A method for underwater acoustic and optical joint target search

Abstract

The present invention discloses an underwater acoustic-optical combined target search method. The underwater robot used in the method includes a robot body and an external environment perception module. The external environment perception module includes a side-scan sonar, a forward-looking sonar, an underwater optical sensor and an underwater fill light. The forward-looking sonar, the optical sensor and the underwater fill light are all installed at the head end of the robot body, and are all arranged on the same side of the robot body that deviates from the bottom. Two side-scan sonars are respectively installed on opposite sides of the bottom of the robot body. The present invention uses a side-scan sonar to traverse and scan the underwater search area, and when a suspected target position appears, a spiral dive is performed, and acoustic and optical images in the direction of the suspected target are collected by the forward-looking sonar and the optical sensor, and acoustic-optical combined recognition is performed to determine whether the target actually exists, thereby significantly improving the accuracy of underwater target recognition.

Description

Technical Field

[0001] The invention belongs to the field of underwater search, and in particular relates to an underwater acoustic-optical combined target search method. Background Art

[0002] With the rapid development of modern science and technology, human beings are increasingly dependent on marine resources. However, due to the complex and changeable marine environment and low underwater visibility, ocean exploration has become extremely challenging. Traditional underwater searches rely on acoustic detection, which has low accuracy and requires manual verification by divers or robots to determine the authenticity of the target. How to balance a large search range and accurate close verification is a technical difficulty.

[0003] In underwater target search operations, the target object is usually detected by controlling an underwater vehicle (Remotely Operated Vehicle, ROV). This not only relies on the unique experience of experts, but also is very cumbersome and labor-intensive when facing a large-scale search. With the development of control technology and image perception technology, highly autonomous underwater vehicles (Autonomous Underwater Vehicles, AUV) have gradually become a flexible and effective tool in the field of ocean detection. Autonomous underwater robots have the characteristics of good concealment, flexibility of underwater movement, and economic applicability. They have received widespread attention in various fields. Autonomous search of underwater targets by AUV can greatly extend the operation time limit of the search task and save labor costs. The acoustic detection and optical detection of the existing underwater target search methods are independent. The acoustic detection method has a long detection distance but low accuracy; compared with optical detection, it has high accuracy, but is easily affected by water quality and has a short detection distance; therefore, effectively combining their respective advantages to jointly search for targets will improve the reliability of underwater search. In view of the above problems, the present invention proposes a target search planning method combining underwater acoustic and optical information. Summary of the invention

[0004] The purpose of the present invention is to provide an underwater acoustic and optical joint target search method for specific underwater targets, using an underwater autonomous robot to operate autonomously underwater, using autonomous coverage path planning and combining motion control to search a designated area, and combining an acoustic wide-area recognition method and an optical recognition method to achieve the purpose of underwater autonomous target search. The present invention aims to provide a complete set of underwater search processes and methods that meet actual application needs and have a high degree of intelligence in target search tasks in actual underwater environments.

[0005] An underwater acoustic-optical combined target search method, the hardware system includes a robot body and an external environment perception module. The external environment perception module includes a side-scan sonar, a forward-looking sonar, an underwater optical sensor, and an underwater fill light. The forward-looking sonar, the optical sensor, and the underwater fill light are all installed at the head end of the robot body, and are all arranged on the same side of the robot body that deviates from the bottom. The side-scan sonars are respectively installed on the opposite sides of the bottom of the robot body.

[0006] The underwater acoustic-optical combined target search method comprises the following steps:

[0007] Step 1: Set the scope of the search area, and the underwater robot moves into the search area.

[0008] Step 2: Wide-area exploration phase.

[0009] The underwater robot moves in the search area and conducts a traversal search. During the movement of the underwater robot, the side-scan sonar reads the sonar data. The obtained data is input into the corresponding recognition network for detection. When the recognition network detects a target with a confidence level exceeding a threshold, it determines that a suspicious target has been identified and enters the verification phase.

[0010] Step 3: Verification stage.

[0011] 3-1. After the side-scan sonar finds a suspicious target, the position coordinates of the suspicious target in the geodetic coordinate system are calculated using the suspicious target position solution method; the vertical axis where the position coordinates of the suspicious target are located is used as the central axis to plan the spiral dive path.

[0012] 3-2. The robot body dives along the spiral diving path and approaches the position coordinates of the suspicious target; during the robot body diving, the forward-looking sonar and underwater optical sensor are located on the side of the robot body close to the central axis of the spiral diving path. The forward-looking sonar and underwater optical sensor collect acoustic images and optical images respectively, and input them into the corresponding recognition network to detect whether there is a target, as well as the confidence and position of the target.

[0013] 3-3. When the target is found in both the acoustic image and the optical image, the acoustic and optical fusion is performed to finally determine whether the target is found. The specific process is as follows:

[0014] The confidence c of the optical image after the corresponding recognition network target c , position coordinates p c , area c , the confidence c of the acoustic image after the corresponding recognition network target s , position coordinates p s , area s .

[0015] Constructing joint target confidence D c, location similarity D p Similarity to the object size D s The expression is as follows:

[0016]

[0017] Among them, λ p and λ s are the coefficients of the preset position similarity and object size similarity, and satisfy 0<λ p <1, 0<λ s <1. dis(p c ,p s ) is the position coordinate p c and p s The Euclidean distance between .

[0018] According to the joint target confidence D c , location similarity D p Similarity to the object size D s The sound-light fusion confidence U is obtained as follows:

[0019] U=C1D c +C2D p +C3D s

[0020] Among them, C1, C2 and C3 are feature weight coefficients.

[0021] When the acoustic-optical fusion confidence U is greater than the recognition threshold L, the target is considered to be recognized. The underwater robot floats to the surface and sends the target position information to the host computer. After that, the robot body dives to the preset depth and returns to the wide-area exploration stage.

[0022] If the robot dives to the depth corresponding to the suspected target location but fails to find the target, it is deemed that the side-scan sonar has made an identification error. The robot body then floats up to the preset depth and returns to the wide-area exploration stage.

[0023] Preferably, the robot body is an under-actuated underwater autonomous robot capable of adjusting propulsion force, pitch angle and azimuth angle underwater.

[0024] Preferably, the orientations of the forward-looking sonar and the optical sensor are both at an angle of 45° to the vertical axis. The orientations of the forward-looking sonar and the optical sensor are both at an angle of 45° to the central axis of the robot body.

[0025] Preferably, in step 2, the underwater robot traverses and searches the search area by moving along a serpentine path at a preset depth within the search area.

[0026] Preferably, in step 2, the recognition network corresponding to the side scan sonar performs real-time target detection based on a machine learning detection method; the specific process is: the waterfall data detected by the side scan sonar is generated into an image using a sliding window method, and after being sent to the side scan sonar recognition network, the confidence and position of the suspicious target are obtained.

[0027] Preferably, the underwater robot is positioned by an inertial navigation system when it is below the water surface. When the condition S×p≥R is met, the underwater robot will return to the surface and calibrate the inertial navigation system by the satellite positioning module. Where S is the distance traveled by the robot body after the last calibration, p is the accuracy error of the inertial navigation system, and R is the preset maximum allowable accuracy error.

[0028] Preferably, the curve of the spiral diving path described in step 3-1 is as follows:

[0029]

[0030] Where r is the radius of the preset cylindrical spiral path; x T ,y T z is the position coordinate of the suspicious target projected on the horizontal plane; P is the current depth; t is time, ω is the preset rotation speed, and k is the preset descent speed; it should be noted that the spiral path depth z d (t) shall not be greater than the target depth z T ;x d (t), y d (t), z d (t) is the coordinate of a point on the spiral diving path.

[0031] Preferably, in step 3-2, the recognition networks of both the acoustic image and the optical image use the YOLO target detection algorithm.

[0032] Preferably, in step 3-3, for the target area s of the optical image c The detection method is as follows:

[0033] The target recognition method is used to obtain the center position (x, y) of the target's bounding box in the image and the size of the box (w, h). The width of the target is T W , length T H , area c It is calculated by:

[0034]

[0035]

[0036] s c =T W ×TH

[0037] Target area s for acoustic image s The detection method is as follows:

[0038] Extract the image within the bounding box of the acoustic recognition image, extract the outline of the core target through the level set algorithm, and select the two boundary points with the largest vertical coordinate span and the two boundary points with the largest horizontal coordinate span. Corresponding to the original acoustic data, we can determine the distance from the sonar to the four points (w1, w2, h1, h2) and the angle χ between the two boundary points with the largest horizontal coordinate span and the angle γ between the two boundary points with the largest vertical coordinate span. Area s c The expression is as follows:

[0039]

[0040] Preferably, the robot body is provided with a communication module, an underwater positioning module and a satellite positioning module. The underwater positioning module includes an inertial navigation system, a Doppler sonic velocity meter and a pressure gauge. The communication module includes a Beidou communication system, a 433MHz wireless communication system and a WiFi communication system.

[0041] The beneficial effects of the present invention are:

[0042] 1. The present invention adopts a two-stage combined search method to search for targets in a specified area. It combines the acoustic long-range detection capability and the optical high reliability to identify the target, giving the underwater robot a highly autonomous target search capability.

[0043] 2. The present invention uses a side-scan sonar with long-range detection capability to perform a full-coverage scan of the underwater search area, identifies suspected targets based on a machine learning method, calculates the position of the suspected targets, and performs a coverage scan of the suspected targets by generating a spiral diving path.

[0044] 3. The present invention collects acoustic images and optical images in the direction of suspicious targets through forward-looking sonar and optical sensors, and performs joint acoustic and optical recognition to determine whether the target actually exists, thereby significantly improving the accuracy of underwater target recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the structure of the underwater robot used in the present invention.

[0046] Figure 2 It is a target search flow chart of the present invention.

[0047] Figure 3 It is a schematic diagram of the search coverage path of the underwater robot in the wide area exploration stage of the present invention.

[0048] Figure 4It is a schematic diagram of the path planning of the underwater robot for target search in the present invention.

[0049] Figure 5 It is a schematic diagram of optical image target recognition in the present invention. DETAILED DESCRIPTION

[0050] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0051] like Figure 1 As shown, an underwater acoustic-optical combined target search method uses an underwater robot comprising: a robot body 1 and an external environment perception module. The robot body 1 in this embodiment is an underactuated underwater autonomous robot capable of adjusting propulsion force, pitch angle and azimuth angle underwater.

[0052] The external environment perception module includes a side-scan sonar 2, a forward-looking sonar 3, an underwater optical sensor 4, and an underwater fill light 5. The forward-looking sonar 3, the optical sensor, and the underwater fill light are all installed at the head end of the robot body 1, and are all arranged on the same side of the robot body 1 that deviates from the bottom. The orientations of the forward-looking sonar 3 and the optical sensor are both at an angle of 45° to the vertical axis. The orientations of the forward-looking sonar 3 and the optical sensor are both at an angle of 45° to the central axis of the robot body 1. The side-scan sonars 2 are respectively installed on the opposite sides of the center of the bottom of the robot body 1, and are respectively used to perform sonar scanning on both sides of the robot body 1.

[0053] The robot body 1 is provided with a communication module, an underwater positioning module and a satellite positioning module. The underwater positioning module includes an inertial navigation system, a Doppler sonic velocity meter and a pressure gauge. The underwater positioning module is used to obtain accurate underwater position information of the robot body 1 to ensure navigation safety. The satellite positioning module is used to obtain high-precision global position coordinates of the robot body 1 by radio when the robot body 1 is on the water surface.

[0054] The communication module includes the Beidou communication system, the 433MHz wireless communication system and the WiFi communication system. The communication module is used to communicate with users under different communication restriction conditions.

[0055] like Figure 2 As shown, the underwater acoustic-optical combined target search method comprises the following steps:

[0056] Step 1: Wide-area search path generation phase.

[0057] Set the scope of the autonomous underwater search area. Input the search area boundary point P1 (longitude 1, latitude 1), boundary point P2 (longitude 2, latitude 2), boundary point Pn-1 (longitude n-1, latitude n-1), boundary point Pn (longitude n, latitude n), boundary point P1 (longitude 1, latitude 1) through the wireless communication system; the closed area formed by the boundary point sequence is the search area. Automatically plan the coverage path and autonomously drive to the starting point of the path; enter the wide-area exploration phase after arriving at the starting point.

[0058] Step 2: Wide-area exploration phase.

[0059] like Figure 3 and 4 As shown, the robot body 1 moves along a serpentine path at a preset depth in the search area, and relies on the side scan sonar 2 to conduct a preliminary traversal search of the search area. The depth determination strategy of the robot body 1 can reduce the interference of the complex environment of the water surface on the search process; the sonar data is read by the side scan sonar 2, and the target is detected in real time based on the detection method of machine learning; first, the waterfall data detected by the side scan sonar 2 is generated into an image using the sliding window method, and after being sent to the side scan sonar 2 recognition network, the confidence of the target is obtained through reasoning. When the confidence exceeds the recognition model threshold, it is considered that a suspicious target is identified and enters the verification stage.

[0060] During the movement of the robot body 1 in the search area, when the following conditions are met, the robot body 1 will return to the water surface, be positioned through the satellite positioning module, and correct the positioning accuracy of the inertial navigation system in the underwater positioning module.

[0061] S×p≥R

[0062] Wherein, S is the distance traveled by the robot body 1 after the last calibration, p is the accuracy error of the inertial navigation system, and R is the preset maximum allowable accuracy error.

[0063] Step 3: Verification stage.

[0064] 3-1. Since the underwater positioning system mainly uses underwater acoustic positioning equipment, which needs to be deployed in advance, it greatly increases the complexity of the operation. Therefore, a high-precision inertial navigation system is mainly used for underwater navigation. In order to ensure the positioning accuracy of the AUV, the positioning accuracy correction method is used to improve the underwater positioning accuracy. At the same time, when the entire area is explored, it will return to the starting position autonomously to end the mission.

[0065] After the side scan sonar 2 finds a suspicious target, the position coordinates of the suspicious target in the geodetic coordinate system are calculated by the suspicious target position solution method; the vertical axis where the position coordinates of the suspicious target are located is used as the central axis, and the spiral dive path is planned as follows:

[0066]

[0067] Where r is the radius of the preset cylindrical spiral path; x T ,y T z is the position coordinate of the suspicious target projected on the horizontal plane; P is the current depth; t is the time, ω is the preset rotation speed, and k is the preset descent speed. It should be noted that the spiral path depth z d (t) shall not be greater than the target depth z T .x d (t), y d (t), z d (t) is the coordinate of a point on the spiral diving path.

[0068] If the target is still not found after reaching the predetermined depth, it will return autonomously.

[0069] 3-2. The robot body 1 dives along the spiral diving path and approaches the position coordinates of the suspicious target; during the diving of the robot body 1, the forward-looking sonar 3 and the underwater optical sensor are located on the side of the robot body 1 close to the central axis of the spiral diving path. The forward-looking sonar 3 / underwater optical sensor and fill light located on the side of the robot body 1 are activated. The suspicious targets are further screened by the forward-looking sonar 3 and the underwater optical sensor.

[0070] The acoustic image collected by the forward-looking sonar 3 and the optical image collected by the underwater optical sensor are respectively input into the corresponding recognition network to detect whether there is a target, as well as the confidence and position of the target. When the confidence is greater than or equal to the preset recognition model threshold, it is considered that the target is found. In this embodiment, the recognition networks of the acoustic image and the optical image both use the YOLO target detection algorithm.

[0071] In this embodiment, the target position detection method for acoustic images is specifically as follows:

[0072] The relative position information of the suspicious target can be extracted from the sonar data, including the distance d from the target, the horizontal angle α and the vertical angle β. The horizontal angle is related to the installation method of the side scan sonar 2. When located on both sides of the robot body 1, the horizontal angles of the sonars on both sides are 90° and -90° respectively. Combined with the posture information of the robot body 1, the target position is obtained by the following method.

[0073]

[0074] Among them, ψ and θ are the azimuth and pitch angles of the robot body 1 respectively; (x d ,y d ,z d ) is the position coordinate of the robot body 1; (x T ,y T,z T ) is to obtain the target position coordinates based on the acoustic image.

[0075] 3-3. When the target is found in both the acoustic image and the optical image, the acoustic and optical fusion is performed to finally determine whether the target is found. The specific process is as follows:

[0076] The optical image and acoustic image are respectively passed through the corresponding recognition network to obtain the confidence c of the target c and c s At the same time, the position of the target p is calculated according to the monocular camera ranging method c The relative angle and distance of the target can be directly obtained from the acoustic image. The estimated position p of the target in the acoustic image can be obtained by the above acoustic target position calculation method. s ; For the optical image and acoustic image, the corresponding target area estimation method is used to calculate the area s of the observed target c and s .

[0077] In this embodiment, the target area s of the optical image c The detection method is as follows:

[0078] The target recognition method is used to obtain the target's bounding box center position (x, y) and box size (w, h). Figure 5 Take the calculation of the width of the real target as an example, w is the length of the box in the image, x is the distance from the center of the bounding box to the central axis, the distance between the camera and the center of the target is d, and the focal length of the optical sensor is fixed l. The real width of the target T W It can be calculated in the following way.

[0079]

[0080] The same method can be used to obtain the actual length of the target The target area is s c =T W ×T H .

[0081] In this embodiment, the target area s of the acoustic image s The detection method is as follows:

[0082] Extract the image within the bounding box of the acoustic recognition image, extract the outline of the core target through the level set algorithm, and select the two boundary points with the largest vertical coordinate span and the two boundary points with the largest horizontal coordinate span. Corresponding to the original acoustic data, we can determine the distance from the sonar to the four points (w1, w2, h1, h2) and the angle χ between the two boundary points with the largest horizontal coordinate span and the angle γ between the two boundary points with the largest vertical coordinate span. The area calculation method is as follows:

[0083]

[0084] Acoustic-optical fusion mainly integrates three kinds of feature data, namely, joint target confidence D c , location similarity D p Similarity to the object size D s , the calculation method is as follows:

[0085]

[0086] Among them, λ p and λ s are the coefficients of the preset position similarity and object size similarity, and satisfy 0<λ p <1, 0<λ s <1. dis(p c ,p s ) is the optical target position p c and acoustic target position p s Finally, the acoustic and optical target recognition results are fused by the following formula to obtain the acoustic and optical fusion confidence U as follows:

[0087] U=C1D c +C2D p +C3D s

[0088] Among them, C1, C2 and C3 are feature weight coefficients, and U represents the target recognition accuracy value.

[0089] The greater the confidence U of the acoustic-optical fusion, the greater the probability that the suspicious target in the image is the real target. Therefore, by setting the final recognition threshold L, when U>L, it is considered that the target has been found.

[0090] If the robot body 1 reaches the position of the suspected target determined by the side scan sonar 2 but fails to find the target, it is determined that the side scan sonar 2 has made an identification error, and the robot body 1 floats to a preset depth and then returns to the wide-area exploration stage.

[0091] If the target is found, the robot body 1 records the position and immediately floats to the surface, and sends the target position information to the upper computer through the wireless communication system. Then, the robot body 1 dives to the preset depth and returns to the wide-area exploration stage.

[0092] In this embodiment, the method for acquiring a data set for training the side scan sonar 2 target recognition network is as follows: the robot body 1 is used to scan specific underwater targets in different directions in a fixed depth mode to obtain waterfall graph data, and a sliding window is used to divide the complete waterfall graph into a limited number of images as a data set.

[0093] The data set acquisition method for the recognition network corresponding to the acoustic image and the recognition network corresponding to the optical image collected by the forward-looking sonar 3 is as follows:

[0094] In underwater environments, optical sensors and front-facing sonar are used to capture images of specific targets in a circular trajectory. Multi-perspective data sets can significantly improve the accuracy of the model in identifying targets.

[0095] The coverage search path of the robot body 1 is not bounded by the boundary line of the area, and aims to search the designated area with the best scanning posture without missing any blind spots. Figure 3 The figure shows a schematic diagram of the planning of the coverage path in the closed area composed of P1-P6; considering that when the robot body 1 turns, the image of the side scan sonar 2 will be distorted and effective observation data cannot be obtained, the turning is not performed in the search area.

[0096] In this embodiment, the path tracking controller method of the robot body 1 is as follows:

[0097] The dynamic equation of the underwater autonomous robot is as follows:

[0098]

[0099] Where M is the inertia matrix; C is the centripetal force and Coriolis force matrix, D is the hydrodynamic drag and lift moment, g(η) is the restoring force and moment; v is the velocity matrix; η is the position and attitude matrix; τ is the input torque.

[0100] First, construct the tracking error in the geodetic coordinate system:

[0101]

[0102] Where x, y and z are the position coordinates of the robot body 1, x d ,y d and z d is the coordinate of the preset arrival position. The calculation method of converting the error in the earth coordinate system to the body coordinate system is:

[0103]

[0104] Where ψ and θ are the azimuth and pitch angles of the robot body 1, respectively. Taking the PID control method as an example, the tracking control law of the underwater autonomous robot is as follows:

[0105]

[0106] where τ u , τ r and τ p P is the forward moment, steering moment and pitch moment of the robot body 1; u , Pr and P p is the preset proportional coefficient; I u ,I r and I p is the preset integral coefficient; D u , D r and D p is the preset differential coefficient.

[0107] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An underwater acoustic-optical combined target search method, characterized in that: The underwater robot used includes a robot body and an external environment perception module; the external environment perception module includes a side-scan sonar, a forward-looking sonar, an underwater optical sensor, and an underwater fill light; the forward-looking sonar, the optical sensor, and the underwater fill light are all installed at the head end of the robot body, and are all arranged on the same side of the robot body deviating from the bottom; two side-scan sonars are respectively installed on the opposite sides of the bottom of the robot body; The underwater acoustic-optical combined target search method comprises the following steps: Step 1: Set the scope of the search area, and move the underwater robot into the search area; Step 2: Wide-area exploration phase; The underwater robot moves in the search area and conducts traversal search. During the movement of the underwater robot, two side-scan sonars read the sonar data. The obtained data is input into the corresponding recognition network for detection. When the recognition network detects a target with a confidence level exceeding the threshold, it determines that a suspicious target has been identified and enters the verification phase. Step 3: Verification stage; 3-1. After the side scan sonar finds a suspicious target, the position coordinates of the suspicious target in the geodetic coordinate system are calculated by the suspicious target position solution method; the vertical axis where the position coordinates of the suspicious target are located is used as the central axis to plan the spiral dive path; 3-2. The robot body dives along the spiral diving path and approaches the position coordinates of the suspicious target; during the diving of the robot body, the forward-looking sonar and the underwater optical sensor are located on one side of the robot body close to the central axis of the spiral diving path; the forward-looking sonar and the underwater optical sensor collect acoustic images and optical images respectively, and input them into the corresponding recognition network to detect whether there is a target, as well as the confidence and position of the target; 3-3. When the target is found in both the acoustic image and the optical image, the acoustic and optical fusion is performed to finally determine whether the target is found. The specific process is as follows: The confidence c of the optical image after the corresponding recognition network target c , position coordinates p c , area c , the confidence c of the acoustic image after the corresponding recognition network target s , position coordinates p s , area s ; Constructing joint target confidence D c , location similarity D p Similarity to the object size D s The expression is as follows: Among them, λ p and λ s are the coefficients of the preset position similarity and object size similarity, and satisfy 0<λ p <1, 0<λ s <1; dis(p c ,p s ) is the position coordinate p c and p s The Euclidean distance between According to the joint target confidence D c , location similarity D p Similarity to the object size D s The sound-light fusion confidence U is obtained as follows: U=C1D c +C2D p +C3D s Among them, C1, C2 and C3 are feature weight coefficients; When the acoustic-optical fusion confidence U is greater than the recognition threshold L, the target is considered to be recognized; the underwater robot floats to the surface and sends the target position information to the upper computer; then, the robot body dives to the preset depth and returns to the wide-area exploration stage; If the robot dives to the depth corresponding to the suspected target location but fails to find the target, it is deemed that the side-scan sonar has made an identification error. The robot body then floats up to the preset depth and returns to the wide-area exploration stage.

2. The underwater acoustic-optical combined target search method according to claim 1, characterized in that: The robot body is an under-actuated underwater autonomous robot, which can adjust the propulsion force, pitch angle and azimuth angle underwater.

3. The underwater acoustic-optical combined target search method according to claim 1, characterized in that: The orientations of the forward-looking sonar and the optical sensor are both at an angle of 45° to the vertical axis; the orientations of the forward-looking sonar and the optical sensor are both at an angle of 45° to the central axis of the robot body.

4. The underwater acoustic-optical combined target search method according to claim 1, characterized in that: In step 2, the underwater robot traverses the search area by moving along a serpentine path at a preset depth within the search area.

5. The underwater acoustic-optical combined target search method according to claim 1, characterized in that: In step 2, the recognition network corresponding to the side-scan sonar performs real-time target detection based on a machine learning detection method. The specific process is as follows: the waterfall data detected by the side-scan sonar is used to generate an image using a sliding window method, and after being sent to the side-scan sonar recognition network, the confidence and position of the suspicious target are obtained.

6. The underwater acoustic-optical combined target search method according to claim 1, characterized in that: The underwater robot uses an inertial navigation system to locate itself underwater. When the condition S×p≥R is met, the underwater robot will return to the surface and calibrate the inertial navigation system through the satellite positioning module; where S is the distance traveled by the robot body after the last calibration, p is the accuracy error of the inertial navigation system, and R is the preset maximum allowable accuracy error.

7. The underwater acoustic-optical combined target search method according to claim 1, characterized in that: The curve of the spiral dive path described in step 3-1 is as follows: Where r is the radius of the preset cylindrical spiral path; x T ,y T z is the position coordinate of the suspicious target projected on the horizontal plane; P is the current depth; t is time, ω is the preset rotation speed, and k is the preset descent speed; it should be noted that the spiral path depth z d (t) shall not be greater than the target depth z T ;x d (t), y d (t), z d (t) is the coordinate of a point on the spiral diving path.

8. The underwater acoustic-optical combined target search method according to claim 1, characterized in that: In step 3-2, the recognition networks of acoustic images and optical images both use the YOLO target detection algorithm.

9. The underwater acoustic-optical combined target search method according to claim 1, characterized in that: In step 3-3, for the target area s of the optical image c The detection method is as follows: The target recognition method is used to obtain the center position (x, y) of the target's bounding box in the image and the size of the box (w, h); the width T of the target W , length T H , area c It is calculated by: s c =T W ×T H Target area s for acoustic image s The detection method is as follows: Extract the image within the bounding box of the acoustic recognition image, extract the contour of the core target through the level set algorithm, and select the two boundary points with the largest vertical coordinate span and the two boundary points with the largest horizontal coordinate span; corresponding to the original acoustic data, the distance from the sonar to the four points (w1, w2, h1, h2) and the angle χ between the two boundary points with the largest horizontal coordinate span and the angle γ between the two boundary points with the largest vertical coordinate span can be determined; area s c The expression is as follows:

10. The underwater acoustic-optical combined target search method according to claim 1, characterized in that: The robot body is provided with a communication module, an underwater positioning module and a satellite positioning module; the underwater positioning module includes an inertial navigation system, a Doppler sonic velocity meter and a pressure gauge; the communication module includes a Beidou communication system, a 433MHz wireless communication system and a WiFi communication system.

Images