Underwater operation safety assessment scheme construction method and system based on deep learning

Abstract

The invention discloses an underwater operation safety assessment scheme construction method and system based on deep learning. The method comprises the following steps: acquiring basic physiological indexes of A simulators before entering a cabin and underwater operation physiological indexes corresponding to different depths, different operation intensities and different operation time when the A simulators enter the cabin to simulate underwater; each simulator calculates a relative physiological index and an operation safety score to obtain each piece of feature data of each simulator, and the operation safety level of each piece of feature data is divided; dividing all the feature data into three feature sets, for each feature set, performing matrix sorting according to a first priority of an operation time sampling point time sequence, a second priority of an operation intensity from low to high, and a third priority of a depth from shallow to deep, and selectively expanding each matrix after sorting to obtain m pieces of expanded feature data; screening and constructing an expanded feature set corresponding to each feature set, and constructing a training set and a verification set; and training and verifying the underwater operation safety evaluation model by using the training set and the verification set to obtain a target underwater operation safety evaluation model.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater operation safety assessment, and particularly to a method and system for constructing an underwater operation safety assessment scheme based on deep learning. Background Art

[0002] Due to the complex and changeable underwater environment, underwater operators are often accompanied by great risks and prone to safety accidents. It can be seen that diving operations are high-risk operations. Therefore, it is particularly important to conduct underwater operation safety assessments for underwater operators.

[0003] In order to conduct underwater operation safety assessments for underwater operators, a pressure simulation chamber is needed to simulate the underwater environment by pressurization. For example, the pressure simulation chamber is pressurized to the pressure corresponding to a certain underwater depth, and the underwater operation simulator simulates the movement corresponding to a certain operation intensity at this underwater depth, and collects the physiological signals of the underwater operation simulator for a period of time to obtain sample data for constructing a model. However, the existing technical problems are that the labor cost and economic cost of underwater operation simulation are very high, and the time consumption is long, which limits the number of underwater operation simulations, resulting in a small amount of sample data obtained. The model trained with less sample data is not accurate. Moreover, the existing use of regression models for underwater operation safety assessment is not accurate enough.

[0004] Based on this, the inventors of the present invention performed data expansion processing on the sample data to obtain a sufficient amount of sample data. Using the sufficient amount of sample data to train an underwater operation safety assessment model constructed using a deep learning model can obtain a more accurate model, thereby achieving a high accuracy rate for underwater operation safety assessment. Summary of the Invention

[0005] The present invention aims at the problems and deficiencies existing in the prior art, and provides a method and system for constructing an underwater operation safety assessment scheme based on deep learning.

[0006] The present invention solves the above technical problems through the following technical solutions:

[0007] The present invention provides a method for constructing an underwater operation safety assessment scheme based on deep learning, characterized by including:

[0008] S1. Collect the basic physiological indexes of A simulators before entering the pressure simulation chamber, and the underwater operation physiological indexes corresponding to different depths, different operation intensities, and different operation times underwater when entering the pressure simulation chamber;

[0009] S2. For each simulator, extract the underwater operation physiological indicators at each depth, each operation intensity, and each operation time sampling point, compare them with the basic physiological indicators, calculate the relative physiological indicators and operation safety scores, obtain the characteristic data of each simulator, and divide the operation safety levels of the characteristic data. The operation safety levels include three levels: high, medium, and low.

[0010] S3. Divide all the characteristic data into three characteristic sets according to the operation safety levels. For each characteristic set, perform matrix sorting respectively according to the time sequence of the operation time sampling points as the first priority, the order of the operation intensity from low to high as the second priority, and the order of the depth from shallow to deep as the third priority. Selectively perform interpolation and add noise to each sorted matrix to obtain the corresponding m augmented characteristic data, where m takes a random value less than the number of rows of the matrix - 1 and greater than (less than the number of rows of the matrix - 1) / 3.

[0011] S4. For all the augmented characteristic data obtained corresponding to each characteristic set, screen and construct the augmented characteristic set corresponding to each characteristic set. The construction principle is that the data volume of each characteristic set is greater than the data volume of the corresponding augmented characteristic set randomly selected. Construct the training set and the validation set. Both the training set and the validation set contain the characteristic data of each characteristic set and the augmented characteristic data of each augmented characteristic set, and the data volume of each characteristic set is greater than the data volume of the corresponding augmented characteristic set.

[0012] S5. Use the training set and the validation set to train and validate the underwater operation safety assessment model constructed by the deep learning model respectively, and obtain the target underwater operation safety assessment model, which is used to conduct underwater operation safety assessment on underwater operation personnel.

[0013] The present invention also provides a system for constructing an underwater operation safety assessment scheme based on deep learning, characterized by including:

[0014] A data acquisition module, which is used to acquire the basic physiological indicators of A simulators before entering the pressure simulation chamber, and the underwater operation physiological indicators corresponding to different depths, different operation intensities, and different operation times in the pressure simulation chamber simulating underwater.

[0015] A feature extraction module, which is used to, for each simulator, extract the underwater operation physiological indicators at each depth, each operation intensity, and each operation time sampling point, compare them with the basic physiological indicators, calculate the relative physiological indicators and operation safety scores, obtain the characteristic data of each simulator, and divide the operation safety levels of the characteristic data. The operation safety levels include three levels: high, medium, and low.

[0016] A feature expansion module, which is used to divide all feature data into three feature sets according to the job safety level. For each feature set, matrix sorting is performed respectively with the time sequence of job time sampling points as the first priority, the order of job intensity from low to high as the second priority, and the order of depth from shallow to deep as the third priority. For each sorted matrix, interpolation and noise addition are selectively performed to obtain corresponding m augmented feature data, where m takes a random value less than the number of rows of the matrix - 1 and greater than (less than the number of rows of the matrix - 1) / 3;

[0017] A sample construction module, which is used to screen and construct an augmented feature set corresponding to each feature set for all the augmented feature data obtained for each feature set. The construction principle is that the data volume of each feature set is greater than the data volume of the corresponding randomly selected augmented feature set, and a training set and a validation set are constructed. Both the training set and the validation set contain the feature data of each feature set and the augmented feature data of each augmented feature set, and the data volume of each feature set is greater than the data volume of the corresponding augmented feature set;

[0018] A model construction module, which is used to train and validate an underwater operation safety assessment model constructed by a deep learning model using the training set and the validation set respectively, and obtain a target underwater operation safety assessment model for performing underwater operation safety assessment on underwater operation personnel.

[0019] The present invention also provides an electronic device, which is characterized by including:

[0020] A processor;

[0021] A memory for storing instructions executable by the processor;

[0022] Wherein, the processor is configured to call the instructions stored in the memory to execute the above method.

[0023] The present invention also provides a computer-readable storage medium, on which computer program instructions are stored. The computer program instructions are characterized in that when executed by a processor, the above method is implemented.

[0024] The positive and progressive effects of the present invention are as follows:

[0025] All the characteristic data simulated for underwater operations in the present invention are classified according to high, medium, and low operation safety levels, and three characteristic sets are classified. For each characteristic set, matrix sorting is performed respectively with the time sequence of operation time sampling points as the first priority, the order of operation intensity from low to high as the second priority, and the order of depth from shallow to deep as the third priority. After interpolation and adding noise to each sorted matrix, the augmented characteristic data is obtained, and then a sufficient amount of sample data is obtained. Subsequently, the training set and the validation set are constructed using the characteristic data and the augmented characteristic data. Both the training set and the validation set contain the characteristic data of each characteristic set and the augmented characteristic data of each augmented characteristic set, and the data volume of each characteristic set is greater than that of the corresponding augmented characteristic set. The underwater operation safety assessment model constructed by the deep learning model is trained and validated using the training set and the validation set respectively to obtain an accurate target underwater operation safety assessment model, which is used to accurately assess the underwater operation safety of underwater operation personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flowchart of a method for constructing an underwater operation safety assessment scheme according to a preferred embodiment of the present invention.

[0027] Figure 2 It is a block diagram of a system for constructing an underwater operation safety assessment scheme according to a preferred embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0029] As Figure 1 shown, an embodiment of the present invention provides a method for constructing an underwater operation safety assessment scheme based on deep learning, including the following steps:

[0030] Step 101: Collect the basic physiological indicators containing multiple basic physiological parameters of A simulators before entering the pressure simulation chamber, and the underwater operation physiological indicators containing multiple underwater operation physiological parameters corresponding to different depths, different operation intensities, and different operation times when simulating underwater in the pressure simulation chamber, where A is a positive integer.

[0031] In this step, first, collect the basic physiological indicators of A simulators in the resting state before entering the pressure simulation chamber, and then collect the underwater operation physiological indicators corresponding to different depths, different operation intensities, and different operation times of A simulators in the underwater operation simulation state after entering the pressure simulation chamber.

[0032] The basic physiological indicators include multiple basic physiological parameters, and the physiological indicators for underwater operations include multiple physiological parameters for underwater operations. The parameters are: electroencephalogram, heart rate, respiratory rate, blood pressure, blood oxygen saturation, and body temperature.

[0033] Step 102: For each simulator, extract the physiological indicators for underwater operations at each depth, each operation intensity, and each operation time sampling point, compare them with the basic physiological indicators, calculate the relative physiological indicators containing multiple relative physiological parameters and the operation safety score, obtain the characteristic data of each simulator, and each characteristic data (depth Dd, operation intensity Dp, operation time sampling point Dt, relative physiological indicator Dz, operation safety score Df), and classify the operation safety levels of each characteristic data. The operation safety levels include three levels: high operation safety level, medium operation safety level, and low operation safety level.

[0034] Step 103: Divide all the characteristic data into three characteristic sets according to the operation safety levels. The three characteristic sets are the characteristic set corresponding to the high operation safety level, the characteristic set corresponding to the medium operation safety level, and the characteristic set corresponding to the low operation safety level. For each characteristic set, perform matrix sorting according to the time sequence of the operation time sampling point as the first priority, the ascending order of the operation intensity as the second priority, and the ascending order of the depth as the third priority. Optionally, perform interpolation and add noise to each sorted matrix to obtain the corresponding m extended characteristic data. m takes a random value less than the number of rows in the matrix - 1 and greater than (less than the number of rows in the matrix - 1) / 3. Each extended characteristic data (extended depth Dd’, extended operation intensity Dp’, extended operation time sampling point Dt’, extended relative physiological indicator Dz’, extended operation safety score Df’). Among them, mean interpolation method is used for interpolation and Gaussian noise is used for adding noise.

[0035] In this step, for each characteristic set, perform multiple matrix sortings according to the time sequence of the operation time sampling point as the first priority. In each matrix, the rows with the same depth and the same operation intensity and the consecutive rows with the same sampling point under the same depth and the same operation intensity are sorted, and the different sampling points are sorted in the time sequence from top to bottom. Optionally, perform interpolation and add noise between adjacent rows in each sorted matrix to obtain m extended characteristic data, where the depth and the operation intensity remain unchanged, and interpolation and add noise are performed on the operation time sampling point, the relative physiological indicator, and the operation safety score.

[0036] For example: If A = 10, there are 10 simulators. The job time sampling points are every 5 minutes. The first column is depth, the second column is job intensity, the third column is job time sampling point, the fourth column is relative physiological index, and the fifth column is job safety score. The first 10 rows are the depth 1, job intensity 1, job time sampling point of 5 minutes, corresponding relative physiological index, and corresponding job safety score of 10 simulators. The second 10 rows are the depth 1, job intensity 1, job time sampling point of 10 minutes, corresponding relative physiological index, and corresponding job safety score of 10 simulators... The last 10 rows are the depth 1, job intensity 1, job time sampling point of 120 minutes, corresponding relative physiological index, and corresponding job safety score of 10 simulators.

[0037] In this step, interpolation and noise addition are performed between adjacent rows of each sorted matrix, but not for every adjacent row. Instead, interpolation and noise addition are selectively performed between adjacent rows to obtain m extended feature data. m takes a random value less than the number of rows of the matrix - 1 and greater than (less than the number of rows of the matrix - 1) / 3, so as to ensure that the extended feature data extended from the matrix is less than the feature data of the matrix. The value of m is such that the amount of extended data is appropriate, neither exceeding the amount of data in the matrix nor being too small.

[0038] For each feature set, multiple matrices are sorted in ascending order of job intensity as the second priority. In each matrix, rows with the same depth, the same job time sampling point, and the same job intensity at the same depth and the same job time sampling point are sorted continuously, and different job intensities are sorted from top to bottom in ascending order. Interpolation and noise addition are selectively performed between adjacent rows of each sorted matrix to obtain m extended feature data, where the depth and job time sampling point remain unchanged, and interpolation and noise addition are performed on job intensity, relative physiological index, and job safety score.

[0039] For each feature set, multiple matrices are sorted in ascending order of depth as the third priority. In each matrix, rows with the same job intensity, the same job time sampling point, and the same depth at the same job intensity and the same job time sampling point are sorted continuously, and different depths are sorted from top to bottom in ascending order. Interpolation and noise addition are selectively performed between adjacent rows of each sorted matrix to obtain m extended feature data, where job intensity and job time sampling point remain unchanged, and interpolation and noise addition are performed on depth, relative physiological index, and job safety score.

[0040] Step 104: For all the augmented feature data obtained corresponding to each feature set, screen and construct the augmented feature set corresponding to each feature set. The construction principle is that the data volume of each feature set is greater than the data volume of the corresponding randomly selected augmented feature set. Construct a training set and a validation set. Both the training set and the validation set contain the feature data of each feature set and the augmented feature data of each augmented feature set, and the data volume of each feature set is greater than the data volume of the corresponding augmented feature set.

[0041] After step 103, the data volume of each feature set is less than the data volume of the corresponding augmented feature set. Based on this, in this step, the data volume of the corresponding augmented feature set is screened according to the principle that the data volume of each feature set is greater than the data volume of the corresponding augmented feature set, so that the data volume of each feature set is greater than the data volume of the corresponding randomly selected augmented feature set. Moreover, when constructing the training set and the validation set, both the training set and the validation set contain the feature data of each feature set and the augmented feature data of each augmented feature set, and the data volume of each feature set is greater than the data volume of the corresponding augmented feature set. This operation makes the constructed training set and validation set relatively appropriate and accurate, and reduces the training error.

[0042] Step 105: Use the training set and the validation set to train and validate the underwater operation safety assessment model constructed by the deep learning model respectively, and obtain the target underwater operation safety assessment model, which is used to conduct underwater operation safety assessment on underwater operation personnel. Among them, the deep learning model adopts a recurrent neural network model.

[0043] In this step, first use the training set to preliminarily train the underwater operation safety assessment model, and then use the validation set to validate the trained underwater operation safety assessment model to obtain the preliminary fitness. If the preliminary fitness is not less than the set fitness, set the particle swarm size to n1, otherwise set the particle swarm size to n2, where n1 < n2. Then use the hyperparameters of the model after preliminary training and their randomly mutated hyperparameters to initialize the particle swarm, and use the particle swarm algorithm to optimize the model hyperparameters to obtain the optimal hyperparameters, and use the optimal hyperparameters to construct the target underwater operation safety assessment model.

[0044] As Figure 2 shown, an embodiment of the present invention also provides a system for constructing an underwater operation safety assessment solution based on deep learning, including a data acquisition module 1, a feature extraction module 2, a feature augmentation module 3, a sample construction module 4, and a model construction module 5.

[0045] The data acquisition module 1 is used to collect the basic physiological indicators containing multiple basic physiological parameters of A simulators before entering the pressure simulation chamber, and the underwater operation physiological indicators containing multiple underwater operation physiological parameters corresponding to different depths, different operation intensities, and different operation times when simulating underwater in the pressure simulation chamber.

[0046] The feature extraction module 2 is used to extract the underwater operation physiological indexes of each simulator at each depth, each operation intensity, and each operation time sampling point, compare them with the basic physiological indexes, calculate the relative physiological indexes containing multiple relative physiological parameters and the operation safety scores, obtain the feature data of each simulator (depth Dd, operation intensity Dp, operation time sampling point Dt, relative physiological index Dz, operation safety score Df), and divide the operation safety levels of the feature data. The operation safety levels include three levels: high, medium, and low.

[0047] The feature expansion module 3 is used to divide all the feature data into three feature sets according to the operation safety levels. For each feature set, matrix sorting is performed respectively according to the time sequence of the operation time sampling point as the first priority, the order from low to high of the operation intensity as the second priority, and the order from shallow to deep of the depth as the third priority. For each sorted matrix, interpolation and noise addition are selectively performed to obtain the corresponding m expanded feature data, where m takes a random value less than the number of rows of the matrix - 1 and greater than (less than the number of rows of the matrix - 1) / 3. Each expanded feature data (expanded depth, expanded operation intensity, expanded operation time sampling point, expanded relative physiological index, expanded operation safety score).

[0048] The sample construction module 4 is used to screen and construct the expanded feature set corresponding to each feature set for all the expanded feature data obtained corresponding to each feature set. The construction principle is that the data volume of each feature set is greater than the data volume of the corresponding randomly selected expanded feature set, and the training set and the validation set are constructed. Both the training set and the validation set contain the feature data of each feature set and the expanded feature data of each expanded feature set, and the data volume of each feature set is greater than the data volume of the corresponding expanded feature set.

[0049] The model construction module 5 is used to train and validate the underwater operation safety assessment model constructed by the deep learning model by using the training set and the validation set respectively, and obtain the target underwater operation safety assessment model, which is used to conduct underwater operation safety assessment on underwater operation personnel.

[0050] An embodiment of the present invention also provides an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0051] An embodiment of the present invention also provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0052] The present invention can be a method, a device, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are uploaded.

[0053] Although the specific embodiments of the present invention have been described above, those skilled in the art should understand that these are only examples. The protection scope of the present invention is defined by the appended claims. Without departing from the principle and essence of the present invention, those skilled in the art can make various changes or modifications to these embodiments, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. A method for constructing an underwater operation safety assessment scheme based on deep learning, characterized by including: S1. Collect the basic physiological indicators of A simulators before entering the pressure simulation chamber, and the underwater operation physiological indicators corresponding to different depths, different operation intensities, and different operation times in the pressure simulation chamber to simulate underwater conditions; S2. For each simulator, extract the underwater operation physiological indicators at the sampling points of each depth, each operation intensity, and each operation time, compare them with the basic physiological indicators, calculate the relative physiological indicators and operation safety scores, obtain the characteristic data of each simulator, and classify the operation safety levels of the characteristic data. The operation safety levels include three levels: high, medium, and low; S3. Divide all the characteristic data into three feature sets according to the operation safety levels. For each feature set, perform matrix sorting according to the time sequence of the operation time sampling points as the first priority, the order of operation intensity from low to high as the second priority, and the order of depth from shallow to deep as the third priority. Optionally perform interpolation and add noise to each sorted matrix to obtain the corresponding m augmented feature data, where m takes a random value less than the number of rows of the matrix - 1 and greater than (less than the number of rows of the matrix - 1) / 3; S4. For all the augmented feature data obtained corresponding to each feature set, screen and construct the augmented feature set corresponding to each feature set. The construction principle is that the data volume of each feature set is greater than the data volume of the randomly selected corresponding augmented feature set. Construct a training set and a validation set. Both the training set and the validation set contain the feature data of each feature set and the augmented feature data of each augmented feature set, and the data volume of each feature set is greater than the data volume of the corresponding augmented feature set; S5. Use the training set and the validation set to train and validate the underwater operation safety assessment model constructed by the deep learning model respectively to obtain the target underwater operation safety assessment model, which is used to conduct underwater operation safety assessment on underwater operation personnel.

2. The method for constructing an underwater operation safety assessment scheme based on deep learning according to claim 1, characterized in that, In S3, for each feature set, perform multiple matrix sortings according to the time sequence of the operation time sampling points as the first priority. In each matrix, the rows with the same depth, the same operation intensity, and the same sampling point at the same depth and the same operation intensity are sorted continuously, and the different sampling points are sorted in the time sequence from top to bottom. Optionally perform interpolation and add noise between adjacent rows in each sorted matrix to obtain m augmented feature data, where the depth and the operation intensity remain unchanged, and interpolation and noise addition are performed on the operation time sampling points, relative physiological indicators, and operation safety scores.

3. The method for constructing an underwater operation safety assessment scheme based on deep learning according to claim 1, characterized in that, In S3, for each feature set, perform multiple matrix sortings according to the order of operation intensity from low to high as the second priority. In each matrix, the rows with the same depth, the same sampling point, and the same operation intensity at the same depth and the same sampling point are sorted continuously, and the different operation intensities are sorted from top to bottom from low to high. Optionally perform interpolation and add noise between adjacent rows in each sorted matrix to obtain m augmented feature data, where the depth and the operation time sampling points remain unchanged, and interpolation and noise addition are performed on the operation intensity, relative physiological indicators, and operation safety scores.

4. The method for constructing an underwater operation safety assessment scheme based on deep learning according to claim 1, characterized in that, In S3, for each feature set, multiple matrix sorts are performed for the third priority in the order from shallow to deep depth. In each matrix, the sampling points with the same operation intensity are the same, and the continuous rows with the same depth under the same operation intensity and the same sampling point are sorted, and the rows with different depths are sorted from top to bottom in the order from shallow to deep. Interpolation and noise addition are selectively performed between adjacent rows in each sorted matrix to obtain m augmented feature data, where the operation intensity and the operation time sampling points remain unchanged, and interpolation and noise addition are performed on the depth, relative physiological index, and operation safety score.

5. The method for constructing an underwater operation safety assessment scheme based on deep learning according to claim 1, characterized in that, In S5, the underwater operation safety assessment model is initially trained with the training set, and then the trained underwater operation safety assessment model is verified with the validation set to obtain the initial fitness. If the initial fitness is not less than the set fitness, the number of particle swarms is set to n1, otherwise the number of particle swarms is set to n2, where n1 < n2. Then, the hyperparameters of the model after initial training and their randomly mutated hyperparameters are used to initialize the particle swarm, and the particle swarm algorithm is used to optimize the hyperparameters of the model to obtain the optimal hyperparameters, and the target underwater operation safety assessment model is constructed using the optimal hyperparameters.

6. The method for constructing an underwater operation safety assessment scheme based on deep learning according to claim 1, characterized in that, In S3, mean interpolation method is used for interpolation, and Gaussian noise is used for noise addition. In S5, the deep learning model uses a recurrent neural network model.

7. The method for constructing an underwater operation safety assessment scheme based on deep learning according to claim 1, characterized in that, in In S1, the basic physiological indicators contain multiple basic physiological parameters, and the underwater operation physiological indicators contain multiple underwater operation physiological parameters, and the parameters are: electroencephalogram, heart rate, respiratory rate, blood pressure, blood oxygen saturation, and body temperature.

8. A system for constructing an underwater operation safety assessment scheme based on deep learning, characterized by including: A data acquisition module, configured to acquire the basic physiological indicators of A simulators before entering the pressure simulation cabin, and the underwater operation physiological indicators corresponding to different depths, different operation intensities, and different operation times when simulating underwater in the pressure simulation cabin; A feature extraction module, configured to, for each simulator, extract the underwater operation physiological indicators of each depth, each operation intensity, and each operation time sampling point of the simulator, compare them with the basic physiological indicators, calculate the relative physiological index and the operation safety score, obtain the respective feature data of each simulator, and divide the operation safety levels of the respective feature data, where the operation safety levels include three levels: high, medium, and low; A feature augmentation module, configured to divide all the feature data into three feature sets according to the operation safety level. For each feature set, matrix sorting is performed for the first priority according to the time sequence of the operation time sampling points, the second priority according to the order from low to high of the operation intensity, and the third priority according to the order from shallow to deep of the depth. Interpolation and noise addition are selectively performed on each sorted matrix to obtain the corresponding m augmented feature data, where m takes a random value less than the number of rows of the matrix - 1 and greater than (less than the number of rows of the matrix - 1) / 3; A sample construction module, which is used to screen and construct an augmented feature set corresponding to each feature set for all the augmented feature data obtained for each feature set. The construction principle is that the data volume of each feature set is greater than the data volume of the corresponding randomly selected augmented feature set, and a training set and a validation set are constructed. Both the training set and the validation set contain the feature data of each feature set and the augmented feature data of each augmented feature set, and the data volume of each feature set is greater than the data volume of the corresponding augmented feature set; A model construction module, which is used to train and validate an underwater operation safety assessment model constructed by a deep learning model by using the training set and the validation set respectively, and obtain a target underwater operation safety assessment model for performing underwater operation safety assessment on underwater operation personnel.

9. An electronic device, characterized in that, Including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1-7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method described in any one of claims 1-7 is implemented.

Images