Unmanned salvage ship capable of recognizing images of tiny foreign matters on water surface and image capturing and processing method of unmanned salvage ship

Abstract

The embodiment of the invention provides a salvage unmanned ship for water surface tiny foreign matter image recognition and an image capturing and processing method thereof, and belongs to the technical field of water surface cleaning and water quality monitoring. The salvage unmanned ship comprises a ship body; the image acquisition and processing module is used for capturing an image of the water surface and identifying tiny foreign matters through a target detection algorithm; the garbage collecting device is mounted on the ship body and can collect tiny foreign matters on the water surface according to an identification result so as to clean garbage floating on the water surface; wherein the target detection algorithm is a Cascade R-CNN target detection algorithm, and the Cascade R-CNN target detection algorithm is combined with a feature pyramid network to process a multi-scale target through feature extraction from bottom to top and feature fusion from top to bottom. The Cascade R-CNN target detection algorithm is combined with the feature pyramid network to effectively solve the problems of low water surface garbage cleaning efficiency, difficulty in small foreign matter detection and the like in the prior art.

Description

Technical Field

[0001] The invention belongs to the technical field of water surface cleaning and water quality monitoring, and relates to an unmanned salvage boat for identifying images of tiny foreign objects on the water surface and an image capture and processing method thereof. Background Art

[0002] With the acceleration of urbanization and the increase of human activities, the problem of garbage pollution on the water surface is becoming increasingly serious. These floating objects not only affect the beauty of the water area, but also pose a serious threat to water quality and the ecological environment. Traditional manual cleaning methods are not only inefficient, but also difficult to cover a wide area of ​​water, especially the cleaning of tiny foreign objects is even more difficult.

[0003] At the same time, with the increasing awareness of environmental protection, the demand for water quality monitoring is growing, and traditional water quality testing methods can no longer meet modern environmental protection needs.

[0004] In view of this, there is a real need to provide a salvage unmanned boat and its image capture and processing method that can automatically clean up garbage on the water surface, especially tiny foreign objects, and monitor water quality in real time. Summary of the invention

[0005] The purpose of the present invention is to at least partially solve the above-mentioned technical problems, and to provide a salvage unmanned boat for surface micro-foreign object image recognition and an image capture and processing method thereof, so as to effectively solve the problems existing in the prior art such as low efficiency of surface garbage cleaning and difficulty in detecting micro-foreign objects.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect of the present invention, there is provided an unmanned salvage vessel for image recognition of tiny foreign objects on a water surface, the unmanned salvage vessel comprising:

[0008] hull;

[0009] Image acquisition and processing module, used to capture images of the water surface and identify tiny foreign objects through target detection algorithms;

[0010] A garbage collection device, installed on the hull and capable of collecting tiny foreign objects on the water surface according to the identification result to clean up the garbage floating on the water surface;

[0011] Among them, the target detection algorithm is the Cascade R-CNN target detection algorithm, which combines the feature pyramid network to process multi-scale targets through bottom-up feature extraction and top-down feature fusion.

[0012] In some embodiments of the present invention, the Cascade R-CNN target detection algorithm identifies and locates tiny foreign objects on the water surface through a step-by-step optimization mechanism of multi-level classifiers and regressors.

[0013] In some embodiments of the present invention, in each level of classifier of Cascade R-CNN, a cross entropy loss function is used, and the formula of the cross entropy loss function is:

[0014]

[0015] where p i ,u i is the predicted category probability of the i-th candidate region, N cls is the number of candidate regions.

[0016] In some embodiments of the present invention, a self-attention mechanism and a bounding box regression loss function are combined to perform bounding box regression. In the bounding box regression process, a smooth L1 loss function is used to minimize the gap between the predicted box and the true box. The bounding box regression loss function is:

[0017]

[0018] The smooth L1 loss function is defined as:

[0019]

[0020] Among them, t ij and v ij is the coordinates of the predicted box and the real box in the x, y, w, h dimensions, N reg is the number of candidate regions.

[0021] In some embodiments of the invention, when balancing the errors in classification and regression, the total loss function of Cascade R-CNN is defined as:

[0022] L=L cls +λL reg

[0023] Among them, λ is the weight coefficient, which is used to adjust the importance of classification loss and regression loss.

[0024] In some embodiments of the present invention, the garbage collection device is installed on both sides of the hull and is equipped with expandable PVC bayonet.

[0025] In some embodiments of the present invention, the unmanned salvage vessel further includes: a water quality detection module for monitoring water quality parameters in real time.

[0026] In some embodiments of the present invention, the unmanned salvage vessel further includes: a control module for coordinating navigation, garbage collection, and water quality detection of the vessel.

[0027] In some embodiments of the present invention, the unmanned salvage vessel further comprises: a power navigation module, which adopts a multi-degree-of-freedom propulsion device and uses GPS and UWB dual positioning technology;

[0028] The solar charging device adopts a mother-and-daughter ship design, is installed at the stern of the hull, and is connected to the battery.

[0029] In a second aspect of the present invention, there is provided an image capture and processing method for the above-mentioned unmanned salvage ship for surface micro-foreign object image recognition, comprising the following steps:

[0030] 1) Preprocessing step, the preprocessing step includes: image size adjustment and color space transformation;

[0031] 2) Target recognition step: using the pre-processed image to identify tiny foreign objects on the water surface using a target monitoring algorithm;

[0032] 3) Marking and classification: garbage and obstacles identified by the algorithm are accurately marked and classified.

[0033] The unmanned salvage boat for surface micro-foreign object image recognition and the image capture and processing method thereof according to the embodiment of the present invention have at least one of the following advantages:

[0034] 1. Efficient garbage collection capability: It can effectively identify and clean floating garbage on the water surface and improve the environmental quality of the water area.

[0035] 2. Real-time water quality monitoring function: Integrate multiple sensors to monitor water quality parameters in real time, providing a scientific basis for water quality management.

[0036] 3. Flexible working mode: Combination of manual and autonomous modes to adapt to different operating requirements and environmental conditions.

[0037] 4. Stable electronic control system: The waterproof design ensures the reliability of the electronic control part in bad weather and enhances the adaptability of the hull.

[0038] 5. Improved detection algorithm: The improved Cascade R-CNN target detection algorithm, combined with a deeper feature pyramid network (FPN) and self-attention mechanism, can accurately identify tiny floating garbage on the water surface and improve detection accuracy.

[0039] 6. Enhanced navigation capability: Improve the decision-making ability of unmanned ships in autonomous navigation and obstacle avoidance by improving the high-quality feature support provided by the algorithm, and achieve efficient path planning;

[0040] 7. The self-attention mechanism further enhances the detection algorithm's regional focusing ability, allowing it to focus more on key target areas, such as floating garbage and surface obstacles. This mechanism dynamically adjusts feature weights to ensure that the model can accurately locate targets when dealing with complex water surface environments. Combined with Cascade R-CNN's hierarchical classifiers and regressors, the detection algorithm can refine the boundaries and categories of targets layer by layer, thereby improving the accuracy of target positioning and the robustness of detection. In the bounding box regression process, a smooth L1 loss function is used to minimize the gap between the predicted box and the true box. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] These and / or other aspects and advantages of the present invention will become apparent and readily understood from the following description of the preferred embodiments in conjunction with the accompanying drawings, in which:

[0042] Figure 1 It is a schematic diagram of the three-dimensional structure of an unmanned salvage boat for image recognition of tiny foreign objects on the water surface according to an embodiment of the present invention;

[0043] Figure 2 yes Figure 1 The structural schematic diagram of the water quality detection module of the unmanned salvage boat and the schematic diagram of the boat cover are shown;

[0044] Figure 3 yes Figure 1 The schematic diagram of the structure of the solar-powered boat for salvaging an unmanned ship is shown;

[0045] Figure 4 yes Figure 1 The workflow diagram of salvaging an unmanned vessel is shown;

[0046] Figure 5 yes Figure 1 The structural diagram of the Cascade R-CNN target detection algorithm used by the salvage unmanned vessel shown. DETAILED DESCRIPTION

[0047] The technical solution of the present invention is further specifically described below by examples and in conjunction with the accompanying drawings. In the specification, the same or similar reference numerals indicate the same or similar components. The following description of the embodiments of the present invention with reference to the accompanying drawings is intended to explain the overall inventive concept of the present invention and should not be construed as a limitation of the present invention.

[0048] See also Figure 1 , shows an unmanned salvage boat for image recognition of tiny foreign objects on the water surface according to an embodiment of the present invention. The unmanned salvage boat comprises a hull 10, an image acquisition and processing module 20 and a garbage collection device 30.

[0049] The hull 10 is used to float on the water surface and is equipped with various hardware devices and sensors to enable image recognition, water quality monitoring and normal operation. With reference to the direction of the unmanned salvage boat, the bow 11 of the hull 10 is streamlined to reduce the resistance of navigation.

[0050] A plurality of propulsion propellers 12 are provided at the rear side of the hull 10, which are specifically arranged at the rear side and both sides of the hull 10, placed underwater and completely submerged in water so as to give full play to their propulsion force to propel the ship. In addition, steering can be achieved by controlling the rotation speed of the propellers on both sides.

[0051] The propellers 12 for bilateral propulsion are connected to a drive motor, which is a brushless DC motor. The advantage of a brushless DC motor is that it removes brushes and does not generate sparks during operation, greatly reducing the interference of sparks on radio equipment and various sensors. Differential control is implemented to avoid obstacles during turns, which protects the hull 10 itself.

[0052] Furthermore, a laser radar 13 is installed on the hull on the front side of the hull 10. The principle of the laser radar 13 is similar to that of sonar, and light is used instead of sound to measure the distance between the laser radar 13 and the obstacle. When the hull 10 detects an obstacle in front of it, the hull 10 will automatically retreat and send a signal to achieve autonomous obstacle avoidance.

[0053] The image acquisition and processing module 20 is used to capture the image of the water surface and identify tiny foreign objects through the target detection algorithm. The image acquisition and processing module 20 includes, for example, a front-mounted high-definition camera disposed on the front side of the hull 10 to acquire real-time images of the water surface.

[0054] The image acquisition processing method of the image acquisition processing module 20 comprises the following steps:

[0055] 1) Preprocessing steps, including: image resizing, color space conversion, etc., to ensure that it is suitable for subsequent processing;

[0056] 2) Target recognition step: using the pre-processed image to identify tiny foreign objects on the water surface using a target monitoring algorithm;

[0057] 3) Marking and classification: garbage and obstacles identified by the algorithm are accurately marked and classified.

[0058] See also Figure 5 , wherein the target detection algorithm is an improved Cascade R-CNN target detection algorithm, which combines a feature pyramid network to process multi-scale targets through bottom-up feature extraction and top-down feature fusion.

[0059] The Cascade R-CNN target detection algorithm is based on hierarchical feature extraction and multi-level target detection mechanism, combining the multi-scale feature extraction capability of the Feature Pyramid Network (FPN) and the self-attention mechanism to improve detection accuracy.

[0060] In the bottom-up path, FPN extracts fine-grained features from the low layers, and combines them with high-level semantic information from the top down to fuse these features to adapt to the detection of objects of different scales. Through this multi-scale feature representation, the algorithm can accurately identify floating garbage and potential obstacles of various sizes on the water surface.

[0061] See also Figure 5 The image acquisition and processing module 20 uses an improved Cascade R-CNN target detection algorithm to improve the detection accuracy of tiny floating garbage on the water surface. On this basis, a deeper feature pyramid network (FPN) is introduced to effectively process multi-scale targets through bottom-up feature extraction and top-down feature fusion. The lateral connection of FPN enhances the complementary ability of features at different levels, enabling the model to flexibly identify different types of floating garbage.

[0062] In one example, the Cascade R-CNN target detection algorithm identifies and locates tiny foreign objects on the water surface through a step-by-step optimization mechanism of multi-level classifiers and regressors.

[0063] The Cascade R-CNN uses a multi-level classifier and regressor optimization mechanism to accurately identify and locate small targets on the water surface. This layer-by-layer optimization reduces the missed detection rate and improves detection accuracy and stability, especially in complex water environments.

[0064] In each level of classifier in Cascade R-CNN, a cross entropy loss function is used, and the formula of the cross entropy loss function is:

[0065]

[0066] where p i ,u i is the predicted category probability of the i-th candidate region, N cls is the number of candidate regions.

[0067] In each level of the Cascade R-CNN classifier, the goal is to minimize the error between the true category and the predicted category through the cross entropy loss function. The above formula describes the classification loss at each level, ensuring that the network can optimize the prediction of the target category step by step.

[0068] Specifically, the bounding box regression is performed by combining the self-attention mechanism and the bounding box regression loss function. In the bounding box regression process, the smooth L1 loss function is used to minimize the gap between the predicted box and the true box. The bounding box regression loss function is:

[0069]

[0070] The smooth L1 loss function is defined as:

[0071]

[0072] Among them, t ij and v ij is the coordinates of the predicted box and the real box in the x, y, w, h dimensions, N reg is the number of candidate regions. The formula is used to optimize the bounding boxes of the candidate regions step by step, so that the network can gradually improve the accuracy of the bounding boxes.

[0073] By combining the self-attention mechanism with the bounding box regression loss function, the Cascade R-CNN object detection algorithm of the present invention focuses more on key areas during feature extraction, and uses a smooth loss function to minimize the gap between the predicted box and the true box during the bounding box regression process.

[0074] In order to balance the errors in the classification and regression processes, the total loss function of CascadeR-CNN is defined as:

[0075] L=L cls +λL reg

[0076] Among them, λ is the weight coefficient, which is used to adjust the importance of classification loss and regression loss. In this improved Cascade R-CNN algorithm, by adjusting the weight coefficient λ, we effectively balance the classification loss L cls and regression loss L reg .

[0077] This balanced setting significantly improves detection accuracy, especially in the detection of multi-scale small targets. It provides reliable information support for autonomous navigation, obstacle avoidance and path planning of unmanned ships, thereby achieving efficient water surface cleaning and monitoring. These improvements enable the algorithm to maintain stable performance under variable water surface conditions, providing technical support for environmental protection and water quality monitoring.

[0078] The garbage collection device 30 is installed on the hull and can collect tiny foreign objects on the water surface according to the identification results to clean up the floating garbage on the water surface. The garbage collection device 30 is installed on one or both sides of the hull and is equipped with an expandable PVC bayonet. The garbage collection device 30, such as a scoop net 32, can be replaced to cope with different environments.

[0079] In one embodiment, the garbage collection device 30 includes a double-sided scoop net 32 ​​that is symmetrically arranged and interconnected with the expandable PVC buckles on both sides of the hull 10. The salvage unmanned boat adopts a double-sided scoop net structure, simplifies the hull structure, reduces the hull volume, is easy to carry and deploy, and is suitable for various water surface environments.

[0080] The two sides of the hull 10 are extended detachable structures, for example, external PVC pipes are specifically used to cooperate with PVC bayonet to achieve the function of intercepting garbage. During the operation of the hull 10, only 1 / 4 of the scoop net 32 ​​is submerged in the water, and the hull resistance is small.

[0081] Furthermore, the unmanned salvage boat also includes different functional devices such as a water quality monitoring module 40, a control module 50, a power navigation module 60, and a solar charging device 70.

[0082] The water quality monitoring module 40 can integrate multiple sensors (such as pH sensor, turbidity sensor, conductivity sensor, etc.) to monitor water quality parameters in real time.

[0083] The control module 50 is used to coordinate the navigation, garbage collection, water quality detection and other functions of the hull 10 to ensure stable and efficient operation of the system.

[0084] The powered navigation module 60 uses a multi-degree-of-freedom propulsion device, combined with GPS and UWB dual positioning technology, to achieve precise positioning navigation and flexible steering in complex waters.

[0085] The unmanned salvage boat has the capabilities of autonomous navigation, flexible steering and path planning, and can operate efficiently in complex water environments. It also has a low-power design and two steering systems. The unmanned salvage boat significantly improves the efficiency of surface garbage cleaning and the accuracy of water quality monitoring, and is particularly suitable for narrow waters and water conditions with complex environments.

[0086] The solar charging device 70 adopts a mother-and-child ship design, is installed at the stern and connected to a battery.

[0087] In some embodiments, the garbage collection device 30 can also provide buoyancy to the hull to stabilize the hull.

[0088] In some embodiments, the unmanned salvage vessel also has a remote control mode to cope with various conditions where automatic driving is not possible.

[0089] In some embodiments, for different fine garbage, different data sets can be changed to achieve special salvage of different foreign objects on the water surface.

[0090] After the target detection is completed, the position information of the target is input to the control module 50. The control module 50 generates appropriate control instructions based on the position and route of the target, and these instructions are transmitted to the motor drive module 14 through the PWM signal. The motor drive module 14 will adjust the speed of the propeller and the direction of the steering gear, so that the salvage unmanned ship can automatically adjust the navigation path according to the detection results, realize autonomous navigation and real-time obstacle avoidance. When the unmanned ship detects an obstacle ahead, the system will immediately calculate an alternative route, adjust the direction of the hull to bypass the obstacle, and continue to perform path planning and garbage cleaning tasks.

[0091] The entire detection and navigation process is a closed-loop system. As the unmanned salvage ship moves forward, the laser radar 13 will continue to capture new images, and the image acquisition and processing module 20 will continue to cycle through detection, decision-making, and execution of instructions. Through this closed-loop feedback mechanism, the unmanned salvage ship can dynamically adapt to changes in the water surface environment and always ensure the accurate execution of detection and navigation tasks.

[0092] With reference to the direction in which the unmanned boat is traveling, garbage collection devices 30 are provided on both sides of the hull 10. The garbage storage device 30 is a PVC pipe scoop net for collecting and storing surface garbage. The garbage collection device 30 can be replaced, for example, it can be made of a plastic net, and this structure has a drainage effect. There are PVC buckles between the garbage collection device 30 and both sides of the hull 10, which facilitates the disassembly of the garbage collection device 30, and facilitates transportation and deployment.

[0093] In one embodiment, two brushless motors are provided. In order to reduce the heat generated by the motor speed, two pumping motors 15 are also provided inside the hull 10. The pumping motor 15 uses an aluminum pipe to wrap around the brushless motors on both sides in a surrounding manner. The pumping port and the discharge port are arranged at the tail of the hull 10. The formed vortex also cooperates with the propellers on both sides. When the hull is sailing, the pump will draw river water into the aluminum pipe to cool the motor.

[0094] Taking the direction of the garbage cleaning ship as a reference, a 180° control steering gear 16 is installed at the rear of the hull. The steering gear 16 is linearly connected to the rudder tail 17 at the stern. When the steering gear 16 rotates clockwise, the rudder tail 17 rotates to the left to control the hull to turn to the left. When the steering gear rotates counterclockwise, the rudder tail rotates to the right to control the hull to turn to the right. There is a height difference between the stern and the propeller, so no collision will occur.

[0095] See also Figure 2The main control module inside the hull adopts the special-shaped edge computing platform Jetson Nano. An acrylic box is pasted in the center of the boat cover 18 for the deployment of the main control unit Jetson Nano. The advantage of this structure is that the main control module is fixed on the boat cover, and the hull will not be bumped or damaged by water during operation. A laser radar 13 is deployed on the front deck of the hull. This laser radar 13 integrates a depth sensor and multiple microphone arrays to provide comprehensive perception capabilities. Time of flight (ToF) technology is used for depth sensing, which enables it to accurately capture three-dimensional spatial information under various lighting conditions. The Jetson Nano of the main control module has powerful computing power and parallel processing capabilities, and can handle multiple tasks and threads at the same time, and efficiently perform various computing tasks.

[0096] In the present invention, this module is connected to the various sensors and high-definition cameras mentioned above, and plays the role of transmitting PWM signal control. In addition, it is also connected to the motor drive module, which mainly generates a voltage model to control the brushless motor. In order to accurately track and locate the exact position and route of the unmanned boat for cleaning, this hull is equipped with dual positioning modules: UWB positioning module and GPS positioning module. The UWB positioning module deploys multiple base stations to realize route monitoring in a small area. If a larger area of ​​water appears, the GPS positioning module can also accurately locate the precise coordinates and sailing speed of the ship. Similarly, this module is also connected to the JetsonNano of the main control unit 13.

[0097] like Figure 2 As shown, three water quality monitors are installed on the left side of the hull, namely pH sensor 41, turbidity sensor 42, and water conductivity sensor 43; pH sensor 41 is used to detect the activity of hydrogen ions in water, turbidity sensor 42 comprehensively judges the turbidity situation through the transmittance and scattering rate in water, and water conductivity sensor 43 generates a small current in water by inputting an alternating driving voltage to the test electrode, thereby obtaining the water conductivity of the water area. The three sensors transmit the obtained data to the main control unit JetsonNano through electrical signals.

[0098] See also Figure 3 A power supply boat 80 can be installed on the rear side of the hull as a water platform for the solar charging module 70. The solar charging module 70 of the system converts solar energy into electrical energy, and adjusts the electrical energy to the current and voltage required by the battery through the PWM controller 81, thereby achieving a continuous power supply effect.

[0099] For this invention, a corresponding UI interface is equipped. A camera real-time interface is designed in the upper left corner of the interface to facilitate remote personnel to observe the water surface conditions in real time. A garbage cleaning unit is designed in the upper right corner of the interface to monitor the world coordinates of the hull and the sailing speed in real time. When the sailing speed of the ship is abnormal, the navigation indicator light in the lower left corner of the UI interface turns red. When a bottle disappears from the screen, the number of garbage will be recorded +1. When the amount of garbage reaches the set value, the net bag indicator light in the lower left corner of the UI interface will turn red. In the lower right corner of the page, a water quality monitoring unit is set, which can collect the pH, turbidity, conductivity and other data of the water quality monitoring module. When the measured value of the measured water area exceeds the set value, the indicator light will turn red. When a certain indicator light turns red, the water quality indicator light in the lower left corner turns red accordingly. The UI interface monitors the data of the unmanned ship and the water area in real time for remote personnel to view.

[0100] This cleaning boat has two working modes: manual and autonomous. In manual mode, the operator can control the boat according to monitoring and remote control; in autonomous working mode, it can sail along the designated route according to GPS positioning, plan the route, and automatically avoid obstacles during the garbage collection process. When the garbage is full, the net bag status indicator turns red, and then according to GPS positioning, the main circuit controls the return automatically.

[0101] The hull surface is sealed and waterproof, the control box is sealed, and all connected lines are waterproof and sealed to prevent the humid water environment and rainy weather from affecting the hull's electronic control part.

[0102] See also Figure 4 , shows the workflow diagram of salvaging an unmanned ship, and the specific process will not be repeated here.

[0103] The unmanned salvage boat for surface micro-foreign object image recognition and the image capture and processing method thereof according to the embodiment of the present invention have at least one of the following advantages:

[0104] 1. Efficient garbage collection capability: It can effectively identify and clean floating garbage on the water surface and improve the environmental quality of the water area.

[0105] 2. Real-time water quality monitoring function: Integrate multiple sensors to monitor water quality parameters in real time, providing a scientific basis for water quality management.

[0106] 3. Flexible working mode: Combination of manual and autonomous modes to adapt to different operating requirements and environmental conditions.

[0107] 4. Stable electronic control system: The waterproof design ensures the reliability of the electronic control part in bad weather and enhances the adaptability of the hull.

[0108] 5. Improved detection algorithm: The improved Cascade R-CNN target detection algorithm, combined with a deeper feature pyramid network (FPN) and self-attention mechanism, can accurately identify tiny floating garbage on the water surface and improve detection accuracy.

[0109] 6. Enhanced navigation capability: Improve the decision-making ability of unmanned ships in autonomous navigation and obstacle avoidance by improving the high-quality feature support provided by the algorithm, and achieve efficient path planning;

[0110] 7. The self-attention mechanism further enhances the detection algorithm's regional focusing ability, allowing it to focus more on key target areas, such as floating garbage and surface obstacles. This mechanism dynamically adjusts feature weights to ensure that the model can accurately locate targets when dealing with complex water surface environments. Combined with Cascade R-CNN's hierarchical classifiers and regressors, the detection algorithm can refine the boundaries and categories of targets layer by layer, thereby improving the accuracy of target positioning and the robustness of detection. In the bounding box regression process, a smooth L1 loss function is used to minimize the gap between the predicted box and the true box.

[0111] Although some embodiments of the present general inventive concept have been shown and described, it will be appreciated by those skilled in the art that changes may be made to these embodiments without departing from the principles and spirit of the present general inventive concept, the scope of which is defined by the claims and their equivalents.

Claims

1. An unmanned salvage boat for image recognition of tiny foreign objects on the water surface, characterized in that: The salvage unmanned vessel comprises: hull; Image acquisition and processing module, used to capture images of the water surface and identify tiny foreign objects through target detection algorithms; A garbage collection device, installed on the hull and capable of collecting tiny foreign objects on the water surface according to the identification result to clean up the garbage floating on the water surface; Among them, the target detection algorithm is the Cascade R-CNN target detection algorithm, which combines the feature pyramid network to process multi-scale targets through bottom-up feature extraction and top-down feature fusion.

2. The unmanned salvage vessel according to claim 1, characterized in that: The CascadeR-CNN target detection algorithm identifies and locates tiny foreign objects on the water surface through a step-by-step optimization mechanism of multi-level classifiers and regressors.

3. The unmanned salvage vessel according to claim 2, characterized in that: In each level of classifier in CascadeR-CNN, a cross entropy loss function is used, and the formula of the cross entropy loss function is: where p i ,u i is the predicted category probability of the i-th candidate region, N cls is the number of candidate regions.

4. The unmanned salvage vessel according to any one of claims 1 to 3, characterized in that: Combining the self-attention mechanism and the bounding box regression loss function for bounding box regression, a smooth L1 loss function is used to minimize the gap between the predicted box and the true box during the bounding box regression process. The bounding box regression loss function is: The smooth l1 loss function is defined as: Among them, t ij and ij is the coordinates of the predicted box and the real box in the x, y, w, h dimensions, N reg is the number of candidate regions.

5. The unmanned salvage vessel according to claim 4, characterized in that: When balancing the errors in classification and regression, the total loss function of CascadeR-CNN is defined as: L=L cls +λL reg Among them, λ is the weight coefficient, which is used to adjust the importance of classification loss and regression loss.

6. The unmanned salvage vessel according to claim 5, characterized in that: The garbage collection device is installed on both sides of the hull and is equipped with expandable PVC bayonet.

7. The unmanned salvage vessel according to claim 6, characterized in that: The unmanned salvage boat also includes a water quality detection module for real-time monitoring of water quality parameters.

8. The unmanned salvage vessel according to claim 7, characterized in that: The unmanned salvage boat also includes a control module for coordinating navigation, garbage collection, and water quality testing of the boat.

9. The unmanned salvage vessel according to claim 8, characterized in that: The salvage unmanned vessel also includes: Power navigation module, adopting multi-degree-of-freedom propulsion device and using GPS and UWB dual positioning technology; The solar charging device adopts a mother-and-daughter ship design, is installed at the stern of the hull, and is connected to the battery.

10. An image capture and processing method for a salvage unmanned boat for surface micro-foreign object image recognition according to any one of claims 1 to 9 comprises the following steps: 1) Preprocessing step, the preprocessing step includes: image size adjustment and color space transformation; 2) Target recognition step: using the pre-processed image to identify tiny foreign objects on the water surface using a target monitoring algorithm; 3) Marking and classification: garbage and obstacles identified by the algorithm are accurately marked and classified.

Images