Intelligent optical sensing system and method for imaging through underwater turbid environment

Abstract

The invention provides an intelligent optical sensing system and method for imaging through an underwater turbid environment, and belongs to the field of underwater optical imaging. The problem of low underwater object recognition efficiency caused by difficulty in obtaining clear images in turbid water by existing equipment is solved; the system comprises an optical imaging module, an image acquisition module and computing equipment which are carried on diving equipment, computer programs of an image preprocessing module, a depth sensing module and a neural network denoising recognition module run on the computing equipment, and the optical imaging module generates a structured light field carrying an orbital angular momentum mode based on a spiral phase plate; the image acquisition module is used for acquiring reflection information of the target object irradiated by the vortex light; the image preprocessing module is used for enhancing the acquired image; the depth sensing module obtains the current water pressure through a pressure sensor, and the current water depth is inversely solved through the computing device; the neural network denoising recognition module carries out real-time recognition and denoising on the image; the method is applied to turbid underwater imaging.

Description

Technical Field

[0001] The present application relates to the field of underwater optical imaging technology, and in particular to an intelligent optical perception system and method for imaging through an underwater turbid environment. Background Art

[0002] In underwater search and rescue missions, efficient underwater optical imaging, target recognition capabilities, and position information transmission are key technologies for improving search and rescue efficiency and success rates in order to achieve rapid positioning and rescue. However, in real underwater environments, conventional camera equipment struggles to obtain clear and reliable image information in complex water conditions such as high turbidity, low illumination, and deep sea conditions. This results in rescue personnel being unable to accurately identify targets in distress, potential obstacles, or risk areas. This imaging degradation phenomenon is primarily caused by scattering effects and environmental absorption, and is particularly severe in strongly scattering media. The inability to quickly and accurately obtain environmental information about the distress area will delay rescue efforts and may even lead to rescue personnel misjudging the environment, threatening their own lives.

[0003] Currently, imaging solutions based on underwater cameras remain the mainstream means of underwater detection, offering advantages such as high resolution, fast acquisition, and intuitive visualization. Methods such as multispectral imaging, polarization imaging, binocular vision, and time-gated imaging demonstrate excellent noise immunity and imaging performance in specific scenarios. These imaging methods can, to a certain extent, reduce interference caused by scattering and improve image clarity. However, existing technologies still have many limitations. First, current imaging systems rely on high-precision components, complex multi-device coordination, and high-frame-rate acquisition, resulting in high system costs and low integration, making them difficult to directly deploy in actual search and rescue equipment. Second, in pursuit of high accuracy and resolution, many imaging solutions sacrifice real-time imaging performance, failing to meet the urgent imaging speed requirements of rescue scenarios. Third, in underwater environments, the lack of reference spatial coordinate information makes it difficult to accurately locate the person in distress even if imaging and identification are successful, thus hindering the timeliness and accuracy of rescue deployment. Therefore, there is an urgent need to develop an intelligent optical perception system that can image through underwater turbid environments, integrate functions such as real-time imaging through underwater turbid environments, target recognition, and real-time location information, and have good environmental adaptability and system integrability to meet the "fast, accurate, and safe" technical requirements of emergency search and rescue missions in turbid waters. Summary of the Invention

[0004] In order to solve the problem that existing equipment has difficulty in obtaining clear images in turbid water, resulting in low efficiency in identifying underwater objects, this application proposes an intelligent optical perception system and method for imaging through underwater turbid environments. Based on vortex light illumination and neural network denoising and recognition technology, it utilizes the good transmission properties of vortex light in turbid water, the efficient feature extraction and robust recognition capabilities of neural networks for image data, and the water pressure-water depth inverse algorithm to achieve real-time imaging and recognition of target objects in complex water environments and intelligent perception of position depth.

[0005] The technical solution adopted in this application is: an intelligent optical perception system for imaging through underwater turbid environments, including a computing device and an optical imaging module, an image acquisition module, and a pressure sensor mounted on the diving equipment. The computing device runs a computer program having an image preprocessing module, a depth perception module, and a neural network denoising and recognition module. The optical imaging module is based on a spiral phase plate and generates a structured light field carrying an orbital angular momentum pattern; the image acquisition module obtains the reflection information of the target object after being irradiated by vortex light through a high-sensitivity imaging device; the image preprocessing module performs illumination and reflection information separation processing based on the Retinex model, frequency domain Gaussian filtering illumination estimation, and adaptive histogram equalization enhancement on the captured image; the depth perception module uses a pressure sensor to obtain the current water pressure, and the computing device inversely resolves the current water depth; the neural network denoising and recognition module performs real-time recognition and denoising on the image.

[0006] Furthermore, the optical imaging module includes a laser, a beam expander and a spiral phase plate. The laser emitted by the laser is expanded by the beam expander and irradiated onto the spiral phase plate, where it is modulated to generate vortex light carrying different orbital angular momentum modes. The vortex light beam is irradiated onto the underwater target.

[0007] Furthermore, the image acquisition module includes a lens and a CCD camera. The vortex light beam is irradiated onto the underwater target, and the reflected light is collected by the lens and imaged onto the CCD camera. The CCD camera sends the collected image to the computing device.

[0008] Furthermore, the neural network denoising recognition module uses the data-driven EnhanceNet neural network. The EnhanceNet neural network has an encoder-decoder structure. During the encoding stage, the input image undergoes multiple compression operations to extract image features. Each compression operation passes through two convolution modules and is downsampled through a maximum pooling. After the compression operation is completed, the features are flattened into one dimension and input into the attention module. The attention module contains multiple consecutive Transformer layers. Each Transformer layer contains a multi-head self-attention layer and a multi-layer perceptron layer. Both the multi-head self-attention layer and the multi-layer perceptron layer are first normalized to stabilize the training process, and residual connections are used.

[0009] The network decoding part contains two decoders, a denoising decoder and a classification decoder, which implement denoising and classification operations on the input image;

[0010] The denoising decoder gradually restores the spatial resolution of the image through multi-layer upsampling and skip connections. Each level of decoding includes feature splicing, convolution, nonlinear activation and batch normalization, and finally outputs an image of the same size as the input, achieving noise suppression and detail restoration.

[0011] The classification decoder flattens the feature information of each layer in the upsampling process and then splices them together, inputs them into the fully connected layer, outputs multiple classifications at the image level, and finally the network outputs the denoised image and image classification information.

[0012] Furthermore, the diving equipment includes two transparent windows: a light outlet and a light receiving port, a pair of horizontal propulsion propellers, a pair of vertical propulsion propellers and a pressure sensor, wherein the light outlet and the light receiving port are arranged in the front of the diving equipment, the pair of horizontal propulsion propellers are symmetrically arranged on both sides of the diving equipment, the pair of vertical propulsion propellers are symmetrically arranged in the middle of the upper and lower surfaces of the diving equipment, and the pressure sensor is arranged on the upper surface of the diving equipment.

[0013] An intelligent optical perception method for imaging through an underwater turbid environment, based on an intelligent optical perception system for imaging through an underwater turbid environment, comprises the following steps:

[0014] Step 1: Illuminate the underwater target through the vortex beam in the optical imaging module;

[0015] Step 2: Collect the original image of the target object and preprocess it through the image preprocessing module;

[0016] Step 3: Enhance and identify the target image through the data-driven EnhanceNet neural network;

[0017] Step 4: Real-time detection and calculation of water depth.

[0018] Furthermore, step 2 is specifically as follows:

[0019] Step 2.1: The light reflected from the target is focused by the lens and imaged on the detection surface of the CCD camera;

[0020] Step 2.2: The original image is captured by the CCD camera and transmitted to the computing device;

[0021] Step 2.3: Preprocess the raw image received by the computing device. Estimate the collected raw image as the product of the illumination component and the reflection component. Use Gaussian filtering to estimate the illumination information. Then, use frequency domain convolution and inverse Fourier transform operations to obtain the illumination information for inverse reflection information. Normalize the reflection information and then process the normalized reflection information using an adaptive histogram equalization algorithm to enhance contrast and improve local details.

[0022] A computer device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0023] A computer-readable storage medium stores a computer program / instruction thereon, which implements the steps of the method when executed by a processor.

[0024] A computer program product comprises a computer program / instructions which, when executed by a processor, implement the steps of the method.

[0025] The beneficial effects of this application compared to the prior art are:

[0026] 1. This application uses a vortex beam to illuminate underwater targets, effectively enhancing the anti-interference capability during underwater imaging and increasing the underwater imaging distance. It can effectively detect targets up to 4 meters away from the system in turbid underwater environments.

[0027] 2. This application combines a neural network with a denoising algorithm to adaptively respond to changes in turbidity environments over a wide dynamic range. It can effectively improve the quality of underwater real-time imaging by more than 10dB, providing rescuers with clearer and more reliable visual and distance information, thereby improving the efficiency and success rate of rescue operations.

[0028] 3. This application significantly improves the recognition accuracy of dynamic targets in turbid environments by combining structured light illumination with data-driven intelligent recognition methods, and is suitable for various complex application scenarios such as underwater search and rescue, water security, and water environment monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present application will be further described below with reference to the accompanying drawings:

[0030] Figure 1 A schematic diagram of the overall system structure provided for an embodiment of the present application;

[0031] Figure 2 A schematic diagram of the optical imaging structure provided in an embodiment of the present application;

[0032] Figure 3 A schematic diagram of the structure of the image denoising and recognition network provided in an embodiment of the present application;

[0033] Figure 4 The results of the imaging experiment provided in the embodiments of the present application to verify the effectiveness of the present application;

[0034] Figure 5 A quality line graph of imaging experiment results provided in the embodiments of the present application to verify the effectiveness of the present application;

[0035] Figure 6 The results of the dynamic imaging experiment provided in the embodiment of the present application to verify the effectiveness of the present application;

[0036] In the figure: 101 is the light outlet, 102 is the light receiving port, 103 is the horizontal propulsion propeller, 104 is the vertical propulsion propeller, 105 is the pressure sensor, 201 is the laser, 202 is the beam expander, 203 is the spiral phase plate, 204 is the target object, 205 is the lens, 206 is the CCD camera, and 207 is the computing device. DETAILED DESCRIPTION

[0037] like Figures 1 to 6 As shown, the present application provides an intelligent optical perception system for imaging through underwater turbid environments, which includes an optical imaging module, an image acquisition module, a pressure sensor and a computing device 207. The optical imaging module, the image acquisition module and the pressure sensor are mounted on the diving equipment. The computing device 207 runs a computer program including an image preprocessing module, a depth perception module and a neural network denoising and recognition module. The optical imaging module generates a structured light field carrying an orbital angular momentum pattern based on a spiral phase plate, thereby enhancing the anti-scattering capability of the underwater turbid environment. The image acquisition module obtains the reflection information of the target after being irradiated by vortex light through a high-sensitivity imaging device and sends it to the computing device 207. The image preprocessing module performs illumination and reflection information separation processing based on the Retinex model, frequency domain Gaussian filtering illumination estimation and adaptive histogram equalization enhancement on the captured image. The depth perception module obtains the current water pressure using the pressure sensor 105, and the current water depth is inversely resolved by the computing device 207. The neural network denoising and recognition module adopts an encoding-decoding structure, combined with feature learning and pattern matching, to perform real-time recognition and denoising on the image.

[0038] This diving device integrates an optical imaging module, combining a vortex beam illumination component with a neural network. This allows for effective capture of target scene images in highly turbid water, while utilizing a deep neural network for image denoising and target recognition, effectively improving the system's recognition range in turbid water. Furthermore, the system utilizes an external pressure sensor and an inverse algorithm to locate underwater targets at depth. This application addresses the technical issues of related technologies, such as poor imaging quality, poor target recognition accuracy, and the inability to accurately locate targets in turbid water environments.

[0039] like Figure 1As shown, this embodiment illustrates the structure of an implementable diving device. This diving device includes two transparent windows: a light outlet 101, a light receiving port 102, a pair of horizontal propellers 103, a pair of vertical propellers 104, and a pressure sensor 105. The light outlet 101 and light receiving port 102 are located at the front of the diving device, the pair of horizontal propellers 103 are symmetrically arranged on either side of the device, the pair of vertical propellers 104 are symmetrically arranged between the upper and lower surfaces of the device, and the pressure sensor 105 is located on the upper surface of the device. Furthermore, a communication module, including both wireless and wired communication modules, can be integrated into the diving device to transmit captured images to a computing device 207, serving as a host computer.

[0040] like Figure 2 As shown, the optical imaging module includes a laser 201. The laser emitted by the laser 201 is expanded by a beam expander 202 and then irradiated onto a spiral phase plate 203, where it is modulated into a vortex beam. The vortex beam irradiates an underwater target 204, and the reflected light is collected by a lens 205 and imaged onto a CCD camera 206. The image is transmitted to a computing device 207, and after preprocessing, it is transmitted to a neural network to output a denoised image and target identification information. The pressure sensor 105 collects pressure information in real time and transmits it to the computing device 207 to display the water depth of the diving equipment in real time.

[0041] like Figure 3 As shown, the neural network denoising recognition module uses the data-driven EnhanceNet neural network. The EnhanceNet neural network has an encoder-decoder structure. During the encoding phase, the input image undergoes multiple compression operations to extract image features. Each compression operation passes through two convolution modules and downsamples through a 2×2 max pooling layer. The convolution module consists of a 3×3 convolution and a Reluctant Unit (ReLU) activation function. After the compression operation, the features are flattened into one dimension and input into the attention module, which consists of eight consecutive Transformer layers. Each Transformer layer consists of a multi-head self-attention layer (MSA) and a multi-layer perceptron (MLP). Both the MSA and MLP layers undergo layer normalization to stabilize the training process and use residual connections. The addition of the attention mechanism enables global, multi-angle, and deep feature extraction, improving the robustness of the neural network.

[0042] The denoising decoder gradually restores the spatial resolution of the image through multi-layer upsampling and skip connections. Each level of decoding includes feature concatenation, convolution, nonlinear activation, and batch normalization, ultimately outputting an image of the same size as the input, achieving noise suppression and detail restoration. The network uses mean squared error as the loss function:

[0043]

[0044] Where y i is the true value of the i-th pixel, is the predicted value of the i-th pixel, and N is the total number of pixels in the image.

[0045] The classification decoder flattens the feature information of each layer during the upsampling process and then splices it together and inputs it into the fully connected layer. The fully connected layer contains 3 hidden layers and an output layer. The output of the output layer is a vector of length 10, which represents the category probability distribution. Finally, the network outputs the denoised image and classification information.

[0046] Based on the above system, this application also proposes an intelligent optical perception method for imaging through underwater turbid environments, including the following steps:

[0047] Step 1: Illuminate the underwater target 204 with the vortex beam in the optical imaging module, specifically:

[0048] Step 1.1: placing a fish model as a target object 204 in turbid water;

[0049] Step 1.2: The laser 201 emits a 532 nm wavelength laser beam which is expanded by the beam expander 202 and then irradiated onto the spiral phase plate 203 for modulation to generate vortex light carrying different orbital angular momentum modes, which is then irradiated onto the target 204.

[0050] Step 2: Acquire the original image of the target object 204 and pre-process it through the image pre-processing module;

[0051] Step 2.1: The light reflected from the target 204 is focused by the lens 205 and formed into an image on the detection surface of the CCD camera 206;

[0052] Step 2.2: The original image is captured by the CCD camera 206 and transmitted to the computing device 207;

[0053] Step 2.3: Preprocess the original image received by the computing device 207, estimate the collected original image as the product of the illumination component and the reflection component, use Gaussian filtering to estimate the illumination information, and then obtain the illumination information through operations such as frequency domain convolution and inverse Fourier transform to inversely resolve the reflection information; normalize the reflection information, and then process the normalized reflection information through an adaptive histogram equalization algorithm to enhance contrast and improve local details.

[0054] The specific calculation process of preprocessing is as follows: Use the Retinex algorithm to separate the lighting information and restore the reflected image. The image information in the Retinex algorithm is modeled as:

[0055] I(x,y)=L(x,y)·R(x,y);

[0056] Where I(x,y) represents the original image, L(x,y) represents the illumination component, and R(x,y) represents the reflection component;

[0057] Step 2.4: Estimate lighting information using Gaussian filtering:

[0058]

[0059] Where G σ (x, y) is the lighting information, σ is the standard deviation, which determines the smoothness of the lighting information;

[0060] Perform frequency domain convolution and inverse Fourier transform to obtain lighting information:

[0061]

[0062] Where, is the spectrum information of lighting, represents Fourier transform;

[0063] Step 2.5: Estimate the reflection information from the image modeling in step 2.3:

[0064] logR(x,y)=logI(x,y)-logL(x,y);

[0065] Normalize the reflection information after restoring the brightness range:

[0066] R′(x,y)=exp(logR(x,y));

[0067]

[0068] Where R′(x,y) is the reflection information after restoring the brightness range, and R″(x,y) is the normalized reflection information;

[0069] Step 2.6: Use the adaptive histogram equalization algorithm in the MATLAB library to process the normalized reflection information to further enhance the contrast and improve local details:

[0070] I0(x,y)=adapthisteq(R″(x,y));

[0071] Where I0(x,y) is the enhanced image in the preprocessing process, and adapthisteq is the adaptive histogram equalization algorithm.

[0072] Step 3: Enhance and identify the target object image based on the data-driven EnhanceNet neural network. Specifically, use the EnhanceNet neural network to process the enhanced image obtained in step 2.6 to obtain a denoised image and classify the image targets.

[0073] Step 4: Real-time water depth detection, specifically:

[0074] Step 4.1: The pressure sensor 105 collects pressure in real time, returns a voltage value, and transmits it to the computing device 207;

[0075] Step 4.2: The computing device 207 determines the underwater pressure using the sensor calibration relationship and determines the current sensor depth using the pressure formula:

[0076] P = k·V + b;

[0077]

[0078] Where P is the underwater pressure, V is the voltage value returned by the sensor, k is the sensor sensitivity coefficient, b is the zero drift, k and b are calibrated by water pressure before the experiment, ρ is the water density, P atm is the current atmospheric pressure, which is calibrated before the experiment; the program converts the returned voltage value into water depth h in real time and outputs image annotation in real time.

[0079] The following experiments verify the validity of this application:

[0080] To verify the imaging performance of the system under varying turbidity levels and target distances (i.e., the distance between the target and the system), a corresponding experimental environment was established. The system position was kept fixed (at a depth of 1 meter from the water surface) and tested in a 4-meter-long flume by varying the turbidity (ranging from 10 to 80 NTU at a target distance of 1 meter) and the target distance (ranging from 0.5 to 4 meters at a turbidity of 15 NTU). To ensure consistency across different experimental conditions, the focal length was adjusted throughout the experiment to maintain a consistent target image size.

[0081] Through analysis Figure 4-6 It can be concluded that:

[0082] (1) Figure 4As shown in the figure, an underwater environment with turbidity increasing from 10 to 80 NTU is constructed at a target distance of 1 meter. As can be seen, the original image becomes blurry as the turbidity increases, and the target object is no longer distinguishable at 80 NTU. However, although the quality of the output image gradually decreases after system optimization, the target object is still clearly visible to the human eye in an underwater environment with a turbidity of 80 NTU. Furthermore, under the condition of a turbidity of 15 NTU, the imaging results of an underwater target are measured at target distances increasing from 0.5 to 4 meters. As can be seen, as the distance between the target and the system increases, the original image becomes blurry, and the target object is no longer distinguishable at a distance of 3 meters. In contrast, although the quality of the output image gradually decreases after system optimization, the target object is still clearly distinguishable to the human eye at a distance of 4 meters.

[0083] The results show that the system can not only cope with the strong scattering and absorption caused by the target distance, but also cope with the scattering noise caused by turbidity, and effectively suppress the noise in the original image directly obtained by the CCD camera.

[0084] 2) If Figure 5 As shown, the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) parameters of all images in underwater turbidity environments of 0-80 NTU and target distances of 0.25-4 meters are calculated.

[0085] At a constant target distance, the PSNR of the original image drops sharply from 35dB to 27dB as turbidity increases, then gradually decreases to 20dB between 20 and 80 NTU. Similarly, the SSIM of the original image drops sharply from 0.9 to 0.28. In contrast, for the optimized output image, although both PSNR and SSIM show a downward trend, the decrease is gradual. The PSNR gradually decreases from 35dB to 32dB, and the SSIM decreases by 0.05.

[0086] Furthermore, at a constant underwater turbidity, the PSNR of the original image drops sharply from 33dB to 25dB, then gradually to 16dB, as the target distance increases. Furthermore, its SSIM drops sharply from 0.85 to 0.33, then gradually to 0.18. In contrast, for the optimized output image, while target distance does affect image quality, the PSNR decreases less significantly, from 33dB to 27dB, and the SSIM decreases by only 0.08.

[0087] The results show that in an underwater environment with high turbidity and a long distance between the target and the system, the image quality after optimization of this system is improved by more than 10dB.

[0088] 3) If Figure 6As shown in the figure, a set of images is generated by extracting frames from a dynamic video at intervals to analyze and describe the imaging performance of the system during dynamic processes (target distance: 1.25 meters, turbidity: 25 NTU). It can be observed that in underwater turbidity environments, the system's output image quality of dynamic targets is stable, with no noticeable artifacts or fluctuations. The experimental results verify the reliability and stability of the system in practical application scenarios.

[0089] 4) We collected 500 scenes without objects and 1000 scenes with objects to verify the effective recognition ability of the system. The experimental results show that only 17 scenes without objects were incorrectly reported, and 15 scenes with objects were missed. After calculation, the false alarm rate of the system is as low as 1.7%, and the missed alarm rate is as low as 3.0%. In addition, the measured density of the water body is 1028kg / m 3 , the returned voltage value is 1.01V, and the calculated water depth of this system is 1.05m.

[0090] In summary, this application proposes for the first time a high-scattering medium imaging and positioning scheme based on structured light illumination and network denoising, and proposes a real-time, low-false-alarm-rate turbid underwater target recognition scheme, providing strong technical support for emergency command and underwater rescue.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent optical perception system for imaging underwater turbid environments, characterized by: The system comprises a computing device and an optical imaging module, an image acquisition module and a pressure sensor mounted on the diving equipment. The computing device runs a computer program including an image preprocessing module, a depth perception module and a neural network denoising and recognition module. The optical imaging module generates a structured light field carrying an orbital angular momentum pattern based on a spiral phase plate. The image acquisition module obtains a reflected image of a target object after being irradiated by vortex light through a high-sensitivity imaging device and transmits the image to the computing device. The image preprocessing module performs illumination and reflection information separation processing based on the Retinex model, frequency domain Gaussian filtering illumination estimation and adaptive histogram equalization enhancement on the collected image. The depth perception module obtains the current water pressure using a pressure sensor, and the current water depth is inversely resolved by the computing device. The neural network denoising recognition module performs real-time recognition and denoising of images.

2. The intelligent optical perception system for imaging underwater turbid environments according to claim 1, characterized in that: The optical imaging module includes a laser, a beam expander and a spiral phase plate. The laser emitted by the laser is expanded by the beam expander and irradiated onto the spiral phase plate, where it is modulated to generate vortex light carrying different orbital angular momentum patterns. The vortex light beam is then irradiated onto the underwater target.

3. The intelligent optical perception system for imaging underwater turbid environments according to claim 2, characterized in that: The image acquisition module includes a lens and a CCD camera. The vortex light beam is irradiated onto the underwater target. The reflected light is collected by the lens and imaged onto the CCD camera. The CCD camera sends the collected image to the computing device.

4. The intelligent optical perception system for imaging underwater turbid environments according to claim 1, characterized in that: The neural network denoising and recognition module uses the data-driven EnhanceNet neural network, which has an encoder-decoder structure. During the encoding phase, the input image undergoes multiple compression operations to extract image features. Each compression operation passes through two convolution modules and is downsampled through a maximum pooling. After the compression operation, the features are flattened into one dimension and input into the attention module. The attention module consists of multiple consecutive Transformer layers, each of which contains a multi-head self-attention layer and a multi-layer perceptron layer. Both the multi-head self-attention layer and the multi-layer perceptron layer are first normalized to stabilize the training process, and residual connections are used. The network decoding part contains two decoders, a denoising decoder and a classification decoder, which implement denoising and classification operations on the input image; The denoising decoder gradually restores the spatial resolution of the image through multi-layer upsampling and skip connections. Each level of decoding includes feature splicing, convolution, nonlinear activation and batch normalization, and finally outputs an image of the same size as the input, achieving noise suppression and detail restoration. The classification decoder flattens the feature information of each layer in the upsampling process and then splices them together, inputs them into the fully connected layer, outputs multiple classifications at the image level, and finally the network outputs the denoised image and image classification information.

5. The intelligent optical perception system for imaging underwater turbid environments according to any one of claims 1 to 4, characterized in that: The diving equipment includes two transparent windows: a light outlet and a light receiving port, a pair of horizontal propulsion propellers, a pair of vertical propulsion propellers and a pressure sensor, wherein the light outlet and the light receiving port are arranged in the front of the diving equipment, the pair of horizontal propulsion propellers are symmetrically arranged on both sides of the diving equipment, the pair of vertical propulsion propellers are symmetrically arranged in the middle of the upper and lower surfaces of the diving equipment, and the pressure sensor is arranged on the upper surface of the diving equipment.

6. An intelligent optical perception method for imaging through an underwater turbid environment, based on the intelligent optical perception system for imaging through an underwater turbid environment according to any one of claims 1 to 5, characterized in that: The following steps are involved: Step 1: Illuminate the underwater target through the vortex beam in the optical imaging module; Step 2: Collect the original image of the target object and preprocess it through the image preprocessing module; Step 3: Enhance and identify the target image through the data-driven EnhanceNet neural network; Step 4: Real-time detection and calculation of water depth.

7. The intelligent optical perception method for imaging in an underwater turbid environment according to claim 6, characterized in that: Step 2 is as follows: Step 2.1: The light reflected from the target is focused by the lens and formed into an image on the detection surface of the CCD camera; Step 2.2: The original image is captured by the CCD camera and transmitted to the computing device; Step 2.3: Preprocess the raw image received by the computing device. Estimate the collected raw image as the product of the illumination component and the reflection component. Use Gaussian filtering to estimate the illumination information. Then, use frequency domain convolution and inverse Fourier transform operations to obtain the illumination information for inverse reflection information. Normalize the reflection information and then process the normalized reflection information using an adaptive histogram equalization algorithm to enhance contrast and improve local details.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to claim 6 or 7.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to claim 6 or 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to claim 6 or 7 are implemented.

Images