A method and system for controlling underwater robot based on machine vision

Abstract

The invention relates to the technical field of underwater robot control, and specifically discloses an underwater robot control method and system based on machine vision. The system comprises: a data acquisition module, which is used to collect environmental data information around the current position of the underwater robot. By first combining the environmental data information around the current position of the robot, the environmental impact currently suffered by the underwater robot is analyzed to determine whether image shooting can be performed, so as to avoid blind shooting in a poor shooting environment, resulting in a large number of unqualified images, resulting in deviations in the subsequent extraction of feature information of a target object. Afterwards, whether the quality of each shot image is qualified is evaluated, so that high-quality images can be obtained for subsequent feature extraction, and unqualified images can be deleted in time to avoid the image occupying memory and reducing the algorithm capability of the robot, resulting in a slow speed of subsequent analysis operations.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater robot control, and in particular to an underwater robot control method and system based on machine vision. Background Art

[0002] An underwater robot based on machine vision is a device that can replace humans to complete certain tasks underwater. It has broad application prospects in underwater exploration, underwater rescue, underwater anti-fouling and other fields.

[0003] In practical applications, underwater robots are usually used to identify and locate underwater targets and determine their position information to implement target grasping tasks. When the underwater robot is in use, in order to ensure the normal implementation of its grasping task, the underwater robot is first controlled to capture the image of the underwater target underwater, and then the feature information of the target object is extracted through the image processing algorithm, and the distance between the underwater robot and the underwater target is obtained through the deep learning model, so as to facilitate the background management personnel to control the underwater robot to approach the underwater target and implement the grasping task.

[0004] In order to ensure the normal progress of the grasping task in the prior art, it is necessary to first control the underwater robot to take images of the underwater target underwater. However, since the images taken by the underwater robot will be affected by the environment, if the images are taken blindly, only a large number of images of unqualified quality will be obtained. This image will not only cause deviations when extracting the characteristic information of the target object through the image processing algorithm, resulting in a decrease in the accuracy of the subsequent distance estimation results between the underwater robot and the underwater target, but also occupy a large amount of memory, resulting in a decrease in the algorithm of the underwater robot, which slows down the subsequent analysis operations, thereby affecting the overall operation efficiency. Summary of the invention

[0005] The purpose of the present invention is to provide an underwater robot control method and system based on machine vision to solve the following technical problems:

[0006] How to improve the accuracy of extracting feature information of target objects and the overall work efficiency.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] A method and system for controlling an underwater robot based on machine vision, the system comprising:

[0009] A data acquisition module is used to collect environmental data information around the current position of the underwater robot;

[0010] The environmental assessment module is used to analyze the environmental impact of the underwater robot in combination with the environmental data collected by the data acquisition module, and determine whether image capture can be performed;

[0011] An image acquisition module is used to capture image data of underwater targets when it is determined that the underwater robot is currently subject to low environmental impact;

[0012] An image evaluation module is used to evaluate whether the quality of each image captured is qualified based on the appearance of the image captured by the image acquisition module;

[0013] The recognition module is used to extract the required features in the image when the image quality is judged to be high, and estimate the position distance data between the underwater robot and the underwater target based on the deep learning algorithm of the convolutional neural network;

[0014] The adjustment module adjusts and corrects the position distance data obtained in the recognition module by combining the data information collected by the data collection module.

[0015] Furthermore, the environmental data collected by the data collection module includes: impurity content of water around the current position of the underwater robot, water flow rate and light intensity.

[0016] Furthermore, the evaluation process of the environmental evaluation module includes:

[0017] By formula Calculate the underwater environmental impact coefficient t in the i-th time period i ;

[0018] Among them, i is any monitoring time period, st i is the impurity content of the water body in the i-th time period underwater, st y is the preset water impurity content, st b for st i The standard value of f m (x)≥x, then let f m (x) = x, otherwise, let f m (x) = 1, ls i is the average flow velocity of the water body in the i-th time period, ls y is the preset flow rate, gx i is the average light intensity in the i-th time period, gx y is the preset light intensity, gx b For gx i The standard value of θ i is the light irradiation angle influence coefficient, which is set according to empirical fitting, a is any time point in a time period, b is the total number of time points in a time period, and fh ai is the voltage load of the fill light at the ath time point in the i-th time period underwater, For all fh ai The average value of .

[0019] Furthermore, the assessment process of the environmental assessment module also includes:

[0020] By calculating the underwater environmental impact coefficient t in the i-th time period i and the preset underwater environmental impact coefficient threshold t 01 Make a comparison;

[0021] If i ≥t 01 ,The system determines that the underwater environment in the current time period has a great impact on the image shooting quality of the underwater robot, and does not perform shooting control operations;

[0022] If i <t 01 ,The system determines that the underwater environment in the current time period has little impact on the image shooting quality of the underwater robot, and the shooting control operation can be carried out.

[0023] Furthermore, the evaluation process of the image evaluation module includes:

[0024] When it is determined that the underwater environment in the current time period has little influence on the image shooting quality of the underwater robot, the underwater robot is controlled to shoot the image;

[0025] And through the formula Calculate the quality influence coefficient r of the cth picture taken in the i-th time period ci ;

[0026] Among them, c is any picture taken, wzx ci is the completeness of the cth picture taken in the i-th time period, wzx b For wzx ci The standard value of bx ci is the deformation area of ​​the cth picture taken in the i-th time period, ab ci is the dark spot area of ​​the cth picture taken in the i-th time period, yy ci is the shadow area of ​​the cth picture taken in the ith time period, sh y is the preset total damaged area, x1, x2 and x3 are the damaged area weight coefficients, which are set based on empirical fitting.

[0027] Furthermore, the evaluation process of the image evaluation module also includes:

[0028] By calculating the quality influence coefficient r of the cth picture taken in the i-th time period ci The preset quality influence coefficient threshold r 01 Make a comparison;

[0029] If r ci ≥r 01, the system determines that the quality of the photo is unqualified and deletes the photo to release memory space;

[0030] If r ci <r 01 , the system determines that the quality of the photograph is unqualified and can be used as a high-quality image for feature extraction.

[0031] Furthermore, the adjustment process of the adjustment module includes:

[0032] First, the required features in the image are extracted through the recognition module, and the position distance data p between the underwater robot and the underwater target in the i-th time period is estimated based on the deep learning algorithm of the convolutional neural network. 01 ;

[0033] And through the formula p i =p 01 *f w (t i *r ci ) Calculate the corrected position distance p between the underwater robot and the underwater target i ;

[0034] Among them, f w To adjust the coefficient comparison table function, according to the empirical data (t i *r ci ) The influence of the range of calculated result values ​​on the corrected position distance between the underwater robot and the underwater target is obtained based on tests.

[0035] A method for controlling an underwater robot based on machine vision, the method comprising:

[0036] S1: Collect environmental data information around the current position of the underwater robot through the data acquisition module;

[0037] S2: Analyze the environmental impact of the underwater robot through the environmental assessment module combined with the environmental data collected by the data acquisition module, and determine whether image capture can be performed. If yes, proceed to step S3, otherwise, continue analysis;

[0038] S3: When it is determined that the underwater robot is currently affected by a low level of environmental influence, the image acquisition module is used to capture image data of the underwater target;

[0039] S4: The image evaluation module is combined with the appearance of the image captured by the image acquisition module to evaluate whether the image quality of each shot is qualified. If yes, proceed to step S5; otherwise, continue shooting;

[0040] S5: Extract the required features in the image through the recognition module, and estimate the position distance data between the underwater robot and the underwater target based on the deep learning algorithm of the convolutional neural network;

[0041] S6: The position distance data obtained in the recognition module is adjusted and corrected by combining the adjustment module with the data information collected by the data collection module.

[0042] Beneficial effects of the present invention:

[0043] (1) The present invention first analyzes the environmental impact of the underwater robot in combination with the environmental data information around the current position of the robot, and determines whether image shooting can be performed. This can avoid blind shooting in a poor shooting environment, resulting in a large number of unqualified images, which will lead to deviations in the subsequent extraction of feature information of the target object. In addition, the quality of each captured image is evaluated to see if it is qualified, so that high-quality images can be obtained for subsequent feature extraction, and unqualified images can be deleted in time to avoid the situation where images occupy memory and reduce the robot's algorithm capabilities, resulting in a slowdown in the subsequent analysis operation.

[0044] (2) The present invention calculates the underwater environmental impact coefficient t in the i-th time period by i and the preset underwater environmental impact coefficient threshold t 01 By comparing the images, it is possible to make an accurate judgment on the impact of the underwater environment on the image shooting quality of the underwater robot in the current time period, and further decide whether the shooting control operation can be performed, thereby ensuring that the underwater robot can take pictures in a good shooting environment to ensure the quality of the captured images and avoid blindly shooting images, resulting in a large number of unqualified images, which will lead to errors in the subsequent extraction of feature information of the target object.

[0045] (3) The present invention calculates the quality influence coefficient r of the cth picture taken in the i-th time period ci The preset quality influence coefficient threshold r 01 By comparing, it is possible to make an accurate judgment on whether the quality of the taken photos is qualified, thereby ensuring that the subsequent extraction of target object feature information is based on high-quality image data, thereby ensuring the quality of feature extraction, and by deleting the photos in time to release memory space when the quality of the taken photos is judged to be unqualified, it can avoid waste films occupying a large amount of memory, resulting in the reduction of the algorithm of the underwater robot, and slowing down the subsequent analysis operations, thereby improving overall operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The present invention will be further described below in conjunction with the accompanying drawings.

[0047] Figure 1 It is a schematic block diagram of an underwater robot control system based on machine vision in the present invention;

[0048] Figure 2 It is a flow chart of a method for controlling an underwater robot based on machine vision in the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1 As shown, in one embodiment, the present application provides a method and system for controlling an underwater robot based on machine vision, the system comprising:

[0051] A data acquisition module is used to collect environmental data information around the current position of the underwater robot;

[0052] The environmental assessment module is used to analyze the environmental impact of the underwater robot in combination with the environmental data collected by the data acquisition module, and determine whether image capture can be performed;

[0053] An image acquisition module is used to capture image data of underwater targets when it is determined that the underwater robot is currently subject to low environmental impact;

[0054] An image evaluation module is used to evaluate whether the quality of each image captured is qualified based on the appearance of the image captured by the image acquisition module;

[0055] The recognition module is used to extract the required features in the image when the image quality is judged to be high, and estimate the position distance data between the underwater robot and the underwater target based on the deep learning algorithm of the convolutional neural network;

[0056] An adjustment module adjusts and corrects the position distance data obtained in the recognition module by combining the data information collected by the data collection module;

[0057] Through the above technical scheme, this embodiment provides a data acquisition module for collecting environmental data information around the current position of the underwater robot. Then, the environmental assessment module is combined with the environmental data collected by the data acquisition module to analyze the environmental impact currently suffered by the underwater robot, and it is determined whether image shooting can be performed. When it is determined that the environmental impact currently suffered by the underwater robot is low, the image data of the underwater target is shot. Then, the image assessment module is combined with the appearance of the image captured by the image acquisition module to evaluate whether the image quality of each shot is qualified, and when it is determined that the image quality is high, the required features in the image are extracted, and the position distance data of the underwater robot and the underwater target are estimated based on the deep learning algorithm of the convolutional neural network. Finally, the position distance data obtained in the recognition module can be adjusted and corrected by combining the data information collected by the data acquisition module.

[0058] Through the above technical scheme, when controlling the underwater robot, by first combining the environmental data information around the current position of the robot, the environmental impact on the underwater robot is analyzed to determine whether image shooting can be performed. This can avoid blind shooting in poor shooting environments, resulting in a large number of unqualified images, which will lead to deviations in the subsequent extraction of feature information of the target object. In addition, the image quality of each shot is evaluated to see if it is qualified, so that high-quality images can be obtained for subsequent feature extraction, and unqualified images can be deleted in time to avoid the situation where images occupy memory and reduce the robot's algorithm capabilities, resulting in a slowdown in subsequent analysis operations.

[0059] The environmental data collected by the data collection module include: impurity content, water flow rate and light intensity of the water body around the current position of the underwater robot;

[0060] Through the above technical scheme, this embodiment provides environmental data collected by the data acquisition module, including the water impurity content, water flow rate and light intensity around the current position of the underwater robot, wherein too high water impurity content will affect the propagation of light and the clarity of the image, turbid water will cause light scattering, making it difficult to accurately estimate the distance between objects, and the water flow rate will cause the camera to shake, thereby affecting the stability of the image, and the light intensity will affect the contrast and color reproduction of the image, and refraction and scattering in the water will cause the image to be blurred. Therefore, by combining this data, accurate data support can be provided for the subsequent calculation of the underwater environmental impact coefficient, thereby improving the accuracy of the calculation results.

[0061] The assessment process of the environmental assessment module includes:

[0062] By formula Calculate the underwater environmental impact coefficient t in the i-th time period i ;

[0063] Among them, i is any monitoring time period, st i is the impurity content of the water body in the i-th time period underwater, st y is the preset water impurity content, st b for st i The standard value can be selected and set according to the allowable error in the empirical data. m (x)≥x, then let f m (x) = x, otherwise, let f m (x) = 1, ls i is the average flow velocity of the water body in the i-th time period, ls y is the preset flow rate, gx i is the average light intensity in the i-th time period, gx y is the preset light intensity, gx b For gx i The standard value of θ can be selected and set according to the allowable error in empirical data. i is the light irradiation angle influence coefficient, which is set according to empirical fitting, a is any time point in a time period, b is the total number of time points in a time period, and fh ai is the voltage load of the fill light at the ath time point in the i-th time period underwater, For all fh ai The average value of

[0064] Through the above technical solution, this embodiment provides the underwater environment impact coefficient t in the i-th time period: i , can be obtained by formula Calculated, where the formula The voltage load fluctuation value of the fill light in the i-th time period can be calculated. This data can reflect the lighting stability of the fill light in the i-th time period from the side. Therefore, when the impurity content of the water body in the i-th time period underwater is higher than the preset value, the difference between the average light intensity in the i-th time period and the preset value is greater, the average flow rate of the water body in the i-th time period is faster and the voltage load fluctuation value of the fill light in the i-th time period is greater, then the underwater environmental impact coefficient t of the i-th time period is i The larger the value, the greater the environmental impact will be, which means that the image shooting in this time period will be greatly affected by the environment, resulting in unqualified image quality; conversely, when the impurity content of the water body in the i-th time period is lower than the preset value, the difference between the average light intensity in the i-th time period and the preset value is smaller, the average flow rate of the water body in the i-th time period is slower, and the voltage load fluctuation value of the fill light in the i-th time period is smaller, then the underwater environmental impact coefficient t in the i-th time period is iThe smaller it is, it means that the image shooting in this time period will not be greatly affected by the environment, and the image quality will be relatively easy to meet the standards.

[0065] The assessment process of the environmental assessment module also includes:

[0066] By calculating the underwater environmental impact coefficient t in the i-th time period i and the preset underwater environmental impact coefficient threshold t 01 Make a comparison;

[0067] If i ≥t 01 ,The system determines that the underwater environment in the current time period has a great impact on the image shooting quality of the underwater robot, and does not perform shooting control operations;

[0068] If i <t 01 ,The system determines that the underwater environment in the current time period has little impact on the image shooting quality of the underwater robot, and the shooting control operation can be carried out;

[0069] Through the above technical solution, this embodiment calculates the underwater environment impact coefficient t i and the preset underwater environmental impact coefficient threshold t 01 By comparing the images, it is possible to make an accurate judgment on the impact of the underwater environment on the image shooting quality of the underwater robot in the current time period, and further decide whether the shooting control operation can be performed, thereby ensuring that the underwater robot can take pictures in a good shooting environment to ensure the quality of the captured images and avoid blindly shooting images, resulting in a large number of unqualified images, which will lead to errors in the subsequent extraction of feature information of the target object.

[0070] The evaluation process of the image evaluation module includes:

[0071] When it is determined that the underwater environment in the current time period has little influence on the image shooting quality of the underwater robot, the underwater robot is controlled to shoot the image;

[0072] And through the formula Calculate the quality influence coefficient r of the cth picture taken in the i-th time period ci ;

[0073] Among them, c is any picture taken, wzx ci is the completeness of the cth picture taken in the i-th time period, wzx b For wzx ci The standard value of bx can be set based on the allowable error in the empirical data. ci is the deformation area of ​​the cth picture taken in the i-th time period, abci is the dark spot area of ​​the cth picture taken in the i-th time period, yy ci is the shadow area of ​​the cth picture taken in the ith time period, sh y is the preset total damaged area, x1, x2 and x3 are the damaged area weight coefficients, which are set based on empirical fitting;

[0074] Through the above technical solution, this embodiment provides the quality influence coefficient r of the cth picture taken in the i-th time period: ci , can be obtained by formula Obviously, when the integrity of the cth picture taken in the i-th time period is lower, and the dark spot area, deformation area and shadow area of ​​the cth picture taken in the i-th time period are larger, then the quality influence coefficient r of the cth picture taken in the i-th time period is ci The higher the image quality is, the lower the image quality is and the image cannot be used as a qualified image for subsequent target object feature information extraction. On the contrary, when the integrity of the cth image taken in the i-th time period is higher and the dark spot area, deformation area and shadow area of ​​the cth image taken in the i-th time period are smaller, the quality influence coefficient r of the cth image taken in the i-th time period is ci The lower the value, the higher the quality of the image, and it can be used as a qualified image for subsequent target object feature information extraction.

[0075] The quality influence coefficient r of the cth picture taken in the i-th time period is calculated by combining the appearance information of the taken image ci , which can provide accurate data support for the subsequent judgment of whether the photos are qualified, thereby improving the accuracy of the judgment results.

[0076] The evaluation process of the image evaluation module also includes:

[0077] By calculating the quality influence coefficient r of the cth picture taken in the i-th time period ci The preset quality influence coefficient threshold r 01 Make a comparison;

[0078] If r ci ≥r 01 , the system determines that the quality of the photo is unqualified and deletes the photo to release memory space;

[0079] If r ci <r 01 ,The system determines that the quality of the photograph is unqualified and can be used as a high-quality image for feature extraction;

[0080] Through the above technical solution, this embodiment calculates the quality influence coefficient r of the cth picture taken in the i-th time period ciThe preset quality influence coefficient threshold r 01 By comparing, it is possible to make an accurate judgment on whether the quality of the taken photos is qualified, thereby ensuring that the subsequent extraction of target object feature information is based on high-quality image data, thereby ensuring the quality of feature extraction, and by deleting the photos in time to release memory space when the quality of the taken photos is judged to be unqualified, it can avoid waste films occupying a large amount of memory, resulting in the reduction of the algorithm of the underwater robot, and slowing down the subsequent analysis operations, thereby improving overall operation efficiency.

[0081] The adjustment process of the adjustment module includes:

[0082] First, the required features in the image are extracted through the recognition module, and the position distance data p between the underwater robot and the underwater target in the i-th time period is estimated based on the deep learning algorithm of the convolutional neural network. 01 ;

[0083] And through the formula p i =p 01 *f w (t i *r ci ) Calculate the corrected position distance p between the underwater robot and the underwater target i ;

[0084] Among them, f w To adjust the coefficient comparison table function, according to the empirical data (t i *r ci ) The influence of the range of values ​​of the calculation results on the position distance between the underwater robot and the underwater target after correction is obtained based on the test;

[0085] Through the above technical solution, this embodiment provides the corrected position distance p between the underwater robot and the underwater target. i , can be obtained by formula p i =p 01 *f w (t i *r ci ) is calculated, through the above technical solution, this embodiment combines the underwater environment impact coefficient t of the i-th time period i The quality influence coefficient r of the cth picture taken in the i-th time period ci For the position distance data p between the underwater robot and the underwater target in the i-th time period 01 Correction can ensure that the data is closer to the real data, thereby improving the accuracy of underwater robot control.

[0086] See also Figure 2 As shown, a method for controlling an underwater robot based on machine vision comprises:

[0087] S1: Collect environmental data information around the current position of the underwater robot through the data acquisition module;

[0088] S2: Analyze the environmental impact of the underwater robot through the environmental assessment module combined with the environmental data collected by the data acquisition module, and determine whether image capture can be performed. If yes, proceed to step S3, otherwise, continue analysis;

[0089] S3: When it is determined that the underwater robot is currently affected by a low level of environmental influence, the image acquisition module is used to capture image data of the underwater target;

[0090] S4: The image evaluation module is combined with the appearance of the image captured by the image acquisition module to evaluate whether the image quality of each shot is qualified. If yes, proceed to step S5; otherwise, continue shooting;

[0091] S5: Extract the required features in the image through the recognition module, and estimate the position distance data between the underwater robot and the underwater target based on the deep learning algorithm of the convolutional neural network;

[0092] S6: adjusting and correcting the position distance data obtained in the recognition module by combining the data information collected by the data collection module through the adjustment module;

[0093] Through the above technical scheme, this embodiment provides an underwater robot control method based on machine vision, firstly, the environmental data information around the current position of the underwater robot is collected through the data acquisition module, and then the environmental assessment module is combined with the environmental data collected by the data acquisition module to analyze the environmental impact currently suffered by the underwater robot, and it is judged whether image shooting can be performed. When it is judged that the environmental impact currently suffered by the underwater robot is low, the image data of the underwater target is shot through the image acquisition module, and the image assessment module is combined with the appearance of the image shot by the image acquisition module to evaluate whether the image quality of each shot is qualified, and then the required features in the high-quality image are extracted through the recognition module, and the position distance data of the underwater robot and the underwater target are estimated based on the deep learning algorithm of the convolutional neural network, and finally the position distance data obtained in the recognition module is adjusted and corrected through the adjustment module combined with the data information collected by the data acquisition module;

[0094] Through the above technical scheme, by first combining the environmental assessment module with the environmental data collected by the data acquisition module, the environmental impact currently on the underwater robot is analyzed, so as to avoid blind shooting in poor shooting environments, resulting in a large number of unqualified images, which leads to deviations in the subsequent extraction of feature information of the target object. After that, the image evaluation module is combined with the appearance of the image captured by the image acquisition module to evaluate whether the quality of each image shot is qualified, which can ensure the quality of the captured image, thereby ensuring that the subsequent extraction of underwater target features is based on high-quality image data, thereby ensuring the accuracy of feature extraction.

[0095] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. An underwater robot control system based on machine vision, characterized in that: The system comprises: A data acquisition module is used to collect environmental data information around the current position of the underwater robot; The environmental assessment module is used to analyze the environmental impact of the underwater robot in combination with the environmental data collected by the data acquisition module, and determine whether image capture can be performed; An image acquisition module is used to capture image data of underwater targets when it is determined that the underwater robot is currently subject to low environmental impact; An image evaluation module is used to evaluate whether the quality of each image captured is qualified based on the appearance of the image captured by the image acquisition module; The recognition module is used to extract the required features in the image when the image quality is judged to be high, and estimate the position distance data between the underwater robot and the underwater target based on the deep learning algorithm of the convolutional neural network; The adjustment module adjusts and corrects the position distance data obtained in the recognition module by combining the data information collected by the data collection module.

2. The underwater robot control system based on machine vision according to claim 1, characterized in that: The environmental data collected by the data collection module include: impurity content of water around the current position of the underwater robot, water flow rate and light intensity.

3. The underwater robot control system based on machine vision according to claim 2, characterized in that: The assessment process of the environmental assessment module includes: By formula Calculate the underwater environmental impact coefficient t in the i-th time period i ; Among them, i is any monitoring time period, st i is the impurity content of the water body in the i-th time period underwater, st y is the preset water impurity content, st b for st i The standard value of f m (x)≥x, then let f m (x) = x, otherwise, let f m (x) = 1, ls i is the average flow velocity of the water body in the i-th time period, ls y is the preset flow rate, gx i is the average light intensity in the i-th time period, gx y is the preset light intensity, gx b For gx i The standard value of θ i is the light irradiation angle influence coefficient, which is set according to empirical fitting, a is any time point in a time period, b is the total number of time points in a time period, and fh ai is the voltage load of the fill light at the ath time point in the i-th time period underwater, For all fh ai The average value of .

4. The underwater robot control system based on machine vision according to claim 3, characterized in that: The assessment process of the environmental assessment module also includes: By calculating the underwater environmental impact coefficient t in the i-th time period i and the preset underwater environmental impact coefficient threshold t 01 Make a comparison; If i ≥t 01 ,The system determines that the underwater environment in the current time period has a great impact on the image shooting quality of the underwater robot, and does not perform shooting control operations; If i <t 01 ,The system determines that the underwater environment in the current time period has little impact on the image shooting quality of the underwater robot, and the shooting control operation can be carried out.

5. The underwater robot control system based on machine vision according to claim 4, characterized in that: The evaluation process of the image evaluation module includes: When it is determined that the underwater environment in the current time period has little influence on the image shooting quality of the underwater robot, the underwater robot is controlled to shoot the image; And through the formula Calculate the quality influence coefficient r of the cth picture taken in the i-th time period ci ; Among them, c is any picture taken, wzx ci is the completeness of the cth picture taken in the i-th time period, wzx b for wzx ci The standard value of bx ci is the deformation area of ​​the cth picture taken in the ith time period, ab ci is the dark spot area of ​​the cth picture taken in the i-th time period, yy ci is the shadow area of ​​the cth picture taken in the ith time period, sh y is the preset total damaged area, x1, x2 and x3 are the damaged area weight coefficients, which are set based on empirical fitting.

6. The underwater robot control system based on machine vision according to claim 5, characterized in that: The evaluation process of the image evaluation module also includes: By calculating the quality influence coefficient r of the cth picture taken in the i-th time period ci The preset quality influence coefficient threshold r 01 Make a comparison; If r ci ≥r 01 , the system determines that the quality of the photo is unqualified and deletes the photo to release memory space; If r ci <r 01 , the system determines that the quality of the photograph is unqualified and can be used as a high-quality image for feature extraction.

7. The underwater robot control system based on machine vision according to claim 1, characterized in that: The adjustment process of the adjustment module includes: First, the required features in the image are extracted through the recognition module, and the position distance data p between the underwater robot and the underwater target in the i-th time period is estimated based on the deep learning algorithm of the convolutional neural network. 01 ; And through the formula p i =p 01 *f w (t i *r ci ) Calculate the corrected position distance p between the underwater robot and the underwater target i ; Among them, f w To adjust the coefficient comparison table function, according to the empirical data (t i *r ci ) The influence of the range of calculated result values ​​on the corrected position distance between the underwater robot and the underwater target is obtained based on tests.

8. A method for controlling an underwater robot based on machine vision, the method adopting an underwater robot control system based on machine vision as claimed in claims 1 to 7, characterized in that: The method comprises: S1: Collect environmental data information around the current position of the underwater robot through the data acquisition module; S2: Analyze the environmental impact of the underwater robot through the environmental assessment module combined with the environmental data collected by the data acquisition module, and determine whether image capture can be performed. If yes, proceed to step S3, otherwise, continue analysis; S3: When it is determined that the underwater robot is currently affected by a low level of environmental influence, the image acquisition module is used to capture image data of the underwater target; S4: The image evaluation module is combined with the appearance of the image captured by the image acquisition module to evaluate whether the image quality of each shot is qualified. If yes, proceed to step S5; otherwise, continue shooting; S5: Extract the required features in the image through the recognition module, and estimate the position distance data between the underwater robot and the underwater target based on the deep learning algorithm of the convolutional neural network; S6: The position distance data obtained in the recognition module is adjusted and corrected by combining the adjustment module with the data information collected by the data collection module.

Images