Pomacea canaliculata egg mass monitoring and treatment method based on cooperation of low-altitude unmanned aerial vehicle and water surface unmanned ship

Abstract

The invention provides a pomacea canaliculata egg mass monitoring and treatment method based on cooperation of a low-altitude unmanned aerial vehicle and a water surface unmanned ship. The pomacea canaliculata egg mass monitoring and treatment method comprises the following steps: performing image acquisition and coordinate positioning on pomacea canaliculata egg masses in a monitored water area through the unmanned aerial vehicle to construct a pomacea canaliculata egg mass preliminary distribution map; the accurate positioning coordinates are divided through a DBSCAN clustering method, so that the monitoring water area is divided into different ecological risk level areas; according to ecological risk level area division information, a density hotspot priority list of ampullaria gigas egg mass removal tasks is set, a high-risk emergency area with large ampullaria gigas egg mass density is processed preferentially, and by combining weather, water flow and obstacle factors in a monitored water area, an unmanned ship completes ampullaria gigas egg mass removal route planning; the unmanned ship accurately removes the pomacea canaliculata egg masses in the monitored water area at fixed points according to the removal route; according to the method, the ampullaria gigas egg masses can be quickly identified and removed in a wide water area, so that dynamic monitoring and continuous treatment are realized, the ecological environment of the water area is effectively protected, and the water quality recovery capability is improved.

Description

Technical Field

[0001] The present invention belongs to the field of data processing systems or methods for supervision or prediction purposes, and particularly relates to a method for monitoring and controlling the egg masses of Pomacea canaliculata by the cooperation of low-altitude unmanned aerial vehicles and unmanned surface vessels, and a preparation method thereof. Background Art

[0002] With the advancement of the urbanization process, many urban inland waters have suffered serious pollution, and the water ecological environment has deteriorated day by day. As an alien species, the egg masses of Pomacea canaliculata reproduce in large numbers in waters, not only damaging the water quality, but also affecting the growth of aquatic species and even endangering human health. Traditional removal methods rely on manual and mechanized operations, which have problems such as low efficiency, narrow coverage, and high costs, and are difficult to meet the governance requirements of large-scale waters.

[0003] In summary, a method that can quickly identify and remove the egg masses of Pomacea canaliculata in a wide range of waters to achieve dynamic monitoring and continuous governance, so as to effectively protect the water ecological environment and improve the water quality restoration ability, is urgently needed to be developed. Summary of the Invention

[0004] In view of the above-mentioned disadvantages of the prior art, to solve the technical problems in the prior art that it is impossible to quickly identify and remove the egg masses of Pomacea canaliculata to achieve dynamic monitoring and continuous governance, so as to effectively protect the water ecological environment and improve the water quality restoration ability;

[0005] The purpose of the present invention is also to provide a method for monitoring and controlling the egg masses of Pomacea canaliculata by the cooperation of low-altitude unmanned aerial vehicles and unmanned surface vessels, including the following steps:

[0006] S1: Collect images of the egg masses of Pomacea canaliculata in the monitored waters by an unmanned aerial vehicle and perform coordinate positioning to construct a preliminary distribution map of the egg masses of Pomacea canaliculata;

[0007] S2: Then divide the accurately positioned coordinates by the DBSCAN clustering method, so as to divide the monitored waters into different ecological risk level regions;

[0008] S3: According to the information on the division of the ecological risk level regions, set a density hot spot priority list for the task of removing the egg masses of Pomacea canaliculata, give priority to processing high-ecological-risk emergency regions with a large density of the egg masses of Pomacea canaliculata, and combine the weather, water flow, and obstacle factors in the monitored waters to complete the route planning for the unmanned surface vessel to remove the egg masses of Pomacea canaliculata;

[0009] S4: The unmanned surface vessel accurately and thoroughly removes the egg masses of Pomacea canaliculata at fixed points in the monitored waters according to the removal route;

[0010] S5: After the first clearance of the unmanned boat in the monitoring waters, the unmanned boat returns to the starting point to standby; the unmanned aerial vehicle takes off again at high altitude to collect images within the monitoring waters and uploads the image data to the cloud in real time for egg mass identification and clearance effect evaluation. The cloud compares the data after the re-flight with the previous clearance records, determines whether the egg masses are completely cleared, and updates the ecological risk zoning map of the waters according to the results; if it is found that the egg masses in some areas are not cleared, the cloud will adjust the operation path and priority, and re-arrange the unmanned boat to perform the clearance task again until all clearance tasks are completed; among them,

[0011] In S1, the image acquisition includes that the unmanned aerial vehicle first flies low along the monitoring area, completes continuous environmental mapping and the collection of the images of the apple snail egg masses through SLAM, and then rises to high altitude to collect the global image of the monitoring area; among them,

[0012] During the collection process of the images of the apple snail egg masses, the apple snail egg masses are detected, identified and filtered out through YOLOv5 image recognition in the cloud to obtain the monitoring and positioning coordinate map of the apple snail egg masses; the NWD loss function is added in the YOLOv5 image recognition; and,

[0013] The risk level is divided as follows: after the DBSCAN clustering method is completed, the density values of the apple snail egg masses are statistically calculated for several clusters obtained by clustering, sorted according to the density values of the apple snail egg masses in the clusters, and the 80% quantile is selected as the critical value of the high ecological risk area on the sorted data; the 40% quantile is used as the critical value of the medium ecological risk area; and the area less than the 40% quantile is regarded as the low ecological risk area.

[0014] Preferably, the low altitude range is 8-20 meters; the high altitude range is 30-80 meters.

[0015] Preferably, in S4, when the unmanned boat arrives in the designated clearance area, it is necessary to re-locate and confirm the apple snail egg masses in this area first. If it is found that the apple snail egg masses are consistent with the corresponding positions in the monitoring and positioning coordinate map of the apple snail egg masses, the high-speed and high-pressure water flow ejected by the water gun installed on the unmanned boat is used to clear the apple snail egg masses; if it is found that they are inconsistent, the position of the unmanned boat needs to be finely adjusted until the corresponding apple snail egg masses are identified, and then the clearance is performed; if the corresponding apple snail egg masses still cannot be identified after multiple fine adjustments, the unmanned boat will go to the next clearance area and repeat the above steps until the current clearance task is completed.

[0016] Preferably, the removal of Pomacea canaliculata egg masses includes information collection, control decision-making, and execution adjustment. Among them, information collection includes collecting the positioning information of the target Pomacea canaliculata egg masses and the water gun spraying angle information; control decision-making includes the angle error and distance error between the water gun pointing and the target, and the PID controller in the information processing module will process the two errors and generate a control signal; execution adjustment includes driving the motor to adjust the horizontal and pitching angles of the pan-tilt according to the control signal, dynamically adjusting the feedback, and regulating the PWM water pump to spray water.

[0017] Preferably, obtain the positioning information of the target Pomacea canaliculata egg masses and the water gun spraying angle information:

[0018] Taking the camera as the origin, the three-dimensional space coordinates of the target Pomacea canaliculata egg masses relative to the camera are P c =(x c , y c , z c );

[0019] Taking the water gun as the origin, the three-dimensional space coordinates of the target Pomacea canaliculata egg masses relative to the water gun are P w =(x w , y w , z w );

[0020] The internal parameter matrix of the camera Among them, f x and f y are the focal lengths of the camera, and cx and cy are the coordinates of the optical center of the camera;

[0021] When the Pomacea canaliculata egg masses appear in the picture of the depth camera, assuming that the two-dimensional pixel coordinates of the target detected by the camera are (u, v), then through the internal parameter matrix k of the depth camera and the depth information z c , the three-dimensional space coordinates of the target in the camera coordinate system can be obtained:

[0022]

[0023] Also, since the relative spatial positions of the camera and the water gun remain unchanged, the target can be transformed from the camera coordinate system to the water gun coordinate system through the external parameter matrix T cw :

[0024] P w =T cw ·P c ; Among them,

[0025] Among them is a 4x4 rigid body transformation matrix; including the rotation matrix R 3x3 , the translation vector t 3x1 ;

[0026] According to the three-dimensional spatial coordinate position (x w , y w , z w ) of the target object in the water gun coordinate system, that is, P w , the pitch angle θ that the pan-tilt needs to adjust can be calculated, that is, the rotation angle and horizontal angle that the water gun needs to adjust in the vertical plane that is, the rotation angle that the water gun needs to adjust in the horizontal plane.

[0027] Preferably, a prediction mechanism for the growth cycle of Pomacea canaliculata egg masses is superimposed on the basis of the density hot spot priority list.

[0028] Preferably, the prediction mechanism for the growth cycle of Pomacea canaliculata egg masses captures the law of Pomacea canaliculata spawning by using the LSTM model for historical environmental climate data, historical water quality data and historical spawning time intervals in the monitoring area, and then predicts the time of the next spawning.

[0029] Preferably, the prediction mechanism for the growth cycle of Pomacea canaliculata egg masses is expressed by the following formula:

[0030] T next = T base + ΔT season + ΔT environment ; where

[0031] T next is the predicted time of the next spawning;

[0032] T base is the reference value, the spawning time when there are no other factors affecting;

[0033] ΔT season is the error caused by seasonal factors;

[0034] ΔT environment is the error caused by water quality environment factors;

[0035] For the reference value, we use the mean model for calculation:

[0036] where is the spawning time of historical observations; N is the total number of historical data.

[0037] Preferably, it further includes a safety margin T safetymargin :

[0038] Use MSE to measure the training error of the LSTM model, and define a weighting factor α safety to control the size of the safety margin;

[0039]

[0040] Wherein:

[0041] MSE: The mean square error predicted by the current model;

[0042] MSE max : The maximum mean square error in history, used as the basis for standardization;

[0043] ∈: A small constant to prevent the denominator from being zero

[0044] To better consider the role of the reference value T base A weighting factor β is introduced base To compare the difference between the reference value and the predicted value:

[0045] Wherein,

[0046] T next : The next spawning time predicted according to the model;

[0047] T base : The reference value, the spawning time when there is no other influence;

[0048] ∈: A small constant to prevent the denominator from being zero;

[0049] β base-max : The maximum value of the limit on β base To prevent the patrol inspection time interval from being overly affected when the prediction error is too large; when the prediction is relatively accurate, set it to 0.2 - 0.5; when the prediction is unstable, set it to 0.5 - 1.0;

[0050] Combined with α safety and β base , the final formula for the safety margin T safetymargin is designed as:

[0051] T safety margin = T next ×(1 - α safety ×MSE)+ β base ×|T base - T next |.

[0052] Preferably, it further includes the design of the patrol inspection time interval:

[0053] If the predicted value is T next , then the patrol inspection time interval should be:

[0054] T inspection = min(T next, T base , T safety margin );

[0055] Wherein:

[0056] T next : The predicted next spawning time according to the model;

[0057] T base : The reference value, the spawning time when assuming no other factors affect;

[0058] T safety margin: The safety margin for the predicted time interval based on the uncertainty of the prediction.

[0059] The preparation method of the monitoring and control method for Pomacea canaliculata egg masses by the cooperation of low-altitude drones and surface unmanned boats given in this case has the following beneficial effects:

[0060] 1) In the present invention, the traditional IoU loss is replaced by the NWD loss function. NWD can effectively solve the inaccurate problem of IoU in the label assignment of small target objects, thereby improving the detection accuracy of small targets such as Pomacea canaliculata egg masses;

[0061] 2) The cloud in the present invention has pre-judged the distribution of egg masses based on the images and coordinates collected by the low-altitude drone. However, due to certain dynamic changes in the distribution of egg masses (such as the influence of water flow, object occlusion, etc.), each egg mass must be reconfirmed; the unmanned boat will use its own vision system to re-identify the specific position, size and shape of the egg mass to ensure the accurate execution of the removal task;

[0062] 3) The design of the inspection time interval provides a scientific and flexible time arrangement for inspectors by combining the predicted spawning interval, the reference spawning time and the safety margin. This method can not only ensure the timeliness of the inspection, but also avoid excessive or insufficient inspections, ensure the reasonable use of resources, and ultimately achieve the effective monitoring and management of the spawning behavior of Pomacea canaliculata.

[0063] 4) The hierarchical cooperation of low-altitude and high-altitude inspections takes into account both detailed collection and large-scale overview; the present invention introduces a monitoring method combining low-altitude flight and high-altitude flight in step S1, so that both detail clarity and area coverage can be obtained: low-altitude images can provide high-resolution pictures required for accurate identification, and high-altitude aerial photography can enable managers to master the overall distribution and hydrological environment. And it can effectively reduce missed detections and fill in blind spots: during low-altitude inspections, if there are local obstacles blocking the line of sight, subsequent high-altitude aerial photography can timely supplement information, and vice versa; improve the inspection efficiency: through the mutual verification of the two types of flight data, the cloud can quickly focus on suspected high-risk areas or missed detection areas, thereby reducing ineffective flight routes and shortening the inspection cycle.

[0064] 5) The present invention proposes the concept of "coordination between low-altitude drones and unmanned surface vessels", which not only refers to the collaborative work of drones (aerial vehicles) and unmanned surface vessels (marine / water surface vehicles), but also implies that it can be extended to various types of water environments such as coastal areas, estuaries, lakes, and inland rivers; the benefits brought by this air-water three-dimensional governance include: adapting to diverse water terrain: in cases such as shoals, tortuous river channels, and tidal changes, drones and unmanned surface vessels can flexibly switch operation routes; reducing the risk of manual operations: there is no need for a large amount of manpower to enter the water or take a boat to search for egg masses at close range, which is both safe and efficient; quickly responding to emergencies: if an extremely high-risk area is found during patrol, the drone can immediately mark the location, and the unmanned surface vessel can quickly go to deal with it.

[0065] 6) It can be integrated with multi-source environmental monitoring data to form comprehensive decision support for water area governance; in the system of the present invention, in addition to the image and coordinate data collected by the drones and unmanned surface vessels themselves, multi-source environmental monitoring data (such as air temperature, water temperature, pH, nutrient salt concentration, river water quality sensor data, etc.) can also be incorporated into the cloud database; a more comprehensive risk assessment can be obtained: the cloud can, while evaluating the density of Pomacea canaliculata egg masses, combine indicators such as water quality conditions or algae concentration to judge the overall health of the water ecosystem; accurately identify the reasons for the high incidence of egg masses: if in some water areas, due to high water temperature or eutrophication, Pomacea canaliculata lays eggs concentratedly, the governance strategy can be changed to make it more targeted, such as subsequent release of biological inhibitors, etc.; long-term monitoring and trend prediction: the accumulated multi-dimensional data can further train a more accurate time series model to conduct a more in-depth analysis and prediction of the evolution of the water ecosystem and the breeding law of Pomacea canaliculata.

[0066] 7) In the traditional manual patrol and salvage and removal mode, operators need to spend a lot of time and effort to search, record, and remove Pomacea canaliculata egg masses. The air-water collaborative system of the present invention has a high degree of automation and intelligence; it can greatly save labor: managers only need to carry out supervision, strategy adjustment, and a small amount of maintenance work; it can achieve 24-hour operation: compared with manual work, the intelligent unmanned platform can work at night or in multiple shifts continuously when necessary, shortening the disposal time before the hatching of Pomacea canaliculata egg masses; it provides sufficient human safety guarantee: effectively reducing the risks encountered during water area operations, such as underwater undercurrents, silt subsidence, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flow chart for the drone of the present invention to collect images and locate Pomacea canaliculata egg masses in the river and construct a map.

[0068] Figure 2 It is a schematic diagram of the risk zoning of Pomacea canaliculata egg masses in the urban inland water area generated by DBSCAN according to the present invention.

[0069] Figure 3This invention is a flowchart of the Pomacea canaliculata spawning inspection mechanism based on LSTM (Long Short-Term Memory, a type of recurrent neural network).

[0070] Figure 4 Schematic diagram of the cooperative work of the drone and the unmanned boat

[0071] Figure 5 This invention is a flowchart of the unmanned boat cruising to remove Pomacea canaliculata egg masses

[0072] Figure 6 This invention is a schematic diagram of the drone's working principle

[0073] Figure 7 This invention is a schematic diagram of the unmanned boat's working principle

[0074] Figure 8 Schematic diagram of the cooperative work of the cloud with the drone and the unmanned boat Detailed implementation manners

[0075] The following specific embodiments illustrate the implementation manners of the present invention. Those skilled in the art can easily understand the other advantages and effects of the present invention from the content disclosed in this specification.

[0076] Embodiment 1:

[0077] A method for monitoring and removing Pomacea canaliculata egg masses through sea-air cooperation, including the following steps:

[0078] I. Drone monitoring: (as Figure 1 shown)

[0079] Visual SLAM mapping: During the flight of the drone, the Visual SLAM (Simultaneous Localization and Mapping) technology is used to achieve high-precision self-localization and environmental mapping; the SLAM system calculates the position and attitude of the drone in the three-dimensional space in real time through the continuous images captured by the camera and the acceleration and angular velocity data provided by the IMU (Inertial Measurement Unit), so as to construct a map of the river channel and the surrounding environment; the drone updates the map through continuous movement, and at the same time matches and optimizes the environmental features to ensure the accuracy and real-time nature of the map. In this process, Visual SLAM can help the drone achieve autonomous positioning in urban environments where there is no GPS signal or the GPS signal is unstable.

[0080] Image acquisition and coordinate recording: After completing the SLAM mapping, the drone will continue to fly low along the river, focusing on acquiring images of the apple snail egg masses on the riverbank wall; the drone automatically controls its flight path to ensure full coverage of the areas where egg masses may exist; while acquiring images, the drone records the precise coordinate information corresponding to each image; the image and its coordinate data will be uploaded to the cloud in real time for subsequent image analysis, egg mass identification, and risk assessment.

[0081] Global image acquisition: After completing the low-altitude flight and egg mass identification, the drone will ascend to a higher altitude to conduct global image acquisition, overlooking the entire river water area; high-altitude image acquisition helps obtain more comprehensive environmental data, including the overall layout of the river, water area scope, flow velocity, bridges, and other factors that may affect path planning and unmanned boat operations; the drone uploads the panoramic images acquired from high altitude to the cloud, and the cloud, through image analysis, uses the number and coordinates of the obtained egg masses to generate a risk zoning map of the entire river using the DBSCAN density clustering algorithm.

[0082] YOLOv5 image recognition: The image data acquired by the drone is uploaded to the cloud server in real time, and the cloud uses the YOLOv5 image recognition model to detect and identify the apple snail egg masses in the images. Since the apple snail egg masses are small targets, in order to improve the detection effect, the NWD (Normalized Wasserstein Distance) loss function is introduced on the basis of the original YOLOv5. Different from the traditional IoU loss function, the NWD loss function is insensitive to objects of different scales, so it is more suitable for measuring the similarity between tiny objects. By replacing the traditional IoU loss, NWD can effectively solve the inaccurate problem of IoU in small target object label assignment, thus improving the detection accuracy of small targets such as apple snail egg masses.

[0083] Risk zoning generation (DBSCAN clustering): Based on the egg mass coordinates identified by YOLOv5, the cloud uses the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm to perform spatial clustering on the egg masses and identify the distribution pattern of the egg masses in the water area. Through DBSCAN, the cloud can divide the water area into different risk level regions, such as Figure 2 shown.

[0084] The risk level division is as follows: After the DBSCAN clustering method is completed, count the apple snail egg mass density values for several clusters obtained by clustering, sort them according to the apple snail egg mass density values of the clusters, and select the 80% quantile as the critical value for the high-risk area on the sorted data; the 40% quantile as the critical value for the medium-risk area; and those less than the 40% quantile are regarded as low-risk areas.

[0085] Path planning and task allocation: Based on the global image data, the precise coordinates of the apple snail egg masses, and the results of the risk zoning, the cloud will use the A*(A-star) path planning algorithm to generate the optimal driving path from the current position of the unmanned boat to the high-ecological-risk area. The path planning will consider various factors in the water area, such as obstacles, water flow, bridges, etc., to ensure that the unmanned boat can safely and efficiently complete the egg mass removal task. The cloud will also adjust the task priorities according to the real-time situation to ensure that the most urgent areas are processed in a timely manner.

[0086] II. Unmanned boat cruising to remove apple snail egg masses (as Figure 5 shown):

[0087] 1. High-precision positioning and path execution: The unmanned boat needs to navigate precisely in a complex water area environment to ensure that it can accurately reach the designated target area for egg mass removal operations; for this purpose, the unmanned boat adopts RTK-GPS (Real-Time Kinematic GPS) combined with LiDAR (Light Detection and Ranging) technology for high-precision positioning.

[0088] 2. Task execution: According to the task instructions and path planning issued by the cloud, the unmanned boat will automatically execute the designated tasks. The cloud generates path planning instructions based on the egg mass distribution in the water area and the ecological risk level, and gives priority to letting the unmanned boat handle the egg masses in high-risk areas. The unmanned boat will travel along the planned path, collect data on the surrounding environment in real time, avoid obstacles and accurately reach the target position.

[0089] 3. Egg mass removal and high-pressure water gun control

[0090] Secondary egg mass recognition: When the unmanned boat arrives at the designated operation area, it uses the on-board camera for secondary egg mass recognition. The cloud has pre-judged the distribution of the egg masses based on the images and coordinates collected by the low-altitude unmanned aerial vehicle. However, due to certain dynamic changes in the distribution of the egg masses (such as the influence of water flow, object occlusion, etc.), it is necessary to confirm each egg mass again. The unmanned boat will use its own vision system to re-identify the specific position, size, and shape of the egg masses to ensure the accurate execution of the removal task. The camera cooperates with the YOLOv5 algorithm to quickly identify and locate through the images captured in real time, and accurately lock the coordinates of the egg masses.

[0091] Egg mass removal: The egg mass removal technology of this design is specifically realized by the depth camera to capture and locate the target apple snail egg masses, the two-dimensional servo to aim at them, and the PWM water pump to spray water with variable frequency. The specific working principle of this step is divided into four main stages: information collection, control decision-making, execution adjustment, and the takeoff and secondary monitoring of the unmanned aerial vehicle.

[0092] A. Information collection: The IMU sensor measures the attitude angles (such as pitch angle and horizontal angle) of the pan-tilt in real time. By obtaining the current orientation of the pan-tilt, the control system can determine whether the actual attitude of the pan-tilt is consistent with the expected one. The specific process is as follows:

[0093] Target object positioning

[0094] Define the camera coordinate system: Take the camera as the origin, and the three-dimensional space coordinates of the target object are P c =(x c , y c , z c );

[0095] Define the water gun coordinate system: Take the water gun as the origin, and the three-dimensional space coordinates of the target object are P w =(x w , y w , z w );

[0096] The internal parameter matrix of the camera f x and f y are the focal lengths of the camera, and cx and cy are the optical center coordinates.

[0097] When the apple snail egg mass appears in the depth camera's image, assuming the camera detects the two-dimensional pixel coordinates of the target object as (u, v), then through the internal parameter matrix k of the depth camera and the depth information z c , the three-dimensional space coordinates of the target object in the camera coordinate system can be obtained:

[0098]

[0099] Also, since the relative spatial positions of the camera and the water gun remain unchanged, the target object can be transformed from the camera coordinate system to the water gun coordinate system through the external parameter matrix T cw (by translation and rotation):

[0100] P w =T cw ·P c

[0101] where is a 4x4 rigid body transformation matrix; including the rotation matrix R 3x3 , the translation vector t 3x1 .

[0102] Calculate the water gun spraying angle

[0103] According to the three-dimensional space coordinate position (x w , y w , z w) The pitching angle θ (the rotation angle that the water gun needs to adjust in the vertical plane) and the horizontal angle ψ (the rotation angle that the water gun needs to adjust in the horizontal plane) that the pan-tilt needs to adjust can be calculated.

[0104]

[0105] B. Control decision:

[0106] Error calculation: Based on the target distance fed back by the depth camera and the current angle information provided by the IMU sensor, the system calculates the angle error and distance error between the current water gun pointing and the target. For example, if the target position is within the view of the pan-tilt but the water flow deviates from the target, the system calculates the angle between the target and the current orientation of the water gun to obtain the error value.

[0107] PID algorithm adjustment: The PID controller in the information processing module processes this error and generates a control signal.

[0108] C. Execute adjustment:

[0109] The motor drives the pan-tilt to adjust the angle: According to the control signal generated by the PID controller, the motor drives the pan-tilt to adjust the angle (including horizontal adjustment and pitching adjustment) to align the water gun with the target.

[0110] Dynamic adjustment feedback: The system monitors the pointing of the water gun and the change of the distance to the target in real time. If a new error appears (such as the target object moves or the direction of the water gun deviates), the control system will calculate the error and perform PID adjustment again to drive the pan-tilt to make further adjustments.

[0111] Regulate the PWM water pump to spray water: After obtaining the specific position of the target object, the water pump will use the PID controller to regulate the duty cycle of the PWM water pump in real time according to the dosing distance and the lift, gradually approaching the ideal range to ensure that the liquid flow finally hits the target; a high duty cycle (such as 80%-100%) corresponds to high power and high flow, suitable for long-distance range or high lift; a low duty cycle (such as 20%-50%) corresponds to lower power and flow, suitable for short-distance range.

[0112] When the egg masses are not recognized after the unmanned boat reaches the designated area, the hull needs to be adjusted again. This situation may occur when the position of the egg masses is not within the scanning range, or due to factors such as water surface fluctuations and obstacle occlusion. Therefore, a mechanism for the unmanned boat to move in a small range is set up to ensure that the target area is covered as much as possible; the specific process is as follows:

[0113] 1. Egg masses not recognized: After the unmanned boat reaches the target area, if the egg masses cannot be detected in the secondary recognition, the control system will trigger the hull to move in a small range. The unmanned boat will make fine adjustments in the local area around the original position to adjust the hull position so that the camera can cover the areas that may be missed.

[0114] 2. Minor adjustments: The unmanned boat will perform small forward, backward, left and right rotation movements to ensure that the camera scans every possible area of the water surface. If the egg mass is still not recognized, the egg mass recognition will be performed again.

[0115] 3. Egg mass not recognized after three adjustments: If the unmanned boat still fails to recognize the egg mass after making three small hull adjustments in the target area, it will determine that there is no egg mass in this area or the distribution of the egg mass is too scattered. At this time, the unmanned boat will go to the next clearing target area according to the instructions issued by the cloud to continue the operation.

[0116] D. Drone takeoff again and secondary monitoring:

[0117] After the task is completed, the unmanned boat will return to the starting point along the predetermined path, and at the same time the drone will perform the takeoff again task for secondary monitoring; the drone collects area images by flying at high altitude and uploads the data to the cloud in real time for egg mass recognition and clearing effect evaluation; the cloud compares the takeoff again data with the previous clearing records, determines whether the egg mass is completely cleared, and updates the risk zoning map of the water area according to the results; if it is found that the egg mass in some areas is not cleared, the cloud will adjust the operation path and priority and re-arrange the unmanned boat to perform the subsequent clearing tasks.

[0118] Embodiment 2:

[0119] The prediction method for Pomacea canaliculata spawning inspection based on LSTM (long short-term memory neural network) includes the following steps: As Figure 3 shown,

[0120] S1. Obtain the following data:

[0121] Environmental climate indicators: daily average temperature, daily average humidity, monthly average precipitation, daily average water temperature;

[0122] Water quality indicators: water body PH, water body dissolved oxygen concentration, permanganate index, salinity, ammonia nitrogen content; and

[0123] The spawning interval of historical Pomacea canaliculata.

[0124] S2. Data processing:

[0125] Normalize and scale the collected data, scale the data range to [0, 1] to avoid the influence caused by different dimensions between features. The calculation formula for the normalized value is:

[0126] where, x min and x max are the minimum and maximum values of this feature respectively

[0127] S3. Model Architecture Design:

[0128] The spawning time of Pomacea canaliculata is affected by various factors. To subdivide the impacts of environmental climate indicators and water quality indicators on its spawning, the prediction process is broken down as follows:

[0129] T next = T base + ΔT season + ΔT eniroment where T next is the predicted next spawning time;

[0130] T base is the baseline value, the spawning time when assuming no other factors are affecting;

[0131] ΔT season is the error caused by seasonal factors;

[0132] ΔT environnent is the error caused by water quality environment factors.

[0133] For the baseline value, we use the mean model for calculation:

[0134] where

[0135] is the spawning time of historical observations;

[0136] N is the total number of historical data.

[0137] For seasonal errors, they are usually closely related to time (months, seasons), temperature, etc. For water quality environment errors, they are usually related to multiple physical and chemical parameters of the water body (pH, water temperature, dissolved oxygen, ammonia nitrogen concentration, permanganate index); therefore, we both use LSTM to capture the temporal characteristics of seasonal changes and learn the impact of water quality environment on spawning time.

[0138] The formula for the seasonal error prediction model can be expressed as:

[0139] ΔT season = f season (time features) where

[0140] ΔT season is the deviation caused by seasonal factors (such as temperature, humidity, months, etc.);

[0141] f season is the seasonal time series function learned by the LSTM model, with the input being historical time features (daily average temperature, daily average humidity, monthly average precipitation, daily average water temperature), and LSTM predicts seasonal errors by capturing the long-term dependencies of the time series.

[0142] The specific structure of the LSTM model with seasonal influence is as follows:

[0143] h t = LSTM(x t , h t-1 ); where:

[0144] x t is the input feature at the time step;

[0145] h t is the hidden state of the LSTM, reflecting the seasonal pattern over a past period of time;

[0146] f season finally outputs the seasonal deviation through the fully connected layer;

[0147] The water quality environment error prediction formula is:

[0148] ΔT environment = f environment (water quality features); where,

[0149] ΔT environment is the deviation caused by water quality environmental factors;

[0150] f environment is the LSTM model, with the input being historical water quality data (pH, water temperature, dissolved oxygen, ammonia nitrogen concentration, permanganate index), and the LSTM captures the impact of water quality on the spawning time.

[0151] The LSTM structure of the water quality error is as follows:

[0152] h t = LSTM(x t , h t- 1); where, x t is the water quality data; f enviroment outputs the water quality environment error

[0153] S4. Outlier Detection and Classification:

[0154] The goal of outlier detection is to identify those spawning time prediction results that do not conform to the normal time series pattern. This design uses outlier detection based on Isolation Forest. Isolation Forest is a tree-based ensemble method that can effectively detect outliers isolated in the data. It randomly selects features and randomly partitions the data, gradually "isolating" the outlier data points, so it is very efficient in dealing with high-dimensional data.

[0155] S5. Loss Function Design:

[0156] To simultaneously consider the prediction accuracy of the model and the influence of outliers, this design adopts a weighted loss function, adding a penalty term for outliers on the basis of a conventional loss function (such as mean squared error, MSE). In this way, the model will not only focus on how to optimize normal data points but also penalize those possible outlier predictions during the training process.

[0157] In a conventional LSTM model, we use mean squared error (MSE) as the loss function, and its calculation formula is:

[0158] where, y i

[0159] is the true spawning time, is the predicted value of the model; when dealing with outliers, we need to impose a greater penalty on the outlier points. When an outlier is detected by the Isolation Forest model, this design penalizes the outlier by introducing a weighting factor to make its contribution to the loss function greater.

[0160] The formula for the weighted loss function is:

[0161] where,

[0162] α i = 1 for normal data points; α i > 1 for outlier data points, and the penalty factor α for outliers is set to 10 to make the influence of outliers on the loss function greater.

[0163] Outlier weighting formula:

[0164]

[0165] L1 Norm Loss (Lasso Regularization):

[0166] To prevent overfitting, this design combines L1 norm loss (Lasso Regularization) to further constrain the complexity of the model. The loss function of L1 regularization is:

[0167] where,

[0168] θ i is the parameter of the model, and λ is the regularization hyperparameter. This regularization term will help the model maintain sparsity and avoid unnecessary complexity.

[0169] In summary, this design will use a loss function that combines weighted mean squared error (MSE) and L1 regularization:

[0170] Among them,

[0171] α i is a weighting factor, and λ is a regularization coefficient.

[0172] S6. Training and optimization:

[0173] To improve the prediction ability of the model, this design uses the Adam optimizer to train the LSTM model. The Adam optimizer can automatically adjust the learning rate to adapt to changes in different parameters, and is especially suitable for training complex deep learning models.

[0174] The update rule of the Adam optimizer is:

[0175] Among them,

[0176] θ t are the model parameters, m t and v t are the mean and variance of the gradient respectively;

[0177] η is the learning rate, and ∈ is a small constant to prevent division by zero;

[0178] Based on the optimized LSTM model, the next spawning time of the apple snail can be predicted, and the final predicted time is:

[0179] T next = T base + ΔT season + ΔT environment

[0180] Inspection strategy: Through the training and prediction based on the LSTM model, we can obtain the next spawning time interval (T next ) of the apple snail. This value is a prediction result, representing the number of days from the current moment to the next spawning of the apple snail. This predicted value fully considers various influencing factors such as historical data, seasonal changes, and water quality environment, so it can relatively accurately reflect the spawning cycle of the apple snail.

[0181] However, the actual occurrence of spawning may be affected by various unexpected factors and may deviate from the model prediction to a certain extent. Therefore, relying solely on the prediction result to arrange inspections is not completely reliable, and a certain safety margin needs to be added to ensure the safety and effectiveness of the inspection work. Therefore, the inspection strategy adopted in this design is as follows:

[0182] Design of the safety margin T safety margin :

[0183] Use the MSE to measure the training error of the LSTM model and define the weighting factor α safety to control the size of the safety margin;

[0184]

[0185] where,

[0186] MSE: The mean squared error predicted by the current model;

[0187] MSE max : The largest mean squared error in history, used as the basis for standardization;

[0188] ∈: A small constant to prevent the denominator from being zero;

[0189] To better consider the role of the reference value T base introduce a weighting factor β base to compare the difference between the reference value and the predicted value;

[0190] where,

[0191] T next : The next spawning time predicted by the model;

[0192] T base : The reference value, the spawning time when no other factors are considered;

[0193] ∈: A small constant to prevent the denominator from being zero;

[0194] β base-max : The maximum value of the limit on β base to prevent excessive influence on the inspection time interval when the prediction error is too large. When the prediction is relatively accurate, set it to 0.2 - 0.5; when the prediction is unstable, set it to 0.5 - 1.0.

[0195] Combining α safety with β base , the final formula for the safety margin T safetymargin is designed as:

[0196] T safety margin =T next ×(1 - α safety ×MSE)+β base ×|T base -T next |;

[0197] If the predicted value is T next , then the inspection time interval should be: T inspection =min(Tnext , T base , T safety margin ); where

[0198] T next : The predicted next spawning time according to the model;

[0199] T base : The reference value, the spawning time when assuming no other factors affect

[0200] T safety margin : Based on the uncertainty of the prediction, it is the safety margin for the prediction time interval.

[0201] Through the above design, the inspection time interval (T inspection ) can provide a flexible and safe time window to perform the inspection task while taking into account the spawning time. For each predicted spawning cycle, the model dynamically adjusts the inspection time node according to the actual environmental and water quality conditions.

[0202] If T next is shorter, that is, the predicted next spawning cycle is shorter, the inspection time interval will be closer, ensuring that the inspection personnel can arrive at the scene in time for inspection.

[0203] If T base is longer, that is, the spawning cycle of Pomacea canaliculata shows a longer cycle in the historical data, the inspection time interval will be relatively looser, and the inspection personnel can appropriately postpone the inspection task.

[0204] The role of the safety margin T safety margin : The safety margin ensures that the inspection time will not be overly delayed, avoiding the risk of missed inspections due to external environmental changes or data prediction errors.

[0205] The above embodiments only illustratively explain the principles and effects of the present invention, rather than limiting the present invention. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes completed by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A method for monitoring and controlling the egg masses of Pomacea canaliculata using a low-altitude drone and a surface unmanned boat, characterized in that: The following steps are involved: S1: Use drones to collect images and locate the coordinates of the Pomacea canaliculata egg masses in the monitored waters to construct a preliminary distribution map of the Pomacea canaliculata egg masses; S2: dividing the accurate positioning coordinates by DBSCAN clustering method, thereby dividing the monitored water area into areas with different ecological risk levels; S3: according to the ecological risk level area division information, set the density hotspot priority list of the golden apple snail egg mass removal task, give priority to the high ecological risk emergency area with large density of golden apple snail egg masses, and complete the route planning of the unmanned boat for the removal of golden apple snail egg masses in combination with the weather, water flow and obstacle factors in the monitored waters; S4: The unmanned boat removes the egg masses of the golden snails in the monitored water area accurately and thoroughly according to the removal route; S5: After the unmanned boat has cleared the monitored waters for the first time, it returns to the starting point and waits for orders; the drone takes a high-altitude go-around to collect images of the monitored waters and uploads the image data to the cloud in real time for egg mass identification and removal effect evaluation. The cloud compares the go-around data with the previous removal records to determine whether the egg masses are completely removed, and updates the ecological risk zoning map of the waters based on the results; if it is found that egg masses in some areas have not been removed, the cloud will adjust the operation path and priority, and reschedule the unmanned boat to perform the removal task again until all removal tasks are completed; among them, The image acquisition in S1 includes the UAV first flying at low altitude along the monitoring area, completing continuous environment mapping and Pomacea canaliculata egg mass image acquisition through SLAM, and then rising to high altitude to acquire global images of the monitoring area; wherein, In the process of collecting the egg mass images of the Pomacea canaliculata, the egg mass of the Pomacea canaliculata is detected, identified and removed by YOLOv5 image recognition in the cloud, so as to obtain a monitoring and positioning coordinate map of the egg mass of the Pomacea canaliculata; an NWD loss function is added to the YOLOv5 image recognition; and, The risk level is divided into: after the DBSCAN clustering method is completed, the density values ​​of the egg masses of the Pomacea canaliculata of several clusters obtained by clustering are counted, and the clusters are sorted according to the density values ​​of the egg masses of the Pomacea canaliculata of the clusters. The 80% quantile is selected as the critical value of the high ecological risk area on the sorted data; the 40% quantile is selected as the critical value of the medium ecological risk area; and the area below the 40% quantile is regarded as a low ecological risk area.

2. The method for monitoring and controlling the egg masses of Pomacea canaliculata using a low-altitude drone and a surface unmanned boat according to claim 1, characterized in that: The low altitude range is 8 to 20 meters; the high altitude range is 30 to 80 meters.

3. The method for monitoring and controlling the egg masses of Pomacea canaliculata using a low-altitude drone and a surface unmanned boat according to claim 2, characterized in that: In S4, when the unmanned boat arrives at the designated clearing area, it is necessary to first perform a secondary positioning confirmation on the golden apple snail egg masses in the area. If it is found that the golden apple snail egg masses at the corresponding position of the golden apple snail egg mass monitoring and positioning coordinate map are consistent, the golden apple snail egg masses are cleared by a high-speed high-pressure water flow shot by a water gun installed on the unmanned boat; if inconsistency is found, the position of the unmanned boat needs to be fine-tuned until the corresponding golden apple snail egg mass is identified, and then the clearing is performed; if the corresponding golden apple snail egg mass still cannot be identified after multiple fine-tuning, the unmanned boat goes to the next clearing area and repeats the above steps until this clearing task is completed.

4. The method for monitoring and controlling the egg masses of Pomacea canaliculata using a low-altitude drone and a surface unmanned boat according to claim 3 is characterized in that: The removal of golden apple snail egg masses includes information collection, control decision and execution adjustment; wherein, information collection includes collecting target golden apple snail egg mass positioning information and water gun spray angle information; control decision includes angle error and distance error between the water gun pointing and the target, and the PID controller in the information processing module will process the two errors and generate a control signal; execution adjustment includes driving the motor to adjust the level and pitch angle of the gimbal according to the control signal, dynamically adjusting feedback and regulating the PWM water pump to spray water.

5. The method for monitoring and controlling the egg masses of Pomacea canaliculata using a low-altitude drone and a surface unmanned boat according to claim 4 is characterized in that: Obtain the location information of the target Pomacea canaliculata egg mass and the water gun spray angle information: Taking the camera as the origin, the three-dimensional space coordinates of the target Pomacea canaliculata egg mass relative to the camera are P c =(x c ,y c , z c ); Taking the water gun as the origin, the three-dimensional spatial coordinates of the target Pomacea canaliculata egg mass relative to the water gun are P W =(x w ,y w , z w ); Camera's intrinsic matrix Among them, f x and f y is the focal length of the camera, cx and cy are the coordinates of the optical center of the camera; When the egg mass of Pomacea canaliculata appears in the image of the depth-of-field camera, assuming that the camera detects the two-dimensional pixel coordinates of the target object as (u, v), the intrinsic parameter matrix k of the depth-of-field camera and the depth information z c , we can get the three-dimensional space coordinates of the target object in the camera coordinate system: Since the relative spatial position of the camera and the water gun remains unchanged, the camera external parameter matrix T cw Convert the target from the camera coordinate system to the water gun coordinate system: P w =T cw ·P c ;in, in is a 4x4 rigid body transformation matrix; including the rotation matrix R 3x3 , translation vector t 3x1 ; According to the three-dimensional space coordinate position (x w ,y w , z w ), namely P w , we can calculate the pitch angle θ that the gimbal needs to adjust. That is, the rotation angle and horizontal angle of the water gun that need to be adjusted in the vertical plane That is, the rotation angle of the water gun needs to be adjusted in the horizontal plane.

6. The method for monitoring and controlling the egg masses of Pomacea canaliculata using a low-altitude drone and a surface unmanned boat in collaboration according to claim 1 or 5, characterized in that: A prediction mechanism for the growth period of the egg mass of the golden apple snail is superimposed on the density hotspot priority list.

7. The method for monitoring and controlling the egg masses of Pomacea canaliculata using a low-altitude drone and a surface unmanned boat according to claim 6, characterized in that: The prediction mechanism for the egg mass growth cycle of the golden apple snail is to capture the spawning law of the golden apple snail by using the LSTM model to analyze the historical environmental climate data, historical water quality data and historical spawning time intervals in the monitoring area, and then predict the time of the next spawning.

8. The method for monitoring and controlling the egg masses of Pomacea canaliculata using a low-altitude unmanned aerial vehicle and a surface unmanned boat according to claim 7 is characterized in that: The prediction of the growth cycle of the Pomacea canaliculata egg mass is expressed by the following formula: T next =T base +ΔT season +ΔT environment ;in, T next is the predicted time of next spawning; T base is the baseline value, assuming that there are no other factors affecting the spawning time; ΔT season It is the error caused by seasonal factors; ΔT environment It is the error caused by water quality and environmental factors; For the benchmark value, we use the mean model for calculation: in, is the spawning time of historical observations; N is the total number of historical data.

9. The method for monitoring and controlling the egg masses of Pomacea canaliculata using a low-altitude unmanned aerial vehicle and a surface unmanned boat according to claim 8, characterized in that: Also includes the safety margin T safetymargin : Use MSE to measure the training error of the LSTM model and define the weighting factor α safety To control the size of the safety margin; in: MSE: mean square error of the current model prediction; MSE max : The largest mean square error in history, used as the basis for standardization; ∈: a small constant to prevent the denominator from being zero; In order to better consider the reference value T base The role of introducing a weighting factor β base To compare the difference between the baseline and the predicted values: in, T next : The next spawning time predicted by the model; T base : Baseline value, assuming that there are no other factors affecting the spawning time; ∈: a small constant to prevent the denominator from being zero; β base-max : For β base The maximum value of the limit is to prevent excessive impact on the inspection time interval when the prediction error is too large; when the prediction is relatively accurate, set it to 0.2~0.5; when the prediction is unstable, set it to 0.5~1.0; Combined with α safety With β base , safety margin T safetymargin The final formula is designed as: T safetymargin =T next ×(1-α safety ×MSE)+β base ×|T base -T next |。 10. The method for monitoring and controlling the egg masses of Pomacea canaliculata using a low-altitude unmanned aerial vehicle and a surface unmanned boat in collaboration according to claim 9, characterized in that: Also includes the design of inspection time intervals: If the predicted value is T next , then the inspection interval should be: T inspection =min(T next , T base , T safetymargin ),in: T next : The next spawning time predicted by the model; T base : Baseline value, assuming that there are no other factors affecting the spawning time; T safetymargin : A safety margin for the prediction time interval based on the uncertainty of the prediction.

Images