a complete monitoring closed loop.

[0073] The electric field excitation module generates an alternating electric field and applies it to the optical sensing module; the optical sensing module generates characteristic light emission signals under electric field excitation and simultaneously senses liquid level, conductivity, and temperature; the visual acquisition module captures the characteristic light emission signals and generates multispectral image data; the edge processing module performs real-time processing of the image data and inversion of water quality parameters; the wireless communication module transmits the processing results to the terminal presentation module; and the energy management module provides adaptive energy supply for each power-consuming module.

[0074] In one specific embodiment, the electric field excitation module includes a signal generation unit, a power amplification unit, an electrode driving unit, and an excitation control unit.

[0075] The signal generation unit employs a direct digital frequency synthesis (DDS) chip or a microcontroller in conjunction with a digital-to-analog converter to generate alternating voltage signals with a frequency adjustment range of 50Hz-1000Hz and an amplitude adjustment range of 10V-100V. The frequency resolution is no less than 0.1Hz, the amplitude resolution is no less than 0.1V, and the waveform types include sine waves, square waves, and triangle waves.

[0076] The power amplifier unit is connected to the signal generation unit and uses a Class D audio power amplifier or a dedicated high-voltage operational amplifier circuit with a gain range of 10-40dB. It is used to amplify the current and match the impedance of the alternating voltage signal, and the output impedance is matched to the equivalent load impedance of the optical sensing module (usually 1kΩ-10kΩ).

[0077] The electrode driving unit is connected to the power amplification unit and the optical sensing module, and includes a high-voltage isolation transformer or a DC-AC inverter circuit for applying the amplified alternating electric field to the electrode substrate unit of the optical sensing module.

[0078] The excitation control unit is connected to the signal generation unit and uses an embedded microprocessor (such as the STM32 series) to switch between three working modes according to the monitoring scenario requirements:

[0079] Standard monitoring mode: frequency 100Hz-500Hz, duty cycle 50%, square wave waveform, suitable for daily water quality monitoring;

[0080] High-sensitivity detection mode: frequency 50Hz-100Hz, sine waveform, improves the excitation efficiency of luminescent materials, suitable for the detection of low-concentration pollutants;

[0081] Rapid screening mode: frequency 500Hz-1000Hz, pulse waveform, shortening the single detection cycle to less than 100ms, suitable for rapid emergency response.

[0082] In one specific embodiment, the optical sensing module and the electric field excitation module are electrically connected via an electrode interface, and are used to generate characteristic light emission signals under alternating electric field excitation and simultaneously sense liquid level, conductivity, and temperature. The optical sensing module includes an electrode substrate unit, a zoned light emission unit, a temperature sensing unit, and an optical window unit.

[0083] The electrode substrate unit includes an insulating substrate and an interdigitated electrode array. The insulating substrate is made of alumina ceramic or polyimide material, with a surface roughness of [missing information]. .

[0084] The interdigitated electrode array is fabricated on the surface of an insulating substrate using photolithography or screen printing, and comprises multiple alternating electrode fingers and electrode gaps. The width of the electrode fingers is... The electrode gap width is The number of electrode pairs is 20-50, and the electrode materials are gold, platinum or indium tin oxide.

[0085] The electrode gap is used to simultaneously sense liquid level and conductivity through the liquid conductivity bridging effect. As the seawater level rises, the electrode pairs are sequentially connected by the seawater, and the number of connected electrode pairs is detected by a scanning circuit. Based on the location and electrode geometry parameters, the liquid level height H is calculated using the following formula:

[0086] ;

[0087] in, The electrode gap spacing, The electrode width is indicated by its reference to ... This is the liquid contact correction factor, typically taken as 0.9-1.0, depending on the surface tension of seawater.

[0088] In another specific embodiment, the current-voltage characteristics of the interdigitated electrode array are measured, and the water conductivity σ is calculated based on the electrode geometry parameters. A small AC voltage of constant amplitude (typically 10mV-100mV, frequency 1kHz-10kHz, distinct from the main excitation frequency) is applied between adjacent electrode pairs, the response current is measured, the inter-electrode resistance R is calculated, and then the water conductivity is calculated using the following formula:

[0089] ;

[0090] in, The length of the electrode gap. The effective contact area is equal to the length of the electrode finger multiplied by the liquid level height.

[0091] The zoned light-emitting units are coated onto the electrode substrate units and are fabricated using spraying, spin coating, or screen printing processes, with a thickness of [missing information]. .

[0092] In one specific embodiment, the partitioned light-emitting unit includes a reference light-emitting component located in the middle and a first sensing component and a second sensing component distributed on both sides of the reference light-emitting component.

[0093] Reference light-emitting element loaded with ZnS:Cu green phosphor (particle size) Cu doping concentration ZnS:Cu emits green fluorescence with a wavelength of approximately 520 nm under alternating electric field excitation, serving as an optical reference to eliminate light source fluctuations and ambient light interference.

[0094] The first sensing component is loaded with platinum metalloporphyrin fluorescent material, specifically platinum(II) octaethylporphyrin (PtOEP), to detect dissolved oxygen concentration through fluorescence quenching. PtOEP emits red fluorescence at a wavelength of approximately 650 nm under alternating electric field excitation. Its fluorescence intensity is sensitive to dissolved oxygen concentration. Dissolved oxygen detection is achieved based on the fluorescence quenching effect caused by oxygen molecule collisions. Quantification is performed by measuring the degree of fluorescence quenching in the 630 nm-670 nm band and combining it with the reference light intensity.

[0095] The second sensing component is loaded with Rhodamine B or Rhodamine 6G fluorescent material to detect pH value through changes in fluorescence intensity. Rhodamine B emits orange-yellow fluorescence with a wavelength of approximately 580 nm under excitation by an alternating electric field. Its fluorescence intensity changes with pH value (showing a linear response in the pH range of 4-10). pH detection is achieved based on changes in the protonation / deprotonation state of the molecule. Quantification is performed by detecting changes in fluorescence intensity in the 540 nm-580 nm band and combining this with a reference light intensity.

[0096] In another specific embodiment, the system also includes a turbidity detection function. Using the stable emission of the partitioned luminescent units as the background light source, the turbidity value is retrieved by analyzing the scattering characteristics of suspended particles in seawater to the background light. Specifically: forward scattered light intensity is positively correlated with turbidity (e.g., for every 1 NTU increase in turbidity, forward scattered light intensity increases by approximately 0.5%-2%), while backscattered light intensity has a non-linear relationship with turbidity (e.g., backscattered light intensity increases rapidly at lower turbidity and tends to saturate at higher turbidity). An inversion model is established based on Mie scattering theory, and the inversion model is implemented using a dual-wavelength ratio method or a neural network fitting method: forward scattered light intensity is collected. and backscattered light intensity Calculate the ratio:

[0097] ;

[0098] By pre-calibration - Turbidity relationship curves (e.g., using cubic spline interpolation or power function fitting) output turbidity values.

[0099] The temperature sensing unit is located at the edge or back of the electrode substrate unit, spatially separated from the partitioned light-emitting unit to avoid light-emitting thermal interference. The temperature sensing unit is fabricated using a conductive polymer material, specifically a poly(3,4-ethylenedioxythiophene):polystyrene sulfonate (PEDOT:PSS) film with a thickness of 100nm-500nm. It is prepared by spin coating and then doped with dimethyl sulfoxide (DMSO) to improve conductivity.

[0100] The temperature sensing unit is used to detect temperature through resistivity-temperature characteristics or thermal radiation features. Temperature detection is based on the resistivity-temperature characteristics of PEDOT:PSS: resistivity decreases with increasing temperature, with a temperature coefficient of approximately -1% / K to -3% / K. Resistance changes are measured using a four-probe method, combined with a conductivity compensation algorithm to eliminate cross-interference from salinity in temperature measurements.

[0101] The optical window unit is positioned above the partitioned light-emitting unit and is made of a transparent hydrophobic material, specifically polydimethylsiloxane (PDMS, refractive index 1.4, thickness 1mm-3mm) or fluorinated ethylene propylene copolymer (FEP, refractive index 1.34, thickness 0.1mm-0.5mm). The refractive index of the optical window unit is close to that of seawater (approximately 1.34), reducing interfacial reflection loss; the surface is treated with a nanostructure to achieve superhydrophobic properties (contact angle >150°), which is used to transmit characteristic fluorescence and isolate seawater, preventing biofouling and optical contamination.

[0102] In one specific embodiment, the workflow of the optical sensing module includes an electric field coupling step, a liquid level sensing step, a conductivity sensing step, and a temperature sensing step.

[0103] Electric field coupling step: The alternating electric field is coupled to the partitioned light-emitting unit through the electrode substrate unit, which excites each sensing component to generate characteristic light emission.

[0104] Liquid level sensing steps: Utilizing the liquid conductivity bridging effect formed by the contact between the electrode gap and seawater, the liquid level height is calculated based on the number and position of the conductive electrode pairs.

[0105] Conductivity sensing steps: Measure the current-voltage characteristics of the interdigitated electrode array and calculate the water conductivity by combining the electrode geometry parameters.

[0106] Temperature sensing steps: The temperature characteristics of the resistance are monitored by the temperature sensing unit, and the salinity interference is eliminated by combining the conductivity compensation algorithm.

[0107] In one specific embodiment, the visual acquisition module and the optical sensing module are optically coupled through an optical window to capture characteristic emission signals and generate multispectral image data. The visual acquisition module includes an optical imaging unit, a spectral resolution unit, an exposure control unit, and a synchronization triggering unit.

[0108] The optical imaging unit includes a fixed-focus lens and a CMOS image sensor. The fixed-focus lens has a focal length of 8mm-16mm, an aperture of f / 2.8-f / 5.6, a working distance of 50mm-200mm, and a field of view covering the entire emitting unit. The CMOS image sensor employs a global shutter architecture, with a resolution of no less than 640×480 pixels and a pixel size of... With a quantum efficiency of >50% (500nm-650nm band) and a dynamic range of >60dB, it is used for optical imaging of characteristic emission signals.

[0109] The spectral resolution unit is disposed in the incident light path of the optical imaging unit and includes a switchable narrowband filter group for separating the characteristic wavelengths of different sensing components. The filter group includes:

[0110] A filter with a center wavelength of 520nm and a bandwidth of 20nm corresponds to a reference light-emitting component;

[0111] A filter with a center wavelength of 650nm and a bandwidth of 30nm corresponds to the first sensing component;

[0112] A filter with a center wavelength of 580nm and a bandwidth of 25nm corresponds to the second sensing component.

[0113] The filter switching is achieved by electromagnetic or piezoelectric drive, with a switching time of <100ms.

[0114] The exposure control unit is connected to the optical imaging unit and employs an automatic exposure algorithm to dynamically adjust the exposure time and gain parameters based on ambient light intensity, ensuring that the emission signals of each sensing component are within the linear response range of the CMOS sensor (typically 20%-80% of the full-well capacity). The exposure time adjustment range is 1μs-1s, and the analog gain adjustment range is 1-16 times.

[0115] The synchronization trigger unit is connected to the electric field excitation module and the optical imaging unit to achieve phase locking between the electric field excitation period and the image acquisition frame rate, eliminating flicker interference. Let the electric field excitation frequency be f and the image acquisition frame rate be FPS, then the synchronization relationship is satisfied. Specifically, when the electric field excitation frequency is low (e.g., f ≤ 100Hz), the following formula is used:

[0116] ;

[0117] That is, the image acquisition frame rate is an integer multiple of the electric field excitation frequency, ensuring that multiple image frames are acquired within each electric field cycle.

[0118] When the electric field excitation frequency is high (e.g., f > 100 Hz), the following formula is used:

[0119] ;

[0120] That is, the image acquisition frame rate is an integer fraction of the electric field excitation frequency, in order to reduce the amount of data.

[0121] Where n is a positive integer. Phase locking is achieved through a phase-locked loop (PLL) circuit or software triggering.

[0122] In one specific embodiment, the edge processing module and the visual acquisition module are connected via a MIPI CSI-2 or USB 2.0 data bus for real-time processing of multispectral image data and water quality parameter inversion. The edge processing module includes a preprocessing unit, a feature extraction unit, a ratio calculation unit, and a parameter inversion unit.

[0123] The preprocessing unit performs denoising, enhancement, and region segmentation on multispectral image data, specifically including three levels of image processing:

[0124] Adaptive denoising: Adaptive Gaussian filtering is used, with kernel standard deviation. The local signal-to-noise ratio is dynamically adjusted, and the specific formula is as follows:

[0125] ;

[0126] in, The standard deviation of the base kernel (usually taken as 1.0-2.0 pixels). For the local window signal-to-noise ratio, This represents the global image signal-to-noise ratio. For regions with high signal-to-noise ratios, the kernel size is reduced to preserve details; for regions with low signal-to-noise ratios, the kernel size is increased to suppress noise.

[0127] Contrast enhancement: The Limiting Contrast Adaptive Histogram Equalization (CLAHE) algorithm is used to divide the image into local sub-regions of 8×8 to 16×16. The histogram of each sub-region is calculated and cropped (the cropping threshold is set to 3-4 times the average pixel value). Then, bilinear interpolation is used to eliminate block artifacts.

[0128] Color space segmentation: Convert the RGB image to the HSV color space and set a dynamic threshold based on the characteristic hue range of each sensing component.

[0129] Reference light-emitting component: H∈[60°,100°], S>0.3, V>0.4;

[0130] First sensing component: H∈[0°,30°]∪[330°,360°], S>0.4, V>0.3;

[0131] Second sensing component: H∈[30°,60°], S>0.3, V>0.4;

[0132] Spatial masks for each luminescent region are extracted using connected component analysis, and the region area, centroid coordinates, and average light intensity are calculated. With light intensity standard deviation .

[0133] The feature extraction unit is connected to the preprocessing unit and includes a lightweight convolutional neural network model for extracting deep semantic and spectral features of the image.

[0134] In one specific embodiment, the lightweight convolutional neural network model is optimized on the edge based on the MobileNetV2 architecture. The model input is a preprocessed 96×96 pixel or 128×128 pixel image patch, and the output is a 128-dimensional feature vector.

[0135] The network structure includes:

[0136] Input layer: 3-channel RGB image, normalized to the range [0,1];

[0137] First hidden layer: 3×3 depthwise separable convolution, 32 channels, stride 2;

[0138] Second hidden layer: Inverted residual block (expansion factor=6, 16 channels → 96 channels → 16 channels), repeated twice;

[0139] Third hidden layer: Inverted residual block (expansion factor=6, 24 channels → 144 channels → 24 channels), step size 2, repeated 3 times;

[0140] Fourth hidden layer: Inverted residual block (expansion factor=6, 32 channels → 192 channels → 32 channels), stride 2, repeated 4 times;

[0141] Fifth hidden layer: Channel attention mechanism, compression ratio 16;

[0142] Global average pooling layer: outputs a 128-dimensional feature vector.

[0143] Model compression strategies include:

[0144] Structured pruning: Removes redundant convolutional kernels, reducing the number of parameters by 30%-50%;

[0145] INT8 quantization: Weights and activation values ​​are quantized into 8-bit integers, compressing the model size to 25% of the original size;

[0146] Operator fusion: Convolution, batch normalization, and ReLU are fused into a single operator, reducing inference latency by 40%.

[0147] The compressed model has less than 500KB of parameters, a single-frame inference time of 10ms-50ms (running on a 480MHz ARM Cortex-M7 processor), and a power consumption of less than 200mW.

[0148] The ratio calculation unit is connected to the feature extraction unit and is used to calculate the normalized ratio between the fluorescence intensity of each sensing component and the reference light intensity of the reference light-emitting component.

[0149] In one specific embodiment, the formula for calculating the normalized ratio is:

[0150] ;

[0151] in, Let i be the normalized ratio of the i-th water quality parameter. Let be the measured fluorescence intensity of the i-th sensing component. As a reference light intensity for the light-emitting component, This is a correction factor for fluorescence quantum efficiency based on real-time temperature. This is a salinity correction function based on real-time conductivity.

[0152] Temperature correction formula:

[0153] ;

[0154] in, Here, T represents the temperature quenching coefficient of the fluorescent material, and T is the measured temperature. The reference temperature is used for calibration (usually 25℃).

[0155] Salinity correction formula:

[0156] ;

[0157] in, The ionic strength influence coefficient (calibrated experimentally, typically -0.001 to -0.01 / (mS / cm)). For the measured conductivity, To calibrate the reference conductivity (usually taken as 50 mS / cm, corresponding to standard seawater).

[0158] The parameter inversion unit is connected to the ratio calculation unit and is used to map the normalized ratio to the quantitative value of water quality parameters based on the preset calibration curve.

[0159] In one specific embodiment, the calibration curve was established through standard solution experiments:

[0160] Dissolved oxygen: Based on the Stern-Volmer equation, the formula is:

[0161] ;

[0162] in, The intensity of fluorescence without oxygen. For actual measured strength, It is the quenching constant;

[0163] pH value: Fluorescence intensity-pH linear relationship based on Henderson-Hasselbalch equation correction;

[0164] Turbidity: A nonlinear model of scattered light intensity-turbidity based on Mie scattering theory;

[0165] Liquid level and conductivity: measured directly from the electrode substrate unit.

[0166] The parameter inversion output includes: pH value (accuracy ±0.1), dissolved oxygen concentration (accuracy ±0.2 mg / L), temperature (accuracy ±0.5℃), turbidity (accuracy ±5%FS), liquid level (accuracy ±5 mm), and conductivity (accuracy ±2%FS).

[0167] In one specific embodiment, the workflow of the edge processing module includes an image preprocessing step, a feature extraction step, a ratio calculation step, and a parameter inversion step.

[0168] Image preprocessing steps: Adaptive denoising, contrast enhancement, and color space segmentation are performed on the original image to extract the effective luminous areas of each sensing component.

[0169] Feature extraction steps: Extract convolutional and spectral features of the image using a lightweight convolutional neural network model.

[0170] Ratio calculation steps: Calculate the normalized ratio of the fluorescence intensity of each sensing component to the reference light intensity of the reference component, and perform temperature and salinity compensation.

[0171] Parameter inversion steps: Based on the calibration curve, the compensated ratio is converted into quantitative values ​​of water quality parameters, and an uncertainty estimate is output.

[0172] In one specific embodiment, the wireless communication module is connected to the edge processing module via a UART or SPI communication interface to transmit the processing results to the terminal device. The wireless communication module supports both a regular monitoring mode and an emergency monitoring mode, with a higher data transmission frequency in the emergency monitoring mode compared to the regular monitoring mode.

[0173] Data transmission formats include:

[0174] Standard monitoring mode: Data packets are transmitted every 30 minutes, containing the average value, standard deviation, and trend slope of 6 water quality parameters;

[0175] Emergency monitoring mode: Transmits real-time data packets every 1-5 minutes, including timestamps, GPS coordinates, raw spectral feature vectors, and anomaly flags.

[0176] The wireless communication module uses Bluetooth Low Energy (BLE 5.0) or LoRa (433MHz / 470MHz) wireless protocols.

[0177] The terminal presentation module and the wireless communication module are connected via a wireless protocol to receive data and perform visualization and early warning management. The terminal presentation module includes a mobile terminal (smartphone APP) or an edge gateway device, providing data visualization (real-time curves, historical trends, spatial distribution maps), multi-level early warnings (blue alerts / yellow warnings / red alerts), and remote configuration functions.

[0178] In one specific embodiment, the energy management module is connected to the electric field excitation module, the vision acquisition module, the edge processing module, and the wireless communication module via a power bus, and is used to dynamically schedule power supply strategies based on environmental energy availability.

[0179] The energy management module includes a wind energy harvesting unit, a solar energy harvesting unit, an energy storage unit, and a power management unit.

[0180] The wind energy harvesting unit includes a rotary triboelectric nanogenerator (TTN) for converting ambient wind energy into electrical energy. The TPN operates in either a contact-separation or sliding mode.

[0181] The solar energy harvesting unit includes solar panels used to convert ambient light energy into electrical energy. The solar panels are made of monocrystalline or polycrystalline silicon materials, with an effective area of ​​10cm²-100cm², a photoelectric conversion efficiency of 15%-22%, and a maximum power point voltage of 2.5V-5V.

[0182] The energy storage unit includes a supercapacitor bank, which is connected to the wind energy harvesting unit and the solar energy harvesting unit to store electrical energy. The supercapacitor bank adopts an activated carbon double-layer structure, with a rated capacity of 1F-100F and a rated voltage of 2.5V-5.5V.

[0183] The power management unit is connected to the energy storage unit and each power consumption module, and is used to switch between active power supply mode and energy storage discharge mode according to the availability of ambient energy.

[0184] In one specific embodiment, the power management unit includes:

[0185] Maximum power point tracking (MPPT) circuitry is used to optimize wind and solar energy harvesting efficiency;

[0186] Overcharge and over-discharge protection circuit to prevent the supercapacitor from being over-voltage (>5.5V) or under-voltage (<2.0V).

[0187] The voltage regulator circuit outputs two stable voltages: 3.3V and 5V.

[0188] The load scheduling circuit dynamically adjusts the power supply priority of each module according to the energy storage status.

[0189] The energy supply strategy switching threshold is set as follows: when the wind speed v ≥ 1.5 m / s or the solar irradiance E ≥ 2000 lux, the system switches to active energy supply mode; otherwise, it switches to energy storage and discharge mode. In active energy supply mode, wind and solar energy are collected together and directly supplied to the load, with surplus energy stored in the energy storage unit; in energy storage and discharge mode, the supercapacitor bank maintains system operation, with a runtime of ≥ 72 hours.

[0190] Reference Figure 2 A shallow sea monitoring method based on multimodal optical sensing and deep learning includes the following steps:

[0191] S100 generates an alternating electric field with adjustable frequency and amplitude;

[0192] S200: Under the excitation of the alternating electric field, a characteristic light emission signal is generated, and the liquid level, conductivity and temperature are sensed simultaneously.

[0193] S300, Capture the characteristic emission signal and generate multispectral image data;

[0194] S400, Perform real-time processing and water quality parameter inversion on the multispectral image data;

[0195] S500: Transmit the processing results to the terminal device;

[0196] S600: Receive the processing results and perform visualization and early warning management;

[0197] S700: Dynamic power supply scheduling strategy based on environmental energy availability.

[0198] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A shallow sea monitoring system based on multimodal optical sensing and deep learning, characterized in that, include: An electric field excitation module is used to generate an alternating electric field with adjustable frequency and amplitude. The electric field excitation module includes a signal generation unit, a power amplification unit, an electrode driving unit, and an excitation control unit. An optical sensing module is electrically connected to the electric field excitation module via an electrode interface. It is used to generate characteristic light emission signals under alternating electric field excitation and simultaneously sense liquid level, conductivity and temperature. The optical sensing module includes an electrode substrate unit, a zoned light emission unit, a temperature sensing unit and an optical window unit. The visual acquisition module, optically coupled to the optical sensing module through an optical window, is used to capture the characteristic emission signal and generate multispectral image data. The visual acquisition module includes an optical imaging unit, a spectral resolution unit, an exposure control unit, and a synchronization triggering unit. An edge processing module is connected to the visual acquisition module via a data bus and is used to perform real-time processing of the multispectral image data and water quality parameter inversion. The edge processing module includes a preprocessing unit, a feature extraction unit, a ratio calculation unit, and a parameter inversion unit. A wireless communication module is connected to the edge processing module via a communication interface and is used to transmit the processing results to the terminal device; The terminal presentation module is connected to the wireless communication module via a wireless protocol and is used to receive data and perform visualization and early warning management. The energy management module is connected to the electric field excitation module, visual acquisition module, edge processing module and wireless communication module via a power bus, and is used to dynamically schedule power supply strategies based on environmental energy availability.

2. The shallow sea monitoring system based on multimodal optical sensing and deep learning according to claim 1, characterized in that, In the electric field excitation module: The signal generating unit is used to generate an alternating voltage signal with adjustable frequency and amplitude; The power amplification unit is connected to the signal generation unit and is used to amplify the current and match the impedance of the alternating voltage signal. The electrode driving unit is connected to the power amplification unit and the optical sensing module, and is used to apply the amplified alternating electric field to the electrode substrate unit. The excitation control unit is connected to the signal generation unit and is used to switch between conventional monitoring mode, high-sensitivity detection mode or rapid screening mode according to the monitoring scenario requirements.

3. The shallow sea monitoring system based on multimodal optical sensing and deep learning according to claim 1, characterized in that, In the optical sensing module: The electrode substrate unit includes an insulating substrate and an interdigitated electrode array disposed thereon. The interdigitated electrode array includes a plurality of alternating electrode fingers and electrode gaps. The electrode gaps are used to simultaneously sense liquid level and conductivity through liquid conductivity bridging effect. The partitioned light-emitting unit covers the electrode substrate unit and includes a reference light-emitting component located in the middle and a first sensing component and a second sensing component distributed on both sides of the reference light-emitting component. The reference light-emitting component is loaded with a chemically stable fluorescent material, and the first sensing component and the second sensing component are respectively loaded with functional fluorescent materials sensitive to different water quality parameters. The temperature sensing unit is disposed on the edge region or the back region of the electrode substrate unit, and is made of conductive polymer material. It is used to detect temperature by resistance temperature characteristics or thermal radiation characteristics. The optical window unit is disposed above the partitioned light-emitting unit and is made of transparent hydrophobic material. It is used to transmit characteristic fluorescence and isolate seawater.

4. The shallow sea monitoring system based on multimodal optical sensing and deep learning according to claim 3, characterized in that, The workflow of the optical sensing module includes: Electric field coupling step: The alternating electric field is coupled to the partitioned light-emitting unit through the electrode substrate unit, thereby exciting each sensing component to generate characteristic light emission; Liquid level sensing step: Utilizing the liquid conductive bridging effect formed by the contact between the electrode gap and seawater, the liquid level height is calculated based on the number and position of the conductive electrode pairs; Conductivity sensing steps: Measure the current-voltage characteristics of the interdigitated electrode array, and calculate the water conductivity by combining the electrode geometric parameters; Temperature sensing step: The temperature characteristics of the resistance are monitored by the temperature sensing unit, and the salinity interference is eliminated by combining the conductivity compensation algorithm.

5. The shallow sea monitoring system based on multimodal optical sensing and deep learning according to claim 3, characterized in that, In the partitioned light-emitting unit: The reference light-emitting component is loaded with ZnS:Cu fluorescent material to provide an optical reference signal; The first sensing component is loaded with platinum metal porphyrin fluorescent material and is used to detect dissolved oxygen concentration through fluorescence quenching effect; The second sensing component is loaded with Rhodamine B fluorescent material and is used to detect pH value by means of changes in fluorescence intensity; The system also includes a turbidity detection function, which uses the stable emission of the partitioned light-emitting unit as the background light source and inverts the turbidity value by analyzing the scattered light distribution characteristics.

6. The shallow sea monitoring system based on multimodal optical sensing and deep learning according to claim 1, characterized in that, In the visual acquisition module: The optical imaging unit includes a fixed-focus lens and a CMOS image sensor, used for optical imaging of the characteristic luminescent signal; The spectral resolution unit is disposed in the incident light path of the optical imaging unit and includes a switchable narrowband filter group for separating the characteristic wavelengths of different sensing components. The exposure control unit is connected to the optical imaging unit and is used to dynamically adjust the exposure parameters according to the ambient light intensity. The synchronous triggering unit is connected to the electric field excitation module and the optical imaging unit to achieve phase locking between the electric field excitation period and the image acquisition frame rate.

7. The shallow sea monitoring system based on multimodal optical sensing and deep learning according to claim 1, characterized in that, In the edge processing module: The preprocessing unit is used to perform denoising, enhancement, and region segmentation processing on the multispectral image data; The feature extraction unit is connected to the preprocessing unit and includes a lightweight convolutional neural network model for extracting deep semantic features and spectral features of the image. The ratio calculation unit is connected to the feature extraction unit and is used to calculate the normalized ratio between the fluorescence intensity of each sensing component and the reference light intensity of the reference light-emitting component. The parameter inversion unit is connected to the ratio calculation unit and is used to map the normalized ratio to a quantitative value of water quality parameters based on a preset calibration curve.

8. The shallow sea monitoring system based on multimodal optical sensing and deep learning according to claim 7, characterized in that, The workflow of the edge processing module includes: Image preprocessing steps: Adaptive denoising, contrast enhancement, and color space segmentation are performed on the original image to extract the effective luminous areas of each sensing component; Feature extraction step: Extract convolutional features and spectral features of the image using the lightweight convolutional neural network model; Ratio calculation steps: Calculate the normalized ratio of the fluorescence intensity of each sensing component to the reference light intensity of the reference component, and perform temperature and salinity compensation. Parameter inversion steps: Based on the calibration curve, the compensated ratio is converted into quantitative values ​​of water quality parameters, and an uncertainty estimate is output.

9. The shallow sea monitoring system based on multimodal optical sensing and deep learning according to claim 1, characterized in that, The energy management module includes: The wind energy harvesting unit includes a rotary triboelectric nanogenerator for converting ambient wind energy into electrical energy; A light energy harvesting unit, including a solar panel, is used to convert ambient light energy into electrical energy; An energy storage unit, including a supercapacitor bank, is connected to the wind energy harvesting unit and the solar energy harvesting unit for storing electrical energy. The power management unit is connected to the energy storage unit and each power module, and is used to switch between active power supply mode and energy storage discharge mode according to the availability of ambient energy.

10. A shallow sea monitoring method based on multimodal optical sensing and deep learning, characterized in that, Includes the following steps: S100 generates an alternating electric field with adjustable frequency and amplitude; S200: Under the excitation of the alternating electric field, a characteristic light emission signal is generated, and the liquid level, conductivity and temperature are sensed simultaneously. S300, Capture the characteristic emission signal and generate multispectral image data; S400, Perform real-time processing and water quality parameter inversion on the multispectral image data; S500: Transmit the processing results to the terminal device; S600: Receive the processing results and perform visualization and early warning management; S700: Dynamic power supply scheduling strategy based on environmental energy availability.

Images