Underwater robot sample identification and grasping method and system

Abstract

The present invention provides a method and system for identifying and grasping samples by an underwater robot, the method comprising: the underwater robot uses a binocular camera to acquire collected image information in real time, and uses a target detection model to detect the number and type of target samples in the image information in real time; based on the image information, the posture information of the identified target sample type underwater is judged, and action instructions of a sampling mechanism corresponding to the target sample type are automatically generated based on the target sample type and the posture information; based on the binocular camera, the depth information of the target sample of the corresponding type underwater relative to the binocular camera is automatically detected, and the action posture of the corresponding sampling mechanism is automatically adjusted based on the depth information of the target sample and the posture information of the target sample; if sampling is successful, the depth information of all samples identified in the image information is acquired, and the sample with the smallest depth information is selected as the new target sample for sampling operation.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater robots, and in particular to a method and system for identifying and grasping samples by underwater robots. Background Art

[0002] At present, underwater robots mainly use a single camera for image acquisition for sample identification and grasping, identify the target sample through image recognition, and grasp it by manually driving the robotic arm. The above-mentioned existing technologies mainly have the following technical problems: 1. The grasping of the robotic arm requires human participation and control, so the underwater robot needs to ensure the stability of the remote control signal underwater. Since the signal is subject to interference from aquatic organisms, water flow, water depth and the electric field of underwater equipment, the stability of the remote control signal is poor. 2. The monocular camera cannot identify the spatial distance between the target sample and the underwater robot, so it is necessary to drive the robot engine, robotic arm and other related equipment multiple times for manual real-time adjustment to achieve accurate positioning and grasping. Summary of the Invention

[0003] One of the inventive purposes of the present invention is to provide a method and system for identifying and grasping samples by an underwater robot. The method and system utilize the robot's target detection technology to identify and locate samples, and automatically generate driving codes for the robot's corresponding sampling mechanism based on the identification and positioning results, thereby realizing automatic sampling of samples by underwater robots using target detection technology. Compared with the traditional manually controlled sampling method of underwater robots, the present invention has better stability and reduces the impact of environmental factors on underwater sampling.

[0004] Another object of the present invention is to provide a method and system for underwater robot sample identification and grasping. The method and system perform feature classification on the identified underwater sample types according to the target detection algorithm, and drive the corresponding type of sampling mechanism of the underwater robot to perform sampling of the corresponding target according to the results of the feature classification. Therefore, the present invention has a wide range of underwater sample sampling capabilities and can improve the comprehensive sampling capabilities of the underwater robot.

[0005] Another object of the present invention is to provide an underwater robot sample identification and grasping method and system, wherein the method and system construct a sample posture recognition method, wherein the posture recognition method pre-stores different posture type images of corresponding sample types, matches the identified underwater target samples with different posture images of corresponding sample types, obtains the posture type of the underwater target sample through matching analysis, and automatically generates a sampling strategy for the target sample based on the posture type of the underwater target sample, thereby improving the success rate of automated sampling of underwater samples.

[0006] Another object of the present invention is to provide an underwater robot sample identification and grasping method and system, which obtains the depth information of the identified target sample in the water in real time through a binocular camera, and drives the real-time adjustment of the movement and posture of the corresponding type of sampling mechanism in combination with the posture information of the sample, thereby improving the sampling success rate of the target sample underwater, and using the depth information obtained after the binocular camera image analysis to perform local sampling of the target sample, thereby improving the efficiency of underwater sampling.

[0007] In order to achieve at least one of the above-mentioned invention objects, the present invention further provides a method for identifying and grasping samples by an underwater robot, the method comprising:

[0008] The underwater robot uses a binocular camera to acquire collected image information in real time, and uses a target detection model to detect the number and type of target samples in the image information in real time;

[0009] Determining the underwater posture information of the target sample type identified according to the image information, and automatically generating an action instruction of a sampling mechanism corresponding to the target sample type according to the target sample type and the posture information;

[0010] Automatically detecting the depth information of the target sample of the corresponding type relative to the binocular camera underwater according to the binocular camera, and automatically adjusting the action posture of the corresponding sampling mechanism according to the depth information and posture information of the target sample;

[0011] Determine whether sampling is successful based on the corresponding type sensor of the sampling mechanism; if sampling fails, re-use the target detection model to track and identify the depth information and posture information of the target sample, and generate a new action posture of the corresponding sampling mechanism based on the re-identified depth information and posture information of the target sample;

[0012] If the sampling is successful, the depth information of all samples identified in the image information is obtained, and the sample with the smallest depth information is selected as a new target sample for sampling operation.

[0013] According to one of the preferred embodiments of the present invention, image information captured by the binocular camera is obtained, and the captured image information is input into a trained YOLO model for target detection, and the YOLO model is used to obtain a sample image area present in the image information, wherein the identified sample image area is rectangularly segmented according to the YOLO model to obtain a target sample block image; a KCF algorithm is used to extract a corresponding gradient histogram from each segmented target sample block image, a circulant matrix describing the area of ​​the sample image is constructed according to the corresponding gradient histogram, a sliding detection window of the circulant matrix is ​​constructed, and the similarity between the segmented target sample block image and the target sample is detected according to the sliding window and the trained classifier. If the similarity is greater than a preset threshold, it is tracked in real time as a target sample.

[0014] According to another preferred embodiment of the present invention, a driving instruction binding relationship corresponding to the underwater sample type and the sampling mechanism type is pre-constructed, wherein the sampling types include aquatic plants, aquatic animals, underwater rocks, and underwater mud, wherein target samples of different sampling types correspond to different sampling mechanisms, and after the sample type is identified by the YOLO model and used as the target sample, a driving instruction of the corresponding type of sampling mechanism is generated according to the driving instruction binding relationship to perform the sampling operation of the target sample.

[0015] According to another preferred embodiment of the present invention, the method for obtaining the posture information of the target sample includes: after obtaining the target sample block image in the image information according to the YOLO model identification, the target sample block image is subjected to edge detection using the Roberts operator to obtain the detected edge pixel features of the target sample in the target sample block image; the standard edge pixel features of different postures of the standard samples of the corresponding sample type are pre-detected using the Roberts operator, the similarity between the detected edge pixel features and the standard edge pixel features is calculated, and the standard edge pixel features of the standard sample posture with the highest similarity are selected as the posture of the target sample.

[0016] According to another preferred embodiment of the present invention, the method for generating motion instructions for the sampling mechanism includes: configuring at least one edge entry point in the standard edge pixel features of the standard sample posture of the corresponding sample type, and generating motion instructions including rotation, movement, and clamping for different control joints of the sampling mechanism based on the edge entry points of the matched standard sample posture of the corresponding sample type, so that the motion instructions will execute the sampling strategy of the corresponding target sample according to the edge entry points.

[0017] According to another preferred embodiment of the present invention, the method for obtaining depth information of the target sample includes: acquisition includes: using a binocular camera to respectively identify and obtain the target sample block image according to the YOLO model, and calculating the coordinate values ​​of the pixels of the target sample block image in the corresponding cameras, respectively calculating the average coordinate values ​​n1 and n2 of all pixels of the target sample block image in the two cameras, and calculating the disparity d=(n2-n1) / B according to the average coordinate values ​​n1 and n2 in the two cameras, where B represents the baseline distance between the binocular cameras, and further calculating the depth information Z of the target sample relative to the binocular camera based on the disparity d.

[0018] According to another preferred embodiment of the present invention, the method for calculating the depth information Z of the target sample includes: obtaining the grayscale value information of the target sample block image, and filtering out the grayscale features of the target sample based on the grayscale value information, performing feature matching based on the target sample grayscale features to obtain the average coordinate values ​​n1 and n2 of all pixels of the target sample in the two cameras, and calculating the disparity d, obtaining the focal length f of the binocular camera, and calculating the depth information Z=f×B / d of the target sample block image relative to the binocular camera according to the calculated disparity d according to the triangulation method.

[0019] According to another preferred embodiment of the present invention, after obtaining the depth information Z of the target sample block image and the posture information of the corresponding target sample, the robot body and / or the sampling mechanism are driven to approach the target sample according to the depth information Z of the block image. When the depth information reaches an appropriate range, a sampling strategy of the corresponding sampling mechanism is further generated according to the edge entry point in the posture information of the target sample.

[0020] In order to achieve at least one of the above-mentioned objects of the invention, the present invention further provides an underwater robot sample grasping and identifying system, which executes the above-mentioned underwater robot sample identification and grasping method.

[0021] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the above-mentioned underwater robot sample identification and grasping method. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Shown is a flow chart of a method for identifying and grasping samples by an underwater robot according to the present invention. DETAILED DESCRIPTION

[0023] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are for illustrative purposes only, and those skilled in the art will readily appreciate other obvious variations. The basic principles of the present invention defined in the following description may be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the present invention.

[0024] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the elements may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0025] Please combine Figure 1 The present invention discloses a method and system for underwater robot sample identification and grasping. The method primarily comprises the following steps: first, the underwater robot is equipped with binocular cameras, which capture image information underwater. The captured image information is then fed into a trained target detection model to output the sample category and quantity in the image information. The sample categories include, but are not limited to, aquatic plants, aquatic animals, underwater rocks, and mud. The target detection model is pre-trained, and in the present invention, the target detection model is preferably a YOLO model. The target sample image obtained by the target detection model is subjected to posture detection, and the underwater posture information of the sample identified by the target detection model is determined. The present invention also identifies the depth information of the target sample in water based on two images captured in the same time sequence by the binocular cameras, and automatically generates a sampling strategy for the target sample based on the target sample type, depth information, and posture information. The sampling strategy comprises a sampling mechanism comprising multiple rotating, telescopic, and mobile mechanisms, such as a conventional mechanical gripper. And because different samples are of different types, it is necessary to configure sampling mechanisms corresponding to the sample types. For example, the sampling mechanism for seabed shellfish can be a mechanical gripper, while the sampling mechanism for marine plankton can be a filter composed of an extension mechanism, etc. Since the sampling mechanisms corresponding to the sample types are all existing mechanical devices, the mechanical structure and form of the sampling mechanism are designed according to different requirements. The above-mentioned mechanical device itself is not the core technical means of the present invention, and the mechanical structure and form of the sampling mechanism are no longer provided with detailed technical details. After calculating the underwater depth and posture information of the target sample image identified by the target detection model, the present invention drives the corresponding type of sampling mechanism to perform automated sampling operations through the sampling strategy of the target sample. Therefore, the technical solution of the present invention can realize the automated sampling of target samples by robots underwater, thereby improving the sampling efficiency and success rate of the target samples.

[0026] Specifically, due to the different types of seabed samples, the difficulty of sampling will be different, especially the seabed environment is also affected by water flow, robot movement, and aquatic organisms. Therefore, the present invention requires real-time identification and tracking of the identified target samples, so that the sampling strategy can adapt to the seabed environment and sample properties and location. The present invention obtains the first image and the second image information in the same time sequence according to the binocular camera configured by the underwater robot, respectively, wherein the first image and the second image information are respectively input into the YOLO model for target sample detection for identification, and the sample type and quantity in the first image and the second image are respectively obtained, wherein the YOLO model is a multi-target detection model, after the corresponding type of target sample is identified by the YOLO model, the corresponding samples in the first image and the second image are rectangularly segmented according to the recognition mechanism of the YOLO model to obtain the corresponding target sample block image; the KCF algorithm is used to extract the gradient histogram of the corresponding target sample block image from each segmented target sample block image, wherein the method for constructing the gradient histogram includes: graying the segmented target sample block image, and automatically marking the target sample area pixels that meet the corresponding grayscale value range after graying, performing convolution calculation using gradient operators in different directions, and calculating the grayscale The target sample block image after the calibration has a gradient component and gradient amplitude in the corresponding direction of each pixel, and the target sample block image is divided into cell units and cell blocks according to a preset pixel size, and the gradient features and gradient amplitudes calculated in the cell units and cell blocks are calculated to obtain corresponding gradient histograms, which are used to describe the grayscale value gradient changes between different cell units and construct a circulant matrix of the annotated area; a classifier is pre-built to calculate the gradient histogram and circulant matrix in the target sample block image in the next frame, and the similarity is calculated based on the corresponding grayscale value in the target sample block image in the next frame obtained by the sliding window. The target sample image region gradient histogram is processed, wherein the circulant matrix is ​​used for Fourier transform calculation to reduce the matrix calculation amount of target tracking, and the sliding window coverage area with a similarity greater than a preset threshold is selected as the target sample tracking area annotated in the previous frame, thereby achieving efficient tracking of the target sample in different frame images. It should be noted that the target sample block image in the present invention is obtained by rectangular cutting of the YOLO model. Therefore, it contains a background image, and after being processed by the KCF algorithm, the influence of the background image on the regional tracking of the target sample image can be effectively reduced.

[0027] Furthermore, the present invention presets a binding relationship between the corresponding sampling mechanism drive instruction and the corresponding sample type, and performs posture feature comparison and recognition based on the target sample block image identified by the YOLO model. The corresponding sampling mechanism can be composed of at least one of a rotating mechanism, a moving mechanism, a telescopic mechanism, and a clamping mechanism, etc., including but not limited to a rotating mechanism, a moving mechanism, a telescopic mechanism, and a clamping mechanism, etc., and its drive instruction includes a rotation instruction, a telescopic instruction, a moving instruction, and a clamping instruction of the corresponding mechanism, etc. The rotation instruction includes a rotation angle and speed, the moving instruction includes a speed and distance of movement, the telescopic instruction includes a telescopic length and speed, and the clamping instruction includes a clamping force angle and strength, etc. The above-mentioned corresponding rotation instruction, telescopic instruction, moving instruction, and clamping instruction are bound by the corresponding type of sampling mechanism and the corresponding sample type. When the YOLO model recognizes the corresponding type of target sample, the corresponding type of sampling mechanism is started, and self-sampling is achieved through the corresponding type of drive instruction. For example, in one preferred embodiment of the present invention, the rotating mechanism may be a stepper motor. The microcontroller driving the stepper motor is configured with a rotation direction control pin, DIR_PIN, and automatically generates the following high-pulse instruction, digitalWrite(STEP_PIN, HIGH). digitalWrite() is a pulse function. The high-pulse instruction, digitalWrite(STEP_PIN, HIGH), generates a pulse to drive rotation via the rotation direction control pin, DIR_PIN. The delay instruction, delay_us(10mm), further delays the corresponding pulse step to control the rotation speed of the corresponding rotating mechanism. The drive instruction also includes a pulse instruction, digitalWrite(STEP_PIN, LOW), and a stepper motor speed, step_motor(), which respectively set the duration parameter of the delay instruction for generating the drive instruction. It should be noted that the drive instruction may be a single drive instruction as described above, or multiple combined drive instructions, used to control the automated sampling of the sampling mechanism. The present invention automatically drives the corresponding sampling mechanism to execute the corresponding sampling drive instruction based on the YOLO model to identify the position of the target sample block image of the corresponding type, the posture of the target sample underwater and the depth information of the target sample.

[0028] The method for acquiring the underwater posture data of the target sample comprises: after obtaining the target sample block image in the image information according to the YOLO model recognition, it is necessary to perform data preprocessing on the target sample block image, wherein the data preprocessing method comprises: using the Roberts operator to perform edge detection on the target sample block image and the standard sample image, obtaining the detection edge pixel features of the target sample in the target sample block image and the standard edge pixel features of the standard sample image in different postures; wherein the detection edge pixel features and the standard edge pixel features of the standard sample image in different postures are detected by the method The method comprises: constructing a horizontal convolution kernel and a vertical convolution kernel according to the Roberts operator, wherein the horizontal convolution kernel and the vertical convolution kernel can be preset as a 2*2 convolution kernel, performing grayscale conversion on the target sample block image and the standard target sample image, and then using the filter2D function to calculate the horizontal and vertical gradient values ​​of the target sample block image and the standard target sample image of different postures, respectively, and merging the horizontal and vertical gradient values ​​to output the corresponding detected edge pixel features and the standard edge pixel features of the standard sample image. In the present invention, a gradient amplitude is obtained according to the merged horizontal and vertical gradient values, and the gradient amplitude is binarized and a threshold is set. The pixel points with a gradient amplitude greater than the preset threshold are used as the pixel points of the corresponding posture edge feature.

[0029] Furthermore, after the present invention obtains the detection edge pixel features and the standard edge pixel features of the standard sample image based on the Roberts operator, the detection edge pixel features actually detected and the standard edge pixel features of the standard sample image with different postures are compared for similarity, and the highest standard edge pixel feature posture is selected as the detection edge pixel feature posture actually detected. An entry point is configured on the edge pixel of the detection edge pixel feature of the identified posture, wherein the setting of the entry point is set according to different sample types. For example, the entry point for fan-shaped shell samples can be set at the edge pixel point of the shell near the shell ear part. Because the shell ear part is suitable for the clamping device to insert the claws from the edge to clamp the shell sample, such an automated clamping method has higher stability. It should be noted that the configuration of the above entry point will be set according to the sample type, and the position of the entry point will be determined in combination with the posture information of the detection sample, and the driving strategy of the sampling mechanism will be automatically generated in combination with the depth information of the target sample relative to the binocular camera. The above entry point is only for example, and the present invention is not limited to this.

[0030] The method for obtaining the depth information of the target sample relative to the binocular camera includes the following steps: obtaining the baseline distance B of the binocular camera after the underwater robot is installed, wherein the baselines between the binocular cameras are installed in parallel, identifying and obtaining the target sample block images of the two cameras respectively according to the YOLO model, recording the first target sample block image P1 and the second target sample block image P2, and respectively calculating the coordinate values ​​of the pixels of the first target sample block image P1 and the second target sample block image P2 in the corresponding cameras, respectively calculating the average coordinate values ​​n1 and n2 of all pixels of the first target sample block image P1 and the second target sample block image P2 in the corresponding cameras, calculating the disparity d=(n2-n1) / B according to the average coordinate values ​​n1 and n2 of all pixels of the first target sample block image P1 and the second target sample block image P2 in the corresponding cameras, wherein B represents the baseline distance between the binocular cameras, and further calculating the depth information Z of the target sample relative to the binocular camera according to the disparity d.

[0031] In one preferred embodiment of the present invention, in order to reduce the influence of the background image in the rectangular cut image on the depth information in the target sample image recognized by the YOLO model, the present invention makes the following improvements: obtaining the grayscale value information of the target sample block image, and filtering out the target sample grayscale features based on the grayscale value information, wherein the screening method can include but is not limited to threshold screening and deep learning model feature analysis screening, etc., performing feature matching based on the target sample grayscale features to obtain the average coordinate values ​​n1 and n2 of all pixels of the target sample in the two cameras, and calculating the disparity d, obtaining the focal length f of the binocular camera, and calculating the depth information Z=f×B / d of the target sample block image relative to the binocular camera according to the calculated disparity d according to the triangulation method. In another preferred embodiment of the present invention, in order to better obtain the different local depth information of the target sample, the present invention can calculate the depth information of the local area of ​​the target sample according to the triangulation method for single pixels or multiple pixels in different areas of the target sample to form pixel blocks, thereby using a sampling strategy for local depth information.

[0032] The automated sampling strategy for the target sample of the present invention includes: after obtaining the depth information Z of the target sample block image and the posture information of the corresponding target sample, driving the robot body and / or sampling mechanism to approach the target sample according to the depth information Z of the block image, wherein the telescopic mechanism can be driven to approach the target sample at a preset speed, and when the depth information reaches the appropriate range, further generating an efficient and accurate sampling strategy for the corresponding sampling mechanism according to the edge entry point in the posture information of the target sample. The sampling strategy also includes searching for the nearest target sample and a sampling strategy. When the sampling is successful, the corresponding sampling sensor can generate a related signal, such as but not limited to a camera, a pressure sensor, a gravity sensor, etc. The sampling operation is performed by obtaining the depth information of all samples identified in the image information and selecting the sample with the smallest depth information as the new target sample.

[0033] The embodiments disclosed in the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. The embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the above-mentioned functions defined in the method of the present application are executed. It should be noted that the computer-readable medium mentioned above in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, a system, device or component of an electrical, magnetic, optical, electromagnetic, infrared segment, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more wire segments, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including, but not limited to, an electromagnetic signal, an optical signal, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, electrical wire, optical fiber cable, RF, etc., or any suitable combination thereof.

[0034] The flow charts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0035] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the implementation methods of the present invention may be subject to any deformation or modification.

Claims

1. A method for identifying and grabbing samples by an underwater robot, characterized in that: The method comprises: The underwater robot uses a binocular camera to acquire collected image information in real time, and uses a target detection model to detect the number and type of target samples in the image information in real time; Determining the underwater posture information of the target sample type identified according to the image information, and automatically generating an action instruction of a sampling mechanism corresponding to the target sample type according to the target sample type and the posture information; Automatically detecting the depth information of the target sample of the corresponding type relative to the binocular camera underwater according to the binocular camera, and automatically adjusting the action posture of the corresponding sampling mechanism according to the depth information and posture information of the target sample; Determine whether sampling is successful based on the corresponding type sensor of the sampling mechanism; if sampling fails, re-use the target detection model to track and identify the depth information and posture information of the target sample, and generate a new action posture of the corresponding sampling mechanism based on the re-identified depth information and posture information of the target sample; If the sampling is successful, the depth information of all samples identified in the image information is obtained, and the sample with the smallest depth information is selected as a new target sample for sampling operation; Obtain image information captured by the binocular camera, and input the captured image information into a trained YOLO model for target detection, using the YOLO model to obtain a sample image area present in the image information, wherein the identified sample image area is rectangularly segmented according to the YOLO model to obtain a target sample block image; extract the corresponding gradient histogram of each segmented target sample block image using a KCF algorithm, construct a circulant matrix describing the sample image area according to the corresponding gradient histogram, construct a sliding detection window of the circulant matrix, and detect the similarity between the segmented target sample block image and the target sample according to the sliding detection window and the trained classifier; if the similarity is greater than a preset threshold, it is tracked in real time as a target sample; Pre-establishing a driving instruction binding relationship corresponding to an underwater sampling type and a sampling mechanism type, wherein the sampling types include aquatic plants, aquatic animals, underwater rocks, and underwater sediments, wherein target samples of different sampling types correspond to different sampling mechanisms, and after the sample type is identified by the YOLO model and used as the target sample, generating a driving instruction for the corresponding type of sampling mechanism according to the driving instruction binding relationship to perform a sampling operation for the target sample; The method for acquiring the posture information of the target sample comprises: after obtaining the target sample block image in the image information according to the YOLO model identification, performing edge detection on the target sample block image using the Roberts operator to obtain the detected edge pixel features of the target sample in the target sample block image; using the Roberts operator to pre-detect the standard edge pixel features of different postures of the standard samples of the corresponding sample type, performing similarity calculation between the detected edge pixel features and the standard edge pixel features, and selecting the standard edge pixel features of the standard sample posture with the highest similarity as the posture of the target sample; The method for generating action instructions for the sampling mechanism includes: configuring at least one edge entry point in the standard edge pixel features of the standard sample posture of the corresponding sample type, and generating action instructions including rotation, movement, and clamping for different control joints of the sampling mechanism based on the edge entry points of the matched standard sample posture of the corresponding sample type, so that the action instructions will execute the sampling strategy of the corresponding target sample according to the edge entry points.

2. The underwater robot sample identification and grasping method according to claim 1, characterized in that: The method for acquiring depth information of the target sample includes: acquisition, including: using a binocular camera to respectively identify and acquire the target sample block image according to the YOLO model, and calculating the coordinate values ​​of the pixels of the target sample block image in the corresponding cameras, respectively calculating the average coordinate values ​​n1 and n2 of all pixels of the target sample block image in the two cameras, and calculating the disparity d=(n2-n1) / B according to the average coordinate values ​​n1 and n2 in the two cameras, where B represents the baseline distance between the binocular cameras, and further calculating the depth information Z of the target sample relative to the binocular camera based on the disparity d.

3. The underwater robot sample identification and grasping method according to claim 2, characterized in that: The method for calculating the depth information Z of the target sample includes: obtaining the grayscale value information of the target sample block image, and filtering out the grayscale features of the target sample based on the grayscale value information, performing feature matching based on the target sample grayscale features to obtain the average coordinate values ​​n1 and n2 of all pixels of the target sample in the two cameras, and calculating the disparity d, obtaining the focal length f of the binocular camera, and calculating the depth information Z=f×B / d of the target sample block image relative to the binocular camera according to the calculated disparity d using the triangulation method.

4. The underwater robot sample identification and grasping method according to claim 2, characterized in that: After obtaining the depth information Z of the target sample block image and the posture information of the corresponding target sample, the robot body and / or the sampling mechanism are driven to approach the target sample according to the depth information Z of the block image. When the depth information reaches an appropriate range, a sampling strategy of the corresponding sampling mechanism is further generated according to the edge entry point in the posture information of the target sample.

5. An underwater robot sample grabbing and identification system, characterized in that: The system executes the underwater robot sample identification and grasping method described in any one of claims 1-4.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the underwater robot sample identification and grasping method according to any one of claims 1 to 4.

Images