A method for predicting the turbidity of deep-sea mining sediment plumes based on digital image processing

Abstract

The present invention provides a method for predicting the turbidity of deep-sea mining sediment plumes based on digital image processing. Based on the principle of different reflectivities of light by seawater and deep-sea mining sediment plumes, a set of methods for predicting the turbidity of deep-sea mining sediment plumes based on digital image processing technology is established. Through program calculation, this algorithm performs a series of processes on the video images of deep-sea mining sediment plumes, such as graying the images of deep-sea mining sediment plumes, image enhancement, and image threshold segmentation, etc., and finally extracts the characteristic values of the images of deep-sea mining sediment plumes. The characteristic values of the images of deep-sea mining sediment plumes can be calibrated with in-situ observation data to realize the inversion of the true turbidity of deep-sea mining sediment plumes.

Description

Technical Field

[0001] The present invention relates to the technical field of ocean engineering geology. Specifically, it particularly relates to a method for predicting the turbidity of deep-sea mining sediment plumes based on digital image processing. Background Art

[0002] In recent years, with the popularization of high-definition camera systems carried by ROVs and subsea observation networks, computer vision-based environmental monitoring technologies have begun to be applied in the deep-sea field. Existing studies mostly adopt traditional methods such as image grayscale histogram statistics or RGB color space conversion, and estimate by establishing an empirical relationship between pixel intensity and turbidity. However, the complex optical environment in the deep sea causes problems such as color deviation attenuation in images (typical attenuation rate: about 0.2 - 0.5 dB per meter in the red light band) and interference from scattering of suspended particles, making it difficult to accurately model the non-linear relationship between traditional image features and turbidity parameters. For example, research by Kawamura et al. (2021) shows that when the suspended solid concentration exceeds 200 mg / L, the prediction error of the traditional HSV color model can reach ±40 NTU. Therefore, the development of new visual feature extraction methods and adaptive prediction models is an urgent need. Summary of the Invention

[0003] To make up for the deficiencies of the prior art, the present invention provides a method for predicting the turbidity of deep-sea mining sediment plumes based on digital image processing. Through the high-definition video image data of deep-sea mining obtained from long-term in-situ observations in the deep sea, the target deep-sea mining sediment plume can be visually observed. However, only relying on the naked eye cannot quantitatively obtain the turbidity of the deep-sea mining sediment plume, and digital image processing technology can provide a solution for the post-processing of the collected underwater video images.

[0004] The present invention is realized through the following technical solutions: A method for predicting the turbidity of deep-sea mining sediment plumes based on digital image processing, specifically including the following steps:

[0005] Step S1, Feature Extraction: Multi-modal Feature Fusion

[0006] First, the size of the deep-sea mining video image is 1080*1920. The obtained high-resolution deep-sea mining video image is processed. The video is frame-divided, and then image processing is performed every 90 frames, including: calculating the average value and standard deviation of the three channels of the RGB image, as well as the average value and variance after grayscale conversion, calculating the entropy of the image, and the contrast of the image. Finally, the video frame image is converted to the HSV space, and the average values of the H, S, and V channels of the image in the HSV space are calculated. Finally, a Gaussian mixture model is used for image detection of the sediment plume and calculation of the plume area ratio, totaling 14 features;

[0007] The RBR multi-parameter water quality instrument observes the turbidity of the water body. After calculating the eigenvalue of the image, it is necessary to determine the observation area of the turbidity probe of the RBR multi-parameter water quality instrument for in-situ observation, that is, area Ⅰ. Then, according to the coordinates of area Ⅰ on the image: 100, 100, a square with a side length of 200, where the length of the image is 1920 and the width is 1080, with the upper left corner as the origin and positive to the right and downwards; extract the corresponding eigenvalues of 14 features in this area;

[0008] Step S2. Model construction: Adaptive deep learning framework

[0009] Step S2-1. Data preprocessing:

[0010] In order to increase the generalization ability of the model and improve the richness of data,

[0011] The principle and function of using data enhancement for data augmentation are: improving the adaptability of the model to real scenarios through diverse training data; adding Gaussian noise (σ = 0.05) to the original sonar signal to simulate deep-sea suspended particle interference;

[0012] The principle and function of using regularization for data regularization are: preventing overfitting by constraining the model complexity, adding a Dropout ratio = 0.3 after the convolutional layer, and applying L2 regularization λ = 0.001;

[0013] The principle and function of using expand dimension to increase the data dimension are: enhancing the feature expression ability by reconstructing the input data, segmenting the one-dimensional signal into multiple time windows to form a two-dimensional input of "time - segment", enhancing the ability to extract temporal features, and enhancing the model's ability to capture local features related to turbidity through multi-dimensional deconstruction and recombination of the spatial features of image data;

[0014] Step S2-2. Model structure selection:

[0015] Based on the grid search method, the optimal learning rate of 0.001, batch size of 32, and number of iterations of 200 are determined, and an early stopping mechanism is introduced: patience value = 10 rounds to prevent training redundancy; a 6-layer convolutional neural network is adopted, with batch normalization Batch Normalization() after each convolutional neural network layer, L2 regularization λ = 0.001 and Dropout ratio = 0.3 are used to suppress overfitting; and a max pooling layer Maxpooling1D(2), the activation function is ReLU, and the output layer is a linear regression unit;

[0016] Step S2-3. Model performance evaluation

[0017] Three indicators, the coefficient of determination R2, the mean squared error MSE, and the mean absolute error MAE, are used to evaluate the performance of the prediction model; the calculation formulas are as follows:

[0018]

[0019]

[0020]

[0021] Refers to the actual turbidity, the actually measured turbidity of the i-th water sample;

[0022] Refers to the predicted turbidity by the model, the turbidity predicted by the model based on the image features of the i-th water sample;

[0023] Refers to the average value of the actual turbidity.

[0024] As an optimal solution, in step S1, the frame rate of the deep-sea mining video is 30, and the observation interval of the RBR turbidimeter is 3 s.

[0025] As an optimal solution, the calculation method of the 14 features in step S1 is

[0026] Step S1-1: Calculate the average values of the three channels of the RGB image: R_mean, G_mean, B_mean, and the standard deviations: R_std, G_std, B_std;

[0027] The calculation formulas for the average values of the three channels of the RGB image are as follows:

[0028]

[0029]

[0030]

[0031] The calculation formulas for the standard deviations of the three RGB channels are as follows:

[0032]

[0033]

[0034]

[0035] Step S1-2, the average Gray_mean and variance Gray_variance after grayscale conversion: For the calculation of the grayscale mean and variance, first convert the RGB image into a grayscale image, and then calculate the average and variance of the grayscale:

[0036] First, convert the RGB image into a grayscale image, and the calculation formula is as follows:

[0037]

[0038] The calculation formula for the grayscale mean is as follows:

[0039]

[0040] The grayscale variance reflects the degree of dispersion of each pixel's grayscale value from the average grayscale value and is the main metric for measuring the amount of information in an image. The larger the variance, the greater the amount of information in the image. The variance calculation formula is as follows:

[0041]

[0042] Step S1-3, the averages of the H hue, S saturation, and V value channels: H_mean, S_mean, V_mean, and the calculation formulas are as follows:

[0043]

[0044] Step S1-4, Contrast: Calculated based on the Gray Level Co-occurrence Matrix (GLCM), representing the local grayscale difference;

[0045] Grayscale brightness Luminance, L:

[0046]

[0047] Using the standard deviation method RMS Contrast, the contrast is measured by the standard deviation of the brightness:

[0048]

[0049]

[0050] Step S1-5, the Entropy of the image: Reflecting the information complexity, the calculation formula is

[0051]

[0052] where p(x) is the probability distribution of pixel intensity;

[0053] Step S1-6, use the Gaussian Mixture Model (GMM) to segment the plume region and calculate the proportion of the plume area:

[0054] Ratio = Number of plume pixels / Total number of pixels.

[0055] Due to the adoption of the above technical solutions, the present invention has the following beneficial effects compared with the prior art:

[0056] Based on the principle of different light reflectivities of seawater and deep-sea mining sediment plumes, the present invention uses digital image processing technology to establish a method for predicting the turbidity of deep-sea mining sediment plumes based on digital image processing technology. This algorithm performs a series of processes on the video images of deep-sea mining sediment plumes through program calculation, such as: grayscale conversion of the deep-sea mining sediment plume images, image enhancement, and image threshold segmentation, etc., and finally extracts the characteristic values of the deep-sea mining sediment plume images. The characteristic values of the deep-sea mining sediment plume images can be calibrated with in-situ observation data to achieve the inversion of the true turbidity of the deep-sea mining sediment plumes.

[0057] By comparing the fitting effects of different models on the deep-sea mining sediment plume images, the optimization of the algorithm is achieved, and finally the best image prediction method for the turbidity of deep-sea mining sediment plumes is established. It can be applied to the on-site prediction of the turbidity of deep-sea mining sediment plumes. This method attempts to apply digital image processing technology to the measurement of the turbidity of deep-sea mining sediment plumes. Its measurement accuracy only depends on the shooting accuracy of the measurement device and the improvement of the subsequent algorithm. Compared with the traditional method for measuring the turbidity of deep-sea mining sediment plumes, it has the advantages of intuitive, continuous observation, and simple operation. Therefore, it has broad application prospects in marine surveys.

[0058] The present invention applies digital image processing technology to the measurement of the turbidity of deep-sea mining plumes. The measurement accuracy of this method only depends on the shooting accuracy of the measurement device and the improvement of the subsequent algorithm, overcoming the principle defects of traditional deep-sea mining plume turbidity measurement technologies, such as being easily affected by the particle size of suspended sediments in the ocean, the particle size distribution and flocculation of deep-sea mining plumes, etc. At the same time, it has the advantages of intuitive, continuous observation, and simple operation. While ensuring the measurement accuracy, the present invention significantly reduces the cost and operation complexity of deep-sea turbidity monitoring, providing an innovative technical means for environmental protection in deep-sea resource development. Therefore, it has broad application prospects in marine surveys. The promotion of this method of digital image processing technology to the field of deep-sea mining plume turbidity also reflects the intersection of multiple disciplinary fields and has good research reference significance.

[0059] 1. Multi-modal feature fusion: Combining RGB, HSV, texture, and statistical features to comprehensively characterize the optical properties of the plume, breaking through the limitations of a single color space.

[0060] 2. Adaptive deep learning framework: Solving the non-linear mapping problem between features and turbidity in the complex deep-sea optical environment through the non-linear modeling ability of CNN.

[0061] 3. Low-cost and high-resolution monitoring: Relying only on video images and algorithm optimization to replace traditional high-cost sensor arrays, the monitoring range can be extended to a kilometer-level working surface.

[0062] 4. Real-time and continuous: Support online video stream processing, achieve turbidity updates in seconds, and meet the requirements of dynamic environment monitoring for deep-sea mining.

[0063] The additional aspects and advantages of the present invention will become apparent in the following description section or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0065] Figure 1 Schematic diagram of the result of segmenting the plume area by the Gaussian mixture model (GMM);

[0066] Figure 2 Schematic diagram of model construction;

[0067] Figure 3 Schematic diagram of the convolutional neural network structure;

[0068] Figure 4 Schematic diagram of the performance evaluation results (coefficient of determination R2, mean square error MSE, and mean absolute error MAE). DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] In order to more clearly understand the above objects, features, and advantages of the present invention, the present invention will be further described in detail below in conjunction with the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0070] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0071] The following combines Figures 1 to 3 to specifically describe the method for predicting the turbidity of deep-sea mining sediment plumes based on digital image processing in the embodiments of the present invention.

[0072] The present invention provides a method for predicting the turbidity of deep-sea mining sediment plumes based on digital image processing, specifically including the following steps:

[0073] Step S1, feature extraction: Multimodal feature fusion

[0074] First, the size of the deep - sea mining video image is 1080*1920. Based on digital image processing technology, the present invention processes the acquired high - resolution deep - sea mining video image, performs frame - by - frame processing on the video, and then performs image processing on every 90th frame (the in - situ observation video has a frame rate of 30 and the RBR turbidity meter observation interval is 3 s), including: calculating the average value and standard deviation of the three channels of the RGB image, as well as the average value and variance after grayscale conversion, calculating the entropy of the image, and the contrast of the image. Finally, the video frame image is converted to the HSV space, and the average values of the H, S, and V channels of the image in the HSV space are calculated. Finally, the Gaussian mixture model is used for image detection of sediment plumes and calculation of the plume area ratio, totaling 14 features;

[0075] The calculation methods of the 14 features are as follows

[0076] Step S1 - 1: Calculate the average values of the three channels of the RGB image: R_mean, G_mean, B_mean, and the standard deviations: R_std, G_std, B_std;

[0077] Calculate the average values and standard deviations of the three channels of the RGB image of the deep - sea mining plume recorded by a high - definition camera at a depth of 5700 meters in the deep sea.

[0078] The calculation formula for the average values of the three channels of the RGB image is as follows:

[0079]

[0080] The calculation formula for the standard deviation of the RGB three channels is as follows: The standard deviation is a statistic that measures the degree of dispersion of pixel values in an image, describes the variation range of pixel values in the image, and can be used to evaluate the contrast and detail richness of the image. The larger the standard deviation, the more extensive the distribution of pixel values in the image and the stronger the contrast.

[0081]

[0082] Step S1 - 2: The average value (Gray_mean) and variance (Gray_variance) after grayscale conversion: The calculation of the grayscale average value and variance first requires converting the RGB image to a grayscale image, and then calculating the average value and variance of the grayscale:

[0083] First, convert the RGB image to a grayscale image, and the calculation formula is as follows:

[0084]

[0085] The calculation formula for the grayscale average value is as follows:

[0086]

[0087] The variance of grayscale, which reflects the degree of dispersion of the grayscale values of each pixel from the average grayscale value, is the main metric for measuring the amount of information in an image. The larger the variance, the greater the amount of information in the image. The variance calculation formula is as follows:

[0088]

[0089] Step S1-3: The average values of the H (hue), S (saturation), and V (value) channels: H_mean, S_mean, V_mean. The calculation formulas are as follows:

[0090]

[0091] Step S1-4: Contrast: Calculated based on the gray-level co-occurrence matrix (GLCM), representing local gray-scale differences;

[0092] Gray-scale luminance (Luminance, L)

[0093]

[0094] Using the standard deviation method (RMS Contrast), the contrast can be measured by the standard deviation of luminance:

[0095]

[0096] Step S1-5: The entropy of the image: Reflecting information complexity, the calculation formula is

[0097]

[0098] where p(x) is the probability distribution of pixel intensity;

[0099] Step S1-6: Use the Gaussian mixture model (GMM) to segment the plume area and calculate the ratio of the plume area: Ratio = number of plume pixels / total number of pixels.

[0100] During in-situ observations of deep-sea mining, there are not only high-resolution seabed video image observations but also RBR multi-parameter water quality instrument observations (which can observe the turbidity of water bodies). Therefore, after calculating the eigenvalue of the image, it is necessary to determine the observation area of the turbidity probe of the RBR multi-parameter water quality instrument for in-situ observations, that is, Region I. Then, according to the coordinates of Region I on the image: 100, 100, a square with a side length of 200, where the length of the image is 1920 and the width is 1080, with the upper left corner as the origin and right and down as positive; extract the 14 corresponding eigenvalues of the features in this area, as shown in Table 1;

[0101] Table 1: Feature selection of training values and predicted values, training values, validation values, predicted values

[0102] Variables datapoints Training phase (0.7) Turbidity(NTU) 15160*0.7 Mean (RGB) 3*15160*0.7 Standard deviation 3*15160*0.7 average_gray, 15160*0.7 entropy 15160*0.7 Contrast 15160*0.7 variance_gray 15160*0.7 Mean(HSV) 3*15160*0.7 ratio 15160*0.7 Testing phase (0.3) Turbidity (NTU) 15160*0.3 Mean (RGB) 3*15160*0.3 Standard deviation 3*15160*0.3 average_gray, 15160*0.3 entropy 15160*0.3 Contrast 15160*0.3 variance_gray 15160*0.3 Mean (HSV) 3*15160*0.3 ratio 15160*0.3 validation phase (1) Mean (RGB) 3*15160 Standard deviation 3*15160 average_gray, 15160 entropy 15160 Contrast 15160 variance_gray 15160 Mean (HSV) 3*15160 ratio 15160 Turbidity (NTU) 15160 predicted value output

[0103] Step S2, Model Construction: Adaptive Deep Learning Framework

[0104] After extracting the above data, the eigenvalue of the image is used as the independent variable, and the turbidity value obtained by in-situ observation is used as the dependent variable. There is a certain correlation between the two. We attempt to use a convolutional neural network to simulate the relationship between them, establish a model for predicting water turbidity from the image eigenvalue, and evaluate the model performance using R2.

[0105] Step S2-1, Data Preprocessing:

[0106] In order to increase the generalization ability of the model and enhance the richness of the data,

[0107] The principle and function of using data enhancement for data augmentation are as follows: By diversifying the training data, the adaptability of the model to real scenarios is improved; Gaussian noise (σ = 0.05) is added to the original sonar signal to simulate the interference of deep-sea suspended particles; its core function is not simply to increase the amount of data, but to simulate the signal interference caused by factors such as suspended particles in the real deep-sea environment, and thus "train" the model. The ultimate goal of doing this is to enhance the adaptability and prediction (or analysis) accuracy of the model in the actual, complex, and noisy deep-sea operation environment. By allowing the model to "practice" in advance how to process signals with interference, the reliability and robustness of the model in real application scenarios can be significantly improved, thereby obtaining more credible analysis results regarding deep-sea suspended particles.

[0108] The principle and function of using regularization for data regularization are as follows: By constraining the model complexity to prevent overfitting, a Dropout ratio of 0.3 is added after the convolutional layer, and L2 regularization λ = 0.001 is applied; the ultimate goal is to construct a model with stronger generalization ability, that is, when applied to new and real (deep-sea) images or sonar data, it can give more accurate and reliable prediction or analysis results.

[0109] The principle and function of using expand dimension to increase the data dimension are as follows: By reconstructing the input data, the feature expression ability is enhanced. The one-dimensional signal is segmented into multiple time windows to form a two-dimensional input of "time - segment", enhancing the ability to extract temporal features. By deconstructing and recombining the spatial features of the image data in multiple dimensions, the ability of the model to capture local features related to turbidity is enhanced;

[0110] Step S2-2, Model Structure Selection: To build a model that can accurately and automatically estimate the turbidity level (turbidity value, in NTU) of water bodies solely by analyzing water images, we selected Convolutional Neural Network (CNN) as the core technology. CNN is particularly adept at automatically learning and extracting visual features relevant to the target task from images. In terms of water turbidity, these features may include the depth of water color, transparency, patterns of light scattering, texture of visible suspended solids or particulate matter, and so on.

[0111] However, the performance of a CNN model depends largely on its "structure" design - such as how many processing units (convolutional layers, pooling layers) it contains, how many "filters" (convolution kernels) are used in each layer to detect different visual patterns, and the size of these "filters", etc. Different structural combinations will affect the effectiveness of the model in capturing turbidity-related information in the image. An overly simple structure may fail to capture the complex visual cues that distinguish subtle turbidity differences; while an overly complex structure may require more data and computing resources and is prone to "memorizing" specific details of the training images (including irrelevant noise), resulting in poor prediction performance for new, unseen water images (this is called overfitting).

[0112] Therefore, finding an "optimal" network structure is crucial, which needs to be able to best learn the mapping relationship between the visual features of water images and the actual turbidity values. To this end, we adopted two key strategies:

[0113] Grid Search for Structure Optimization: The aim is to systematically explore different CNN structure possibilities to find the configuration that is most suitable for extracting turbidity-related information from water images. A series of candidate structural parameters are predefined (for example, try using 2, 3, or 4 convolutional layers; try using 16, 32, or 64 convolutional kernels in each layer; try convolutional kernel sizes of 3x3 or 5x5, etc.). Grid search will automatically try all (or some selected) combinations of these parameters like permutations and combinations. For each structural combination, we use the same batch of water image data with known turbidity values to train the model and evaluate the accuracy of its predicted turbidity (for example, R², MSE, MAE). By comparing the model performance under different structures, grid search helps us find the CNN structure configuration that can most accurately infer the turbidity value from the visual cues in the image. This avoids blind guessing or relying solely on experience, making the structure selection process more scientific and data-driven.

[0114] Early Stopping for Generalization: The purpose is to prevent the model from overfitting to specific images in the training data (even including irrelevant noise or lighting conditions in the images), ensuring that the model learns general turbidity judgment rules that can be applied to new water body images. During model training, we not only adjust the model parameters using training images but also periodically use a batch of validation images that the model has never "seen" to test its current turbidity prediction performance. We continuously monitor the prediction error of the model on the validation images. Once it is found that the prediction error of the model on the validation images no longer decreases or even starts to increase (even if its error on the training images may still be decreasing), we immediately stop the training. This indicates that the model has learned enough general turbidity discrimination features, and continuing to train is likely to start learning details unique to the training data that are not conducive to generalization. Through early stopping, we can obtain a model with stronger generalization ability, which not only performs well on the trained images but, more importantly, can also give relatively reliable and stable turbidity prediction results when encountering new water body images taken in different environments in the future, which is crucial for the practical application of the model (such as field water quality monitoring).

[0115] Determine the optimal learning rate of 0.001, batch size of 32, and number of iterations of 200 based on the grid search method, and introduce the early stopping mechanism: patience value = 10 rounds to prevent training redundancy; finally, determine to use a 6-layer convolutional neural network, with batch normalization Batch Normalization() after each layer of the convolutional neural network, use L2 regularization λ = 0.001 and Dropout ratio = 0.3 to suppress overfitting; and a max pooling layer Maxpooling1D(2), use the ReLU activation function, and the output layer is a linear regression unit;

[0116] Step S2-3, Model Performance Evaluation

[0117] Use the coefficient of determination R 2 , mean squared error MSE, and mean absolute error MAE as three metrics to evaluate the performance of the prediction model; the specific meanings and calculation formulas of the three metrics are as follows:

[0118] Coefficient of determination (R²): Measures the ability of the model to explain the variability of the data (target value > 0.85), indicating to what extent the model explains the changes in the actual turbidity, and focuses on the goodness of fit and explanatory power of the model. It shows to what extent the image features can capture the variation law of turbidity. If R² is very low, it means that the image features themselves may not be sufficient to predict turbidity well, or the model structure is not properly selected.

[0119] Mean Squared Error (MSE): Quantifies the average of the squared errors between the model's predicted values and the true values. It calculates the error between the true turbidity and the predicted turbidity for each water body sample, squares these errors, and then computes the average of these squared errors. The unit of the Mean Squared Error (MSE) is the square of the turbidity unit (NTU). 2 , since the calculation involves the squared term of the error between the true turbidity value and the predicted turbidity value, it is particularly sensitive to large turbidity prediction errors. One or several extreme prediction errors (such as a very low predicted turbidity for a very turbid water body) will significantly increase the value of MSE.

[0120] Mean Absolute Error (MAE): Quantifies the average of the absolute errors between the model's predicted turbidity values and the true turbidity values. It calculates the absolute value of the error between the true turbidity value and the predicted turbidity value and then takes the average. The unit of the Mean Absolute Error (MAE) is the same as the turbidity unit (NTU). If MAE = 5 NTU, it indicates that the difference between the turbidity value predicted by the model and the actually measured turbidity value is 5 NTU (it could be 5 NTU higher or 5 NTU lower). MAE gives the same weight to all magnitudes of turbidity prediction differences and does not amplify the influence of extreme errors, thus better reflecting the general average prediction accuracy of the model. The smaller the MAE, the smaller the average deviation of the turbidity value predicted by the model from the true value.

[0121]

[0122] refers to the actual turbidity, the actually measured turbidity of the i-th water sample;

[0123] refers to the predicted turbidity, the turbidity predicted by the model based on the image features of the i-th water sample;

[0124] refers to the average value of the actual turbidity.

[0125] In the description of the present invention, the term "a plurality of" refers to two or more, unless otherwise clearly defined. The orientation or positional relationship indicated by terms such as "upper", "lower", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention; terms such as "connection", "installation", "fixation", etc. should all be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected, or indirectly connected through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0126] In the description of this specification, the description of terms such as "one embodiment", "some embodiments", "specific embodiments", etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0127] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the turbidity of deep - sea mining sediment plumes based on digital image processing, characterized in that , which specifically includes the following steps: Step S1, Feature Extraction: Multimodal Feature Fusion; First, the size of the deep-sea mining video image is 1080*1920. Process the obtained high-resolution deep-sea mining video image, perform frame division on the video, and then perform image processing on every 90th frame, including: calculating the average value and standard deviation of the three channels of the RGB image, as well as the average value and variance after grayscale conversion, calculating the entropy of the image, and the contrast of the image. Finally, convert the video frame image to the HSV space and calculate the average values of the H, S, and V channels of the image in the HSV space. Finally, use the Gaussian mixture model for image detection of sediment plumes and calculation of the plume area ratio, totaling 14 features; The RBR multi-parameter water quality instrument observes the turbidity of the water body. After calculating the eigenvalue of the image, it is necessary to determine the observation area of the turbidity probe of the RBR multi-parameter water quality instrument for in-situ observation, that is, Region I. Then, extract the corresponding eigenvalue of the 14 features of this region according to the coordinates of Region I on the image. The coordinates of Region I on the image are (100, 100), a square with a side length of 200 pixels, where the length of the image is 1920 and the width is 1080, with the upper left corner of the image as the origin and positive to the right and downwards; Step S2, Model Construction: Adaptive Deep Learning Framework; Step S2-1, Data Preprocessing: In order to increase the generalization ability of the model and enhance the richness of the data, The principle and function of using data enhancement for data enhancement are: improving the model's adaptability to real scenarios by diversifying the training data; adding Gaussian noise σ = 0.05 to the original sonar signal to simulate deep-sea suspended particle interference; The principle and function of using regularization for data regularization are: preventing overfitting by constraining the model complexity, adding a Dropout ratio = 0.3 after the convolutional layer, and applying L2 regularization λ = 0.001; The principle and function of using expand dimension to increase the data dimension are: enhancing the feature expression ability by reconstructing the input data, segmenting the one-dimensional signal into multiple time windows to form a "time-segment" two-dimensional input, enhancing the ability to extract temporal features, and enhancing the model's ability to capture local features related to turbidity by multi-dimensional deconstruction and recombination of the spatial features of the image data; Step S2-2, Model Structure Selection: Based on the grid search method, determine the optimal learning rate of 0.001, batch size of 32, and number of iterations of 200, and introduce an early stopping mechanism: patience value = 10 rounds to prevent training redundancy; adopt a 6-layer convolutional neural network, with batch normalization Batch Normalization after each layer of the convolutional neural network, adopt L2 regularization λ = 0.001 and Dropout ratio = 0.3 to suppress overfitting; and a max pooling layer Maxpooling1D(2), use the ReLU activation function, and the output layer is a linear regression unit; Step S2-3, Model Performance Evaluation: Using the coefficient of determination R 2 , mean squared error MSE, and mean absolute error MAE to evaluate the performance of the prediction model; the calculation formulas are as follows: ， ， ， refers to the actual turbidity, the actually measured turbidity of the i-th water sample; Refers to the predicted turbidity by the model, which is the turbidity predicted by the model based on the image features of the i-th water sample; Refers to the average value of the true turbidity.

2. The method for predicting the turbidity of deep-sea mining sediment plume based on digital image processing according to claim 1, wherein , In step S1, the frame rate of the deep - sea mining video is 30, and the observation interval of the RBR turbidimeter is 3 s.

3. The method for predicting the turbidity of deep-sea mining sediment plume based on digital image processing according to claim 1, wherein , In step S1, the calculation methods of the 14 features are as follows: Step S1 - 1: Calculate the average values of the three channels of the RGB image: R_mean, G_mean, B_mean, and the standard deviations: R_std, G_std, B_std; The calculation formula for the average values of the three channels of the RGB image is as follows: ， ， ， The calculation formula for the standard deviations of the RGB three channels is as follows: ， ， ， Step S1 - 2: The average value Gray_mean and variance Gray_variance after grayscale conversion: The calculation of the grayscale mean and variance first requires converting the RGB image into a grayscale image, and then calculating the average value and variance of the grayscale: First, convert the RGB image into a grayscale image, and the calculation formula is as follows: ， The calculation formula for the grayscale mean is as follows: ， The grayscale variance reflects the degree of dispersion of each pixel's grayscale value from the average grayscale value and is the main metric for measuring the amount of information in an image. The larger the variance, the greater the amount of information in the image; the variance calculation formula is as follows: ， Step S1 - 3: The average values of the H hue, S saturation, and V value channels: H_mean, S_mean, V_mean, and the calculation formula is as follows: ， Step S1 - 4: Contrast: Calculated based on the gray - level co - occurrence matrix GLCM, representing the local gray - level difference; Gray - level luminance Luminance, L: ， Using the standard deviation method RMS Contrast, the contrast is measured by the standard deviation of the luminance: ， ， Step S1 - 5: The entropy Entropy of the image: Reflects the information complexity, and the calculation formula is: ， where p(x) is the probability distribution of pixel intensity; Step S1 - 6: Use the Gaussian mixture model GMM to segment the plume area and calculate the proportion of the plume area: Ratio = number of plume pixels / total number of pixels.

Images