A reservoir dam fish passing system for crossing river cableway sling fish and unmanned aerial vehicle fish transportation

Abstract

The application discloses a reservoir dam fish passing system of cross-river cableway sling fish gathering and unmanned aerial vehicle fish conveying, relates to the technical field of data processing, and comprises image acquisition, feature calculation, batch division, shunt control, air-drop optimization and air-drop execution modules. Fish features are extracted by using a model based on image acquisition, and the volume and mass of single bodies are estimated. Based on ecological compatibility, suitable flow speed and load constraints, batches are executed, fish groups are guided to temporary breeding tanks, and jet flow is controlled to generate a matching flow speed of rotational water flow to unify the posture of fish groups. After the fish groups are sucked into the air by the unmanned aerial vehicle, optimization calculation is performed in combination with environmental wind speed and a dynamics model to obtain optimal air-drop height and speed. During air-drop, a jet flow is emitted to generate a water surface buffer area according to a delay time, and a water bag is thrown. The application overcomes the limitation of high dam terrain, realizes intelligent batch grouping of multiple fish species, and effectively guarantees the survival rate of live fish in high-drop release by means of a fluid active load shedding mechanism.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a fish transport system for reservoir dams that uses cross-river cableway for fish collection and drone-based fish transport. Background Technology

[0002] While water conservancy projects provide comprehensive benefits such as flood control, power generation, and navigation, they also alter the connectivity of natural river channels, blocking the migration routes of migratory fish and adversely affecting their critical life cycle events. This leads to the degradation of aquatic ecosystems and the decline of fish resources. To mitigate the damage to the ecological environment caused by dam blockage, the construction of fish passage facilities has become a key technical means to restore aquatic ecological connectivity and protect fish populations. Traditional fish passage facilities mainly include fixed fishways, fish gates, and fish lifts. Chinese invention patent application publication number CN119026814A discloses a method and device for optimizing the construction of fish passage facilities based on multidimensional impact assessment. This technical solution identifies the critical life cycle events of target fish in the watershed, combines the distribution range, movement path, and movement direction of the target fish, and identifies the corresponding dam data. Based on this, the solution calculates the first and second mitigation effects of the obstruction of fish free movement before and after the construction of fish passage facilities, and obtains an evaluation index through weighted summation to guide the site selection for the construction of dam fish passage facilities. The aforementioned existing technologies focus on the macro-level site selection assessment and planning of fish passage facilities, but still have limitations in the specific physical fish transfer and refined execution aspects. This approach remains based on the traditional construction concept of fixed fish passage facilities, failing to address the physical challenges of difficult transfer channel layout and adaptability to drastic water level changes under the complex terrain of high-head dams. In actual fish collection and transportation processes, different fish species exhibit significant differences in physiological indicators such as ecological compatibility, suitable swimming flow velocity, and shock tolerance. Existing macro-assessment technologies lack real-time, accurate visual recognition and intelligent batching and diversion mechanisms for individual characteristics within mixed fish populations crossing dams, easily leading to physiological stress or swimming fatigue due to flow velocity mismatch during mixed transport. In the fish transport and final release stages, which require overcoming large drops in elevation, existing technologies fail to fully construct a system of impact dynamics characteristics of live fish entering the water, lacking effective active fluid buffering and dynamic de-loading control mechanisms. When fish fall into the water, they are very susceptible to extremely high instantaneous overload impact due to improper water-cutting posture or lack of buffer medium, which can cause mechanical tissue rupture of internal organs, making it impossible to guarantee the survival rate of the fish. Summary of the Invention

[0003] The technical problem solved by this invention is that existing technologies are difficult to achieve flexible cross-air transport while overcoming the limitations of complex terrain of high dams, and at the same time, they are unable to complete intelligent batch matching based on the multidimensional physiological and dynamic characteristics of live fish, and lack an effective active fluid buffering and unloading mechanism to ensure a high survival rate during the high drop release and transport process.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a fish transport system for reservoir dams using cross-river cableway suspension for fish collection and drone-based fish transport, comprising: The image acquisition module is used to acquire images of the fish school and obtain a fish school image matrix; The feature calculation module is used to input the fish image matrix into a pre-trained fish feature extraction model, output the category number and rotation box parameters of each fish, and calculate the estimated volume and estimated mass of a single fish by combining it with a pre-stored fish parameter database, and generate fish feature records. The batch division module is used to perform a batching algorithm on the fish feature records, divide the fish feature records into different transportation batches, and obtain the target flow rate and safe overload lower limit corresponding to the transportation batches; The diversion control module is used to control the fish to enter the corresponding holding tank according to the transport batch, and to control the jet nozzle array on the side wall of the holding tank to generate a circumferential swirling water flow according to the target flow rate. The airdrop optimization module is used to control the drone to descend the spherical fish transport water tank and dock with the temporary holding box to suck up the fish. After the drone takes off, it performs airdrop optimization calculations and outputs the optimal airdrop plan. The airdrop execution module is used to control the UAV to fly to the drop point, unfold the bottom hatch to form a water bag, calculate the jet launch delay time based on the optimal airdrop scheme, and execute the airdrop.

[0005] Furthermore, the fish feature extraction model includes a main feature branch, a feature pyramid fusion network, a classification prediction detection head, and a regression prediction detection head; The parameters of the rotation frame are the principal axis pixel length, secondary axis pixel width, deflection angle, and center point pixel coordinates of the smallest bounding rectangle of the fish target. The process of calculating the estimated volume and estimated mass of a single fish specifically includes: Based on a preset size conversion factor, the main axis pixel length and the secondary axis pixel width are converted into physical projection length and physical projection width; The corresponding body proportion correction factor is queried from the fish parameter database according to the category number, and the product of the body proportion correction factor and the physical projection width is used as the fish body thickness. The fish body shape is abstracted as a triaxial ellipsoid, with the physical projection length as the major axis, the physical projection width and the fish body thickness as the two minor axes, and the estimated volume of a single fish is calculated. The preset average fish density is used, and the product of the estimated volume and the average fish density is used as the estimated mass of a single fish.

[0006] Furthermore, the batching algorithm process specifically includes: When the first fish feature record is obtained, the first batch is created and stored in that fish feature record; When a subsequent fish feature record is obtained, it is compared with the fish feature records in the existing batch to check whether the merged batch meets the batch constraints. If all constraints are met, the fish feature record is stored in the first existing batch that meets all constraints, and the cumulative mass and cumulative volume of the batch are updated. If none of the existing batches meet the constraints, a new batch is created and stored if the total number of batches has not reached the upper limit. The batch constraints include a first mass constraint, a second volume constraint, a third ecological compatibility constraint, and a fourth velocity intersection constraint.

[0007] Furthermore, the batching algorithm process also includes: When the number of batches already created reaches the maximum limit for the total number of batches that can be created and no new batches can be created, the batch division module performs a judgment on the current fish characteristic record according to the following logic: Check whether the fish characteristic record meets the first quality constraint and the second volume constraint of any existing batch. If none of the existing batches meet the constraints, then classify the record into the warning batch. For candidate batches that meet the first quality constraint and the second volume constraint, further check whether they meet the third eco-compatibility constraint. If all candidate batches do not meet the constraint, the record is classified as a warning batch. If there are candidate batches that satisfy the first quality constraint, the second volume constraint, and the third ecological compatibility constraint, but are unable to be included in the batch only because they do not meet the fourth velocity intersection constraint, then a downgrade process is triggered. The downgrade process specifically includes: The candidate batches that meet the first three constraints are selected as fusion candidate batches. The average estimated quality of the fish already contained in them is calculated. The current fish feature records are forcibly assigned to the fusion candidate batch with the smallest average estimated quality and updated as the fusion batch.

[0008] Furthermore, the process of obtaining the target flow rate and safe overload lower limit corresponding to the batch specifically includes: For a fusion batch that has undergone degradation processing, the mean of the appropriate flow rate range for each record within it is extracted, and a weighted average is performed with the estimated quality as the weight to obtain the target flow rate of the fusion batch. The minimum value of the critical acceleration of fish impact damage recorded in the fusion batch is obtained by multiplying each of the recorded critical accelerations by a preset reduction factor and then multiplying it by a preset safety redundancy factor. The lower limit of the safe overload of the fusion batch is then calculated. For normal batches that have not undergone degradation processing, query the critical acceleration of fish impact damage in all fish characteristic records within the normal batch, take the minimum value and multiply it by the preset safety redundancy coefficient as the lower limit of safe overload for the normal batch, calculate the midpoint of the intersection interval of suitable swimming velocity for the normal batch, and use it as the target velocity for the normal batch.

[0009] Furthermore, the process of controlling the jet nozzle array on the side wall of the temporary holding tank to generate a circumferential swirling water flow according to the target flow velocity specifically includes: Send a switching command to the electric gate of the diversion pool to open the corresponding target holding tank channel and close the other channels, so that the water carries the same batch of fish into the designated holding tank. The target flow velocity is substituted into a preset linear conversion formula to calculate the target driving frequency of the water pump. The safe driving frequency is output, and the frequency converter is controlled to drive the water pump so that the tangential jet nozzle array on the side wall of the temporary holding tank generates a circumferential swirling water flow with a linear velocity equal to the target flow velocity.

[0010] Furthermore, the process of performing the airdrop optimization calculation specifically includes: Construct an airdrop kinematic sequence and calculate the free fall time of the water bag based on the drop height; Calculate the vertical water-cutting velocity based on the free fall time; The ground speed of the drone's horizontal flight is used as the initial horizontal absolute speed at the moment of water bag release. The horizontal water cutting speed at the moment of water impact is calculated by combining the horizontal wind speed and the preset wind deflection response coefficient. The combined entry angle is calculated based on the vertical cutting velocity and the horizontal cutting velocity.

[0011] Furthermore, the process of performing the airdrop optimization calculation also includes: A dynamic model of water impact is constructed, and the radius of the equivalent sphere is inversely calculated based on the volume of the spherical water tank. The moment when the bottom of the water tank touches the water is set as the starting point of time iteration, and the initial time, initial immersion depth, initial horizontal absolute velocity component and initial vertical absolute velocity component are set as the initial integral values. In each time step of the iterative calculation, the immersion depth is calculated based on the current vertical velocity integral, and the instantaneous cross-sectional area is dynamically calculated. The instantaneous dynamic impact acceleration overload is calculated. The instantaneous dynamic impact acceleration overload is proportional to the square of the instantaneous cross-sectional area and the resultant velocity. The calculation incorporates the water momentum correction coefficient, the water density of the reservoir area, and the total mass of the fish-water mixture in the current batch. The direction of the drag acceleration is always opposite to the instantaneous velocity vector. Its horizontal and vertical components are distributed in real time according to the proportion of the current velocity component. The simultaneous differential equations of the horizontal and vertical velocity components are simulated and iterated using a first-order Runge-Kutta integrator with a preset time step until the instantaneous vertical velocity decays to zero. The immersion depth, instantaneous cross-sectional area, and instantaneous velocity components are updated synchronously within each time step. The absolute peak acceleration on the time axis of the entire water entry process is extracted and converted into a dimensionless overload factor. Perform a triple security check, which includes a first check, a second check, and a third check; The first determination is that the overload multiple is less than or equal to the lower limit of safe overload; The second determination limits the synthetic cutting angle to within a preset safety range; The third determination is that the horizontal offset distance between the main water bag and the jet contact point is less than the offset tolerance value.

[0012] Furthermore, the process of performing the airdrop optimization calculation also includes: Within the feasible solution space that satisfies the triple safety criteria, an energy consumption evaluation function is constructed with the goal of minimizing the UAV's flight energy consumption for optimization search, and the optimal horizontal flight speed and optimal drop height are output. If the triple safety determination under the current environmental parameters results in an empty set of feasible solutions, the second determination is abandoned, the throwing height is forcibly set to the preset safety limit lower limit, and the horizontal speed of the single target is re-searched with the minimum overload multiple under the premise of strictly maintaining the third determination. If the obtained minimum overload multiple exceeds the batch safe overload lower limit, it is determined that the current airdrop is not feasible, and the bottom hatch of the fish transport water tank is locked; If the obtained minimum dimensionless overload multiple does not exceed the batch safety overload lower limit, the horizontal speed and the preset safety limit lower limit of the drop height obtained by the re-search will be sent to the UAV control system.

[0013] Furthermore, the process of calculating the jet emission delay time specifically includes: Calculate the free fall time of the main water tank based on the optimal throwing height; The jet flight time is calculated based on the jet exit velocity and the optimal throwing height. The difference between the free fall time of the main water bag and the flight time of the jet is taken as the jet launch delay time.

[0014] The beneficial effects of this invention are as follows: This invention introduces a parallel batch processing algorithm based on multidimensional physical and biological constraints, comprehensively considering the lifting mass limit of the UAV, the buffer volume threshold of the water tank, the ecological compatibility of fish species, and the appropriate flow velocity intersection range, to divert and group scattered and disordered fish groups; This invention introduces an active attitude control mechanism based on hydrodynamics in the fish temporary holding and lifting and transfer process. The diversion control module adjusts the tangential jet nozzle array on the side wall of the temporary holding tank according to the target flow velocity calculated for each batch, generating a circumferential swirling water flow that matches the swimming ability of the fish group in that batch. Utilizing the natural flow-oriented instinct of fish, this mechanism forces the mixed fish group to spontaneously adjust and maintain a uniform hovering attitude that is parallel to the current. The airdrop optimization module of this invention comprehensively considers wind deflection compensation, water bag falling kinematics, and instantaneous dynamic impact overload. Under the triple safety judgment of satisfying the lower limit of safe overload, the no-bounce entry angle range, and the effective buffer offset tolerance of the jet, it calculates the optimal horizontal flight speed and drop height that minimize energy consumption. Attached Figure Description

[0015] Figure 1 This is a basic flowchart of a reservoir dam fish passage system for cross-river cableway fish collection and drone fish transport, provided as an embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example, refer to Figure 1 A fish-passing system for reservoir dams, comprising a cross-river cableway for fish collection and a drone-based fish transport system, includes: The image acquisition module is used to acquire images of the fish school and obtain a fish school image matrix; The feature calculation module is used to input the fish image matrix into the pre-trained fish feature extraction model, output the category number and rotation box parameters of each fish, and calculate the estimated volume and estimated mass of a single fish by combining it with the pre-stored fish parameter database, and generate fish feature records. The batch segmentation module is used to perform a batching algorithm on fish feature records, divide the fish feature records into different transportation batches, and obtain the target flow rate and safe overload lower limit corresponding to the transportation batch; The diversion control module is used to control the fish to enter the corresponding holding tank according to the transportation batch, and to control the jet nozzle array on the side wall of the holding tank to generate a circumferential swirling water flow according to the target flow rate. The airdrop optimization module is used to control the drone to descend and dock with the spherical fish transport tank and the temporary holding box and suck up the fish. After the drone takes off, it performs airdrop optimization calculations and outputs the optimal airdrop plan. The airdrop execution module is used to control the drone to fly to the drop point, unfold the bottom hatch to form a water bag, calculate the jet launch delay time based on the optimal airdrop scheme, and execute the airdrop.

[0018] The fish transfer facility in this embodiment is equipped with four independent holding tanks, each corresponding to an independently dispatchable transport batch and a docking drone. Each drone carries a spherical fish transport tank equipped with a life-sustaining circulating water pump to maintain dynamic water circulation and prevent oxygen depletion in the fish population. The spherical tank is surrounded by an annular pure water jacket with a 5-liter capacity, pre-filled with pure water and equipped with a micro-pump for metered water replenishment from the main tank. At the bottom of the jacket is a jet booster pump and a jet nozzle with a fixed downward tilt angle of 90° (vertically downward spraying). The rated operating pressure of the jet booster pump can accelerate the pure water in the jacket to an outlet velocity of 35 m / s. An industrial camera deployed directly above the fish transport pipe from the fish collection box to the temporary holding box in the dam cableway continuously captures top-down images of the fish sliding down the pipe at a frame rate of 60Hz. Each frame is converted from analog to digital and output as a three-dimensional digital matrix. The matrix has a width of 512 pixels, a height of 512 pixels, and a depth of 3 channels. The three channels correspond to the intensity values ​​of red, green, and blue, respectively. Each pixel in each channel represents the reflected light intensity at that wavelength using a dimensionless integer from 0 to 255. This three-dimensional digital matrix is ​​denoted as the fish swarm image matrix. .

[0019] The fish feature extraction model includes a main feature branch, a feature pyramid fusion network, a classification prediction detection head, and a regression prediction detection head; The parameters of the rotation box are the principal axis pixel length, secondary axis pixel width, deflection angle, and center point pixel coordinates of the smallest bounding rectangle of the fish target. As the cableway fish collection box is raised to the water surface and begins to discharge fish into the fish transport pipe, the mixed fish swarms slide down randomly with the water flow. The fish swarm image matrix is ​​extracted frame by frame. For each frame of the image, multi-object detection and quality estimation are performed. Specifically, the fish school image matrix is ​​input into a pre-trained fish feature extraction network, and the class number, rotation rectangle parameters and center point pixel coordinates of each fish in the fish school image are output. The fish feature extraction network adopts an improved network based on the target detection architecture, including a backbone feature branch, a feature pyramid fusion network, a classification prediction detection head and a regression prediction detection head; The main feature branch uses a ResNet-50 architecture, containing 50 convolutional layers, to process the input fish swarm image matrix. Extract multi-scale high-dimensional spatial feature tensors; The feature pyramid fusion network is connected at the end of the main branch, and its downsampling ratio relative to the input image is 4 times, 8 times, 16 times and 32 times respectively, and the number of channels is 256, 512, 1024 and 2048 respectively. The classification prediction detection head and the regression prediction detection head are connected in parallel at the output of the feature pyramid fusion network. The classification prediction detection head outputs the category number for each fish target appearing in the image. The regression prediction detection head uses a rotating bounding box prediction mechanism, which outputs the principal axis pixel length, secondary axis pixel width, deflection angle, and center point pixel coordinates of the smallest bounding box for each target. The fish feature extraction network can output the category number, rotation rectangle parameters, and center point pixel coordinates of all fish in the image in parallel in a single frame. The pre-training process of the fish feature extraction network specifically includes: An industrial camera was used to continuously capture 10,000 frames of top-view images of multiple targets, multiple poses, and partial occlusion in the fish transport pipeline. Each frame was labeled by aquaculture technicians with the true species category of each fish and the true pixel coordinates center, principal axis length, secondary axis width, and deflection angle of the minimum bounding rectangle of rotation. During training, images and annotations are input into the network. The classification detection head outputs classification predictions, and the regression detection head outputs position, size, and angle predictions. The total loss function consists of two weighted sums: Focal Loss for classification and Smooth L1 loss for regression for the rotated rectangle. The loss values ​​for center point coordinate offset, principal axis length offset, secondary axis length offset, and angle offset are calculated separately. The weight of each loss term is set to 1.0 and they are summed in a 1:1:1:1 weighted ratio. The angle loss is obtained by calculating the Smooth L1 loss of the difference between the sine and cosine values ​​of the predicted deflection angle and the true deflection angle. The Adam optimizer is used to backpropagate and update the weights and biases of all layers. The initial learning rate is set to 0.0001 and the batch size is set to 32. After each complete traversal of the training set, if the average detection accuracy on the validation set does not improve for 10 consecutive rounds, training is stopped and the network parameters are fixed. In this embodiment, each row of the fish parameter database corresponds to the ID of a known fish species within the reservoir basin. This ID is pre-established and written into the database based on a fish resource survey of the basin. The fish parameter database contains five fixed fields, the first of which is an ecological compatibility classification label. , stored in positive integer form, fish with the same ecological compatibility classification label value represent fish whose diets do not constitute a predator-prey relationship and whose habitats overlap, and are allowed to be mixed in the same batch; The second field is the lower limit of the suitable flow velocity for swimming. The unit is meters per second; The third field is the upper limit of the suitable flow rate for swimming. The unit is meters per second. The lower limit and upper limit of the suitable swimming flow velocity were obtained through incremental flow velocity experiments in swimming tunnels. The fourth field is the critical acceleration for impact damage to fish. With gravitational acceleration The value is stored in multiples of the value, representing the maximum upward acceleration limit that the fish species can withstand at the moment of entering the water. Exceeding this value will cause mechanical tissue rupture of the internal organs or swim bladder. The critical acceleration for fish impact damage is determined by impact experiments, such as placing the fish in a water bag with an embedded acceleration sensor and dropping it into the water from different heights or speeds, or using a catapult device to simulate the impact of falling into the water. The fish are then dissected and the survival rate is observed to determine the maximum critical acceleration value that will not cause mechanical tissue rupture. The fifth field is the body proportion correction factor. This indicates the ratio of the fish's thickness to its width. The method for obtaining the data is to catch at least 30 samples of different sizes for each type of fish, measure their body length, body width and body thickness, calculate the ratio of body thickness to body width, take the average value and pre-store it in the database.

[0020] The process of calculating the estimated volume and estimated mass of a single fish includes: Based on the preset size conversion factor, the main axis pixel length and secondary axis pixel width are converted into physical projection length and physical projection width; The physical projection length is the product of the main axis pixel length and the preset size conversion factor, and the physical projection width is the product of the secondary axis pixel width and the preset size conversion factor. The preset size conversion factor is obtained through checkerboard calibration and is expressed in meters per pixel. To overcome the technical limitation that monocular top-down vision cannot perceive the fish's thickness dimension, this embodiment queries the corresponding body proportion correction factor from the fish parameter database according to the category number, and uses the product of the body proportion correction factor and the physical projection width as the fish's thickness. The fish's body shape is abstracted as a triaxial ellipsoid, with the physical projection length as the major axis and the physical projection width and fish body thickness as the two minor axes, to calculate the estimated volume of a single fish; The mathematical expression for estimating volume is: ; in, Estimate the volume of a single fish in cubic meters. The physical projection length. The physical projection width, The thickness of the fish's body; The preset average fish density is called, and the product of the estimated volume and the average fish density is used as the estimated mass of a single fish. The average fish density is 1020 kg / m³. This value is obtained by measuring the volume and weighing 10 common economic fish species in the reservoir basin, such as silver carp, bighead carp, and grass carp, using the displacement method, calculating the density, and then taking the average. This invention addresses the technical limitation of monocular top-down vision in directly perceiving the thickness dimension of fish bodies. By combining a pre-trained fish feature extraction network with body proportion correction factors from a pre-stored fish parameter database, it can estimate the physical volume and mass of a single fish online after obtaining the two-dimensional projection parameters of a school of fish, providing individual feature parameter support for subsequent safe batching algorithms.

[0021] The batching algorithm process specifically includes: An improved parallel batch processing algorithm is executed for each fish feature record. This algorithm divides scattered fish feature records into different transport batches. The execution process of the improved parallel batch processing algorithm specifically includes: When the first fish feature record is obtained, the first batch is created and stored in that fish feature record; After completing the species identification, volume estimation, and mass estimation for each fish, a set of structured data corresponding to the fish is obtained as a fish characteristic record. Each fish characteristic record includes the estimated mass, estimated volume, and ecological compatibility tags, suitable flow velocity lower limit, suitable flow velocity upper limit, and impact damage critical acceleration retrieved from the fish parameter database. When a subsequent fish feature record is obtained, it is compared with the fish feature records in the existing batch to check whether the merged batch meets the batch constraints. If all constraints are met, the fish feature record is stored in the first existing batch that meets all constraints, and the cumulative mass and cumulative volume of the batch are updated. If none of the existing batches meet the constraints, a new batch is created and stored if the total number of batches has not reached the upper limit. Batch constraints include the first mass constraint, the second volume constraint, the third ecological compatibility constraint, and the fourth velocity intersection constraint; During the initialization phase, when the first fish feature record is acquired, the first batch is created, denoted as . And the characteristics of the fish are recorded and stored. ,at this time The cumulative mass is the estimated mass in the feature record of this fish, and the cumulative volume is the estimated volume in the feature record of this fish. When the second fish characteristic record is obtained, it is compared with... The existing fish characteristic records are compared to check if the merged record meets the four batch constraints. If all constraints are met, the fish characteristic record is also stored. and update The cumulative mass and cumulative volume; if either is not satisfied, a second batch is created, denoted as... Store it ; When each subsequent fish feature record arrives, it is divided sequentially according to the batch order. The first batch that satisfies all constraints is stored. If none of the existing batches are satisfied, a new batch is created before the total number of batches reaches the upper limit. To prevent system deadlock caused by the number of created batches exceeding the total number of holding tanks due to compatibility exclusion or individual physical limits during continuous high-density multi-species fish transfer, in this embodiment, the fish transfer facility is equipped with 3 independent holding tanks as regular transport batch carrying units, and 1 spare holding tank as an abnormal handling unit. Therefore, in this embodiment, the maximum number of created batches is set to 3. The first mass constraint determination condition is that when the spherical water tank is full, the fish body mass corresponding to the existing fish feature records of the current batch and the fish feature records to be included is added together, and the total mass of the fish and water mixture is calculated together with the mass of the life support water body on the base, and it is determined whether the total mass of the fish and water mixture does not exceed the maximum safe lifting mass of the UAV. The spherical water tank has an open overflow structure when fish are sucked into the holding tank. When a fish is sucked into the tank, an equal volume of holding water is discharged, keeping the total fluid volume within the tank constant. After the fish is sucked in, the tank is full. The mathematical expression for the total mass of the fish and water mixture is: ; in, The total mass of the fish and water mixture is expressed in kilograms. The volume of the spherical water tank is... The density of the reservoir water is taken as 1000 kg / m³. This is the estimated volume of the i-th fish in the batch. The average density of the fish is taken as 1020 kg / m³; In this embodiment, the full-load volume of the spherical water tank is set to... cubic meters, the maximum safe lifting mass of the drone is 105 kg, calculated after adding the characteristic records of the fish to be included. If the weight does not exceed 105 kg, then the first constraint is satisfied; The second volume constraint determination condition is: sum the fish body volume corresponding to the existing fish feature records in the current batch and the fish feature records to be included, and determine whether the total volume obtained does not exceed the preset buffer threshold. In this embodiment, the preset buffer threshold is 0.03 cubic meters. Based on the water tank volume of 0.10 cubic meters, the total volume of the batch of fish is required to not exceed 0.03 cubic meters. If this condition is met, the second constraint is satisfied. The specific calibration experiment for the buffer threshold includes: In a 0.10 cubic meter spherical water tank with the same volume as the water tank in this embodiment, simulated fish bodies with different volume proportions were placed. Silicone models with a density similar to that of fish bodies were used, resulting in fish body volume proportions of 20%, 25%, 30%, 35%, and 40%, corresponding to buffer water volume proportions of 80%, 75%, 70%, 65%, and 60%. With the water tank fully loaded, repeated air-drop experiments were conducted from a height of 20 meters, 10 times per group. The morphology of the water bag upon contact with the water was observed using high-speed video, and the surface of the silicone model was checked for signs of direct collision with the water tank wall. The experimental data are shown in Table 1. Experimental results show that when the buffer water volume ratio is not less than 70%, the fish model does not directly hit the wall, and the number of continuous water film ruptures is controlled within an acceptable range. When the buffer volume ratio drops to 65%, the number of wall hits and the number of film ruptures increase significantly. Therefore, the buffer water volume ratio threshold is set at 70%, based on a water tank volume of 0.10 cubic meters, corresponding to a total volume of no more than 0.03 cubic meters for the batch of fish. The third ecological compatibility constraint is determined by: whether the ecological compatibility classification label in the record to be included is the same as the ecological compatibility classification label of all existing records in the current batch. Fish species with the same ecological compatibility tag, whose diets do not predate each other and whose habitats overlap, are allowed to be mixed-species. If they are to be included in the ecological compatibility tag record... If the ecocompatibility label is the same as all existing records in the current batch, then the third constraint is satisfied; The fourth flow velocity intersection constraint determination condition is: extract the maximum value of the lower limit of suitable swimming flow velocity and the minimum value of the upper limit of suitable swimming flow velocity from all fish feature records in the current batch and the fish feature records to be included, and determine whether the maximum value is less than or equal to the minimum value; If the maximum value is less than or equal to the minimum value, that is, if there is an intersection in the suitable flow velocity range, then the fourth constraint is satisfied.

[0022] This invention introduces an improved parallel batch processing algorithm, which comprehensively sets constraints on lifting mass, water tank buffer volume threshold, ecological compatibility, and appropriate flow velocity intersection. This effectively avoids the risk of interspecies predation during cross-species transport, while also ensuring the safe lifting load of the UAV and the fluid buffer redundancy space when the water bag touches the water at the physical level.

[0023] The batching algorithm process also includes: When the number of batches already created reaches the maximum limit for the total number of batches that can be created and no more new batches can be created, the batch division module performs a judgment on the current fish characteristic record according to the following logic: Check whether the fish characteristic record meets the first quality constraint and the second volume constraint of any existing batch. If none of the existing batches meet the constraints, the record is assigned to the warning batch. It is diverted to the backup holding box and a warning is issued. The warning batch directly corresponds to the backup holding box as an independent abnormal handling channel. Its creation and counting are independent of the aforementioned regular transport batches and do not occupy the total number of created batches of up to 3. For candidate batches that meet the first quality constraint and the second volume constraint, further check whether they meet the third eco-compatibility constraint. If all candidate batches do not meet the constraint, the record is classified as a warning batch. If there are candidate batches that meet the first quality constraint, the second volume constraint, and the third ecocompatibility constraint, but are excluded from the batch only because they do not meet the fourth velocity intersection constraint, then a downgrade process is triggered. The downgrade process specifically includes: The candidate batches that meet the first three constraints are selected as fusion candidate batches. The average estimated quality of the fish already contained in them is calculated. The current fish feature records are forcibly assigned to the fusion candidate batch with the smallest average estimated quality and updated as the fusion batch.

[0024] In continuous high-density multi-species fish passage scenarios, when a new fish species cannot be included in the batch only due to flow velocity intersection constraints, this invention forces it to be assigned to the fusion candidate batch with the lowest average estimated quality. This effectively prevents scheduling deadlock caused by resource exhaustion and improves the operational robustness in dealing with complex fish passage scenarios.

[0025] The process of obtaining the target flow rate and safe overload lower limit corresponding to a batch specifically includes: When the addition of a fish feature record will cause the current batch to fail to meet the first or second constraint, the batch is marked as closed. After the batch is closed, threshold extraction is performed on the closed batch to obtain the target flow rate and the batch safe overload lower limit. For a fusion batch that has undergone degradation processing, the mean of the appropriate flow rate range for each record within it is extracted, and a weighted average is performed with the estimated quality as the weight to obtain the target flow rate of the fusion batch. After the downgrade process is performed, the resulting target flow rate is a weighted result based on the overall optimal principle of the batch, which may deviate from the suitable flow rate range for individual newly added fish. The critical acceleration of fish impact damage recorded in all batches is multiplied by a preset reduction factor of 0.85, and the minimum value is taken. Then, it is multiplied by a preset safety redundancy factor of 0.9 to calculate the lower limit of safe overload for the batch. For normal batches that have not undergone downgrade processing, query the critical acceleration of fish impact damage in all fish characteristic records within the normal batch, take the minimum value and multiply it by the preset safety redundancy coefficient of 0.9 as the lower limit of safe overload for the normal batch, calculate the midpoint of the intersection interval of suitable swimming velocity for the normal batch, and use it as the target velocity for the normal batch.

[0026] This invention is based on closing all fish feature records within a batch and dynamically extracting the target flow velocity and safe overload lower limit of the batch, so that subsequent physical environment control parameters and buffering and load reduction strategies can be adapted to the overall shock resistance limit and swimming ability of the mixed fish population in the current batch.

[0027] The process of generating a circumferential swirling water flow by controlling the jet nozzle array on the side wall of the temporary holding tank according to the target flow velocity specifically includes: After the batch is closed, a switching command is sent to the electric gate of the diversion pool to open the flow channel of the target holding tank corresponding to the batch and close the other channels, so that the water carries the fish of the same batch into the designated holding tank. The target driving frequency of the water pump is calculated by substituting the target flow velocity into the preset linear conversion formula. The target driving frequency that exceeds the safe operating range of the water pump is cut off, and a safe driving frequency is output. The frequency converter is controlled to drive the water pump so that the tangential jet nozzle array on the side wall of the temporary holding tank generates a circumferential swirling water flow with a linear velocity equal to the target flow velocity.

[0028] After a batch is closed, a switching command is sent to the electric gate of the diversion pool to open the flow channel of the target holding tank corresponding to the batch that should be closed. At the same time, the drone corresponding to the holding tank is summoned to fly and hover directly above it. The fish are transported in a hydraulic pipeline mode with water flowing throughout, and the batch diversion is achieved by guiding the water flow. The end of the fish transport pipeline is connected to a buffer diversion pool. The diversion pool is equipped with multiple water outlet slides. Each slide corresponds to an independent holding tank and is equipped with a controllable electric water diversion gate. During the fish transport stage, after the fish characteristics are recorded and classified, the electric gate of the diversion pool is switched according to the classification results. The flow channel corresponding to the target holding tank is opened and the other channels are closed. The water carries the fish of the same batch along the slide and flows into the designated holding tank by gravity. The mathematical expression for the target driving frequency is: ; in, The target driving frequency, in Hertz. and In this embodiment, the constant obtained by calibration in the test water tank using this device is... Hertz per second per meter hertz; The calibration process specifically includes: installing the same model of frequency converter, water pump and tangential jet nozzle array as in this embodiment in the test water tank, setting the output frequency of the frequency converter to 10Hz, 15Hz, 20Hz, 25Hz, 30Hz, 35Hz, 40Hz, 45Hz and 50Hz in sequence, measuring the water flow velocity at 0.3 meters downstream of the nozzle outlet using an electromagnetic flow meter, repeating the measurement 5 times at each frequency point and taking the average value. The experimental data are shown in Table 2. The above data were subjected to least squares linear fitting, and the relationship was converted into a frequency-to-flow velocity expression. To simplify the online calculation in the control system and retain the adjustment margin, the coefficients were rounded to α=8.0 Hz·s / m and β=2.5 Hz. If the target drive frequency exceeds the minimum safe operating frequency of 5 Hz and the maximum rated frequency of 50 Hz for the water pump, a cutoff operation is performed. The safe drive frequency is output to the frequency converter via digital-to-analog conversion. The frequency converter drives the water pump to generate a circumferential swirling water flow with a stable linear velocity of the target flow rate within the enclosure of the sidewall tangential jet nozzle array. The reason for performing this operation is that fish have a natural tendency to follow currents. When the incoming flow velocity is within their suitable range, the fish will spontaneously maintain a uniform hovering posture with their heads against the current and their bodies parallel to the direction of the water flow. As the hemispherical lifting device at the bottom of the enclosure smoothly lifts the fish to the water's surface and docks with the spherical water tank lowered by the drone, the full-depth circumferential water flow maintained by the tangential jets on the sidewalls keeps the fish at a linear velocity. The fish are sucked into the water tank in a relatively uniform longitudinal posture. This longitudinal trend ensures that during the subsequent drone drop phase, the fish enter the water in a mainstream posture where the longitudinal axis of the head or body cuts into the water surface at an acute angle. This can significantly reduce the impact of entering the water. This guiding posture can also greatly reduce the risk of lateral swimming fish being sheared and injured during the mechanical closing process.

[0029] This invention generates a circumferential swirling water flow with the same velocity as the target flow in a batch by controlling the tangential jet nozzle array on the side wall of the holding tank. By utilizing the fish's instinct to follow the flow, the fish spontaneously maintain a uniform hovering posture with their heads facing the current. This not only reduces the risk of mechanical shearing injury during the upward lifting and closing process, but also guides the fish to be sucked into the water tank with a mainstream posture that cuts into the water surface at an acute angle along the longitudinal axis. Through hydrodynamic mechanisms, the impact of subsequent airdrops into the water is effectively weakened.

[0030] The process of performing airdrop optimization calculations specifically includes: The drone descends into the spherical water tank to complete the docking. It uses the negative pressure inside the water tank to suck the fish along with the life-sustaining water into the water tank. After the water tank is full, the bottom door is closed. The drone carries the spherical water tank full of fish and water mixture from the temporary holding tank and flies to the designated release point upstream of the reservoir. It then performs airdrop optimization calculations under multi-dimensional fluid dynamics constraints. The airspeed, ground speed, and heading of the UAV are continuously collected at a frequency of 10Hz. These parameters are acquired by the UAV's flight control system, which integrates an airspeed tube, inertial measurement unit, GNSS receiver, and magnetometer. This system can calculate and output the flight parameters in real time. The environmental wind speed vector is obtained from the airspeed, ground speed, and heading, forming an environmental wind speed sequence. The component of the environmental wind speed vector along the current UAV cruising flight direction is extracted from the environmental wind speed sequence. Taking the headwind direction as positive, a 2-second moving average filter is applied to the environmental wind speed sequence to obtain the horizontal wind speed. The unit is meters per second; Get current horizontal wind speed The lower limit of safe overload corresponding to the current batch Based on the airdrop height, the freefall time, vertical water-cutting speed, and combined cutting angle are calculated. The bottom hatch of the main water tank opens completely at a smooth rate. The water containing the fish remains in a concentrated state under the action of gravity and water surface tension and slides down as a whole, forming a fish-water mixture with a continuous shape and fish inside. In this embodiment, the fish-water mixture is used as a water bag. Construct an airdrop kinematic sequence and calculate the free fall time of the water bag based on the drop height; The mathematical expression for the free fall time from when the water bag leaves the drone until it touches the surface of the reservoir is: ; in, The time taken for free fall, in seconds. This refers to the airdrop height, in meters. The acceleration due to gravity is taken as 9.80 m / s². Calculate the vertical water-cutting velocity based on the free fall time; The vertical shear velocity at the instant the water bag hits the water is calculated. The mathematical expression for the vertical shear velocity is: ; in, The vertical water cutting velocity is expressed in meters per second. The ground speed of the drone's horizontal flight is used as the initial horizontal absolute speed at the moment of water bag release. The horizontal water cutting speed at the moment of water impact is calculated by combining the horizontal wind speed and the preset wind deflection response coefficient. The combined angle of entry is calculated based on the vertical and horizontal water-cutting velocities.

[0031] The initial horizontal absolute velocity at the moment of water bag release is equal to the ground speed of the drone in horizontal flight. ; Considering the drift effect caused by environmental wind resistance during the descent, horizontal wind speed is introduced. Perform wind deflection compensation, wind deflection response coefficient It is a lumped parameter that reflects the attenuation efficiency of the horizontal velocity of the water bag per unit wind speed. It is related to the aerodynamic shape, mass and fall time of the water bag. Its calibration method is as follows: In a known stable horizontal wind field, a UAV hovers at a specific height and releases the water bag. The horizontal drift distance and fall time of the water bag's contact point are measured by high-speed camera. The actual horizontal velocity is calculated in reverse. Then, it is calculated based on the wind speed and the attenuation of the horizontal velocity. The experimental data are shown in Table 3. After conducting multiple experiments at different wind speeds, the statistical mean was taken as a preset constant. In this embodiment... The curing value is 0.1; The mathematical expression for the horizontal water-cutting velocity at the instant of impact is: ; in, The horizontal water-cutting velocity at the moment of impact with the water. The ground speed of the drone during horizontal flight. The preset wind deflection response coefficient, Horizontal wind speed; The mathematical expression for the composite angle of entry is: ; in, To synthesize the angle of entry, The vertical water cutting velocity, The horizontal cutting speed at the moment of impact with the water, expressed in radians converted to degrees.

[0032] This invention constructs an airdrop kinematic sequence and introduces a wind deflection response coefficient based on the environmental wind speed vector in the horizontal water-cutting velocity calculation. It can solve the actual displacement velocity and the combined angle of entry under the influence of actual environmental wind resistance drift, and provide reliable initial boundary conditions for accurately evaluating the dynamic evolution of the water bag at the moment of water drop.

[0033] The process of performing airdrop optimization calculations also includes: A dynamic model of water impact is constructed, and the radius of the equivalent sphere is inversely calculated based on the volume of the spherical water tank. The moment when the bottom of the water tank touches the water is set as the starting point of time iteration, and the initial time, initial immersion depth, initial horizontal absolute velocity component and initial vertical absolute velocity component are set as the initial integral values. A dynamic model of water ingress impact is constructed. Due to the extreme complexity of the water ingress process, this model is simplified based on the dominant effect. The impact force mainly comes from the inertial force overcome by the water body in front of the water ingress. Its magnitude is proportional to the product of the instantaneous cross-sectional area and the square of the resultant velocity. Viscosity and secondary impact are ignored in order to obtain an approximate impact overload that can be calculated online. To achieve rapid online calculation, the quasi-steady assumption, widely used in engineering, is adopted. The instantaneous resistance during the water ingress process is equivalent to the standard fluid resistance form, and complex unsteady and multiphase flow effects are all incorporated into the coefficients. middle; The transient cross-sectional area of ​​the water-filling process is approximated as the cross-sectional area of ​​a spherical cap. Specifically, based on the volume of the spherical water tank... The mathematical expression for the radius of its equivalent sphere is derived as follows: ; in, Let the radius be the equivalent sphere. It is a spherical water tank volume; Let the moment the bottom of the water bag touches the water be the starting point for time iteration, and set the initial value for integration, including the initial time. Initial immersion depth Initial horizontal absolute velocity component Initial vertical absolute velocity component ; In each time step of the iterative calculation, the immersion depth is calculated based on the current vertical velocity integral, and the instantaneous cross-sectional area is dynamically calculated. At each time step of the iterative calculation, based on the current vertical velocity Integral calculation of immersion depth And dynamically calculate the instantaneous cross-sectional area. This model ensures that the impact force increases continuously from zero, truly reflecting the transient impact process; The instantaneous dynamic impact acceleration overload is calculated. The instantaneous dynamic impact acceleration overload is proportional to the square of the instantaneous cross-sectional area and the resultant velocity. The calculation incorporates the water momentum correction coefficient, the water density of the reservoir area, and the total mass of the fish-water mixture in the current batch. According to the principle of the rate of change of momentum, the instantaneous dynamic impact acceleration overload experienced by the water body at the moment of entry into the water is proportional to the sum of the squares of the resultant velocity and the instantaneous cross-sectional area. The mathematical expression for the instantaneous dynamic impact acceleration overload is: ; in, This refers to the instantaneous dynamic impact acceleration overload, measured in meters per second². This is a preset correction factor for the water momentum of the water inlet. The density of the reservoir water is fixed at 1000 kg / m³. This refers to the total mass of the fish and water mixture in the current batch; The water ingress momentum correction coefficient is used to correct the error introduced when simplifying the complex water ingress impact process into a momentum change rate model. Through offline water tank impact experiments, the measured peak acceleration is compared with the model prediction for reverse calibration. In this embodiment, ; The specific experimental procedure included: loading 100 kg of water into a spherical water tank of the same specifications as in this embodiment, with a volume of 0.10 cubic meters, and performing 10 standard releases from a height of 20 meters. The peak acceleration value at the moment of water entry was recorded using a piezoresistive accelerometer built into the center of the water tank, with a sampling rate of 10 kHz and a range of ±200 g. Meanwhile, the numerical integral model of this scheme was run with Cm=1.0 as the initial assumption, and the same release height and water bag mass were input to calculate the predicted peak acceleration value. The experimental data are shown in Table 4. As shown in Table 4, the mean ratio of the 10 experimental groups is 1.036 and the standard deviation is 0.016, indicating that the model has good prediction accuracy when Cm=1.0 and no additional coefficient correction is needed. Therefore, Cm is fixed at 1.0. The direction of the drag acceleration is always opposite to the instantaneous velocity vector. Its horizontal and vertical components are distributed in real time according to the proportion of the current velocity component. The simultaneous differential equations of the horizontal and vertical velocity components are simulated and iterated using a first-order Runge-Kutta integrator with a preset time step until the instantaneous vertical velocity decays to zero. The immersion depth, instantaneous cross-sectional area, and instantaneous velocity components are updated synchronously within each time step. The absolute peak acceleration on the time axis of the entire water entry process is extracted and converted into a dimensionless overload factor. The direction of the drag acceleration is always opposite to the instantaneous velocity vector, and its horizontal and vertical components are distributed in real time according to the proportion of the current velocity components. The iterative equations for the current instantaneous horizontal and vertical velocities after deceleration by fluid resistance, after discretization, are expressed as follows: ; ; in, The horizontal velocity component at the next moment. The horizontal velocity component at the current moment, For instantaneous dynamic impact acceleration overload, The vertical velocity component at the current moment. For the integration time step, The vertical velocity component at a given moment; It is the acceleration due to gravity; Using a first-order Runge-Kutta integrator, with a preset microsecond-level time step... Seconds, in the time domain The preceding simulation iterations are performed on the above system of simultaneous differential equations, where the integration termination time is... The criterion is instantaneous vertical velocity. The moment the decay reaches zero corresponds to the end of the deceleration process of the water entering the water bag; The integrator synchronously updates the immersion depth, instantaneous cross-sectional area, and instantaneous velocity components at each time step, and finally searches in parallel to extract the absolute peak acceleration over the entire water entry process time axis. Convert it to a dimensionless overload factor , for With gravitational acceleration The ratio; A triple security check is performed, consisting of a first check, a second check, and a third check. The first criterion is that the overload multiple is less than or equal to the lower limit of safe overload. This criterion limits the upper limit of the combined water-cutting speed determined by the airdrop height and the horizontal flight ground speed, thus defining that the airdrop height cannot be too large. It also requires that the horizontal water-cutting speed must be positive, excluding the state of the drone hovering at zero horizontal speed for airdrop, and prohibiting 90-degree vertical water drop. The second judgment limits the synthesis angle. It is within the preset safety range; that is ,when Below 30°C, the water tank experiences violent tail-slapping and generates a lethal secondary load peak; above 30°C, the load peaks. When the vertical component was too large, the initial overload amplification became uncontrollable. After repeated entry angle tests at different heights (10 meters, 15 meters, 20 meters, and 25 meters), the critical angle fluctuation of the water bladder ricochet and secondary load was less than ± , interval This range is used as a general hard constraint because it can always cover all safe operating conditions. The experimental data of the angle of entry at different heights are summarized in Table 5. Experiments show that within a release height range of 10 to 25 meters, the angle of attack is lower than... ricochet or near-ricochet occurs at all times, with an entry angle of... to No ricochets occurred, and the angle of entry reached [value missing]. Although no ricochets occurred, the initial peak load had significantly increased. The boundary value of the safe entry angle fluctuated by less than ±2° at different altitudes. Therefore, the anti-ricochet entry angle range was uniquely determined as [missing value]. ; The third criterion is that the horizontal offset distance between the main water jet and the jet contact point is less than the offset tolerance value, and the initial horizontal velocities of the main water jet and the jet before release are the same. Because the main water jet is affected by the wind resistance of the environment and will be horizontally deflected, while the wind deflection of the high-speed pilot jet is negligible, the contact points of the two will be horizontally offset. Because the jet triggering has a delay time, the total horizontal displacement of the jet contact point is The main water tank takes a long time to discharge water. Its actual horizontal water-cutting velocity after being affected by wind deflection is Its actual displacement is Therefore, the horizontal offset distance between the two contact points is given by the formula Real-time calculations, in meters; This offset distance Must be smaller than the current Corresponding offset tolerance value Offset tolerance value It refers to a specific In combination, the buffer zone created by the jet allows the peak acceleration of the water encapsulated to meet the lower limit of safe overload. Maximum permissible horizontal offset distance; This tolerance value is obtained by looking up a preset two-dimensional interpolation table for jet offset tolerance. This table takes horizontal flight ground speed and drop altitude as input and offset tolerance as output. The calibration method for this table is as follows: at different... and Under the combined operating conditions, a standard water jet equipped with an accelerometer is released above the water tank, and the jet is simultaneously activated. High-speed video is used to observe and record the horizontal relative position of the buffer zone formed by the water jet and the jet, determining the minimum overload experienced by the water jet. The horizontal distance between the maximum water jet contact point and the jet contact point, where the overload value is below the safety benchmark, is defined as the offset tolerance under this operating condition. The safety benchmark value is taken as the lower limit of safe overload for the batch. 0.8 times, to leave an extra safety margin during calibration, ultimately forming arrive A two-dimensional data table; The typical experimental process for each data point in this two-dimensional interpolation table is based on the working condition ( =5m / s, Taking a standard water tank (10m) as an example, a flight simulation rail was erected 10 meters above the test water tank. The release mechanism dragged a standard water tank (total mass 100 kg) at a horizontal speed of 5 m / s. Simultaneously, the jet was activated at the calculated launch delay. By precisely adjusting the jet launch delay time, different horizontal offsets were artificially created between the water tank contact point and the jet contact point, successively set at 0m, 0.1m, 0.2m, 0.3m, 0.5m, 0.7m, and 0.9m. Each offset was repeated 5 times. The experiment was conducted using a built-in acceleration sensor. The instrument records the peak value of water overload in the water jacket and calculates the load reduction rate. The load reduction rate is the ratio of the difference between the peak value of overload without jet buffer and the peak value of overload with jet buffer to the peak value of overload without jet buffer. When the load reduction rate drops to 65%, the corresponding offset is recorded as the offset tolerance of the current working condition. Under this working condition, it was found that the load reduction rate is 65% when the offset is 0.75m, so d0(5,10)=0.75m. After rounding, it is recorded as 0.75m and written into the interpolation table. The calibration method for other working conditions is the same, and finally a two-dimensional data table is formed. Table 6 shows some examples of two-dimensional interpolation tables for jet offset tolerance. For values ​​between Combination, its offset tolerance Obtained through bilinear interpolation, if If the solution is not feasible, then the solution set is not feasible.

[0034] This invention defines the physical and dynamic safety boundary of the airdrop process by constructing and solving an online water entry impact dynamics model based on the rate of change of momentum, and combining it with a triple safety judgment mechanism that includes overload multiple, anti-bounce safety entry angle range, and jet horizontal deviation tolerance. This eliminates the violent tail slap and secondary load peak caused by improper water cutting posture, and protects the internal organs of the fish from mechanical tissue rupture.

[0035] The process of performing airdrop optimization calculations also includes: Within the feasible solution space that satisfies the triple safety criteria, an energy consumption evaluation function is constructed with the goal of minimizing the UAV's flight energy consumption to perform optimization search, and the optimal horizontal flight speed and optimal drop height are output. The mathematical expression for the energy consumption evaluation function is: ; in, This is the energy consumption evaluation value. and To determine the energy consumption weighting constant based on the UAV power curve, this embodiment takes... , ; The search algorithm employs a sequential quadratic programming method with boundary constraints. Using the UAV's current horizontal ground speed and its current flight altitude relative to the reservoir surface as initial values, it rapidly converges to a unique optimal solution within the feasible region, outputting the optimal horizontal flight speed. and optimal airdrop height ; If, under the current environmental parameters, the triple safety decision results in an empty solution space, the second decision is abandoned, and a mandatory throwing height is set. To preset a lower limit for safety, this embodiment sets it at 5 meters, and minimizes the overload multiple while strictly maintaining the third criterion. Re-search horizontal speed for a single target ; If the minimum overload factor is obtained Exceeding the batch safety overload limit This indicates that the current wind conditions and hydrodynamic boundaries can no longer guarantee the survival of the fish, and it is determined that airdrop is not feasible at present. The bottom hatch of the fish transport water tank is then locked. If the smallest dimensionless overload factor is obtained Not exceeding the batch safety overload limit The horizontal speed obtained from the re-search and the preset safety lower limit of the throwing height The data was sent to the drone control system.

[0036] The process of calculating the jet emission delay time specifically includes: Calculate the free fall time of the main water tank based on the optimal throwing height; Drone arrives and maintain After steady-state cruise, based on the current Calculate the time taken for the main water tank to fall freely. ; Calculate the jet flight time based on the jet exit velocity and the optimal throwing height; Jet flight time The jet exit velocity is 35 m / s and Calculations show that ; The difference between the free fall time of the main water tank and the jet flight time is used as the jet launch delay time.

[0037] Jet launch delay The objective is to ensure that the high-speed jet tip and the bottom of the water jacket arrive at the theoretical water entry point of the water jacket simultaneously, thereby effectively buffering the water jacket. Therefore... The calculations are based directly on the flight times of each of them. for and The difference; The UAV control system receives the optimal horizontal flight speed and optimal airdrop height Upon reaching the launch coordinates, the main water tank's bottom hatch is controlled to open smoothly and fully. The water containing the fish, under the influence of gravity and surface tension, remains concentrated and slides down as a whole, forming a water jacket. The main water jacket then begins freefall, and after a jet launch delay... Then, a signal is sent to the electromagnetic servo valve at the bottom of the annular pure water jacket, and the jet pressurization pump sprays the pre-filled pure water in the jacket vertically downwards at an outlet velocity of 35 m / s. The jet will carry the same initial horizontal velocity as the water tank at the initial moment. After the jet is launched, the water replenishment pump replenishes water from the main water tank to the jacket. Under ideal, windless conditions, the jet will precisely hit the contact point of the main water jet; however, in actual operation, due to wind deflection, the jet will be positioned further away from the actual landing point of the water jet. Upon contact with the water, a high-speed water curtain is formed, obliquely cutting into the water surface. By displacing the water, a low-pressure, low-density buffer zone is created along the predetermined trajectory of the water jet. The offset distance caused by this wind deflection... It has been strictly limited to the effective tolerance range of the buffer zone during the airdrop optimization phase; The main water bag is used to synthesize the angle of entry. Upon entering the buffer zone, the high-speed water curtain creates a low-pressure buffer, delaying and reducing the peak inrush load. The impact overload experienced by the fish is controlled within the batch's safe overload limit. Within this range, live fish can be safely and safely introduced into the water.

[0038] This invention overcomes the technical limitation of monocular top-down vision in perceiving fish thickness by constructing a fish feature extraction model that includes a main feature branch network and a feature pyramid fusion network. Combined with the body size ratio correction factor in the fish parameter database, it achieves high-precision online estimation of the individual category, volume and mass of mixed fish groups. On this basis, a parallel batch processing algorithm based on multidimensional physical and biological constraints is introduced. Taking into account the lifting mass limit of the UAV, the buffer volume threshold of the water tank, the ecological compatibility of fish species and the intersection range of suitable flow velocity, the scattered and disordered fish groups are diverted and grouped. With the help of degradation fusion and independent early warning processing mechanism, the system deadlock under high-density fish passage conditions is effectively avoided. This invention introduces an active attitude control mechanism based on hydrodynamics in the temporary holding and lifting and transportation of fish. The diversion control module adjusts the tangential jet nozzle array on the side wall of the holding tank according to the target flow velocity calculated for each batch, generating a circumferential swirling water flow that matches the swimming ability of the fish in that batch. Utilizing the fish's natural instinct to follow the flow, this mechanism forces the mixed fish to spontaneously adjust and maintain a uniform hovering posture that is parallel to the current. This not only significantly reduces the risk of mechanical shearing injury during the longitudinal lifting and closing process, but also ensures that the fish can cut into the water surface at an ideal acute angle along the longitudinal axis when they are sucked into the fish transport tank and finally airdropped, thus weakening the impact at the moment of contact with the water from the attitude source. The airdrop optimization module of this invention comprehensively considers wind deflection compensation, water bag falling kinematics, and instantaneous dynamic impact overload. Under the triple safety judgment of satisfying the lower limit of safe overload, the no-bounce entry angle range, and the effective buffer offset tolerance of the jet, it calculates the optimal horizontal flight speed and drop height that minimize energy consumption. At the moment of airdrop execution, a high-speed jet leader is launched downward according to the calculated delay time. The high-speed water curtain penetrates and displaces the water in advance, actively constructing a low-pressure, low-density fluid buffer zone on the predetermined water contact trajectory of the main water bag. The main water bag then falls into the buffer zone, and its water entry kinetic energy is smoothly released. The peak load is significantly reduced, ensuring that the transient impact on the fish is strictly controlled within the critical value of mechanical damage. Thus, while overcoming the limitations of complex terrain, it achieves a high survival rate and safe transfer of fish over high-head dams.

[0039] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A fish-passing system for a reservoir dam using a cross-river cableway for fish collection and unmanned aerial vehicle (UAV) transport, characterized in that, include: The image acquisition module is used to acquire images of the fish school and obtain a fish school image matrix; The feature calculation module is used to input the fish image matrix into a pre-trained fish feature extraction model, output the category number and rotation box parameters of each fish, and calculate the estimated volume and estimated mass of a single fish by combining it with a pre-stored fish parameter database, and generate fish feature records. The batch division module is used to perform a batching algorithm on the fish feature records, divide the fish feature records into different transportation batches, and obtain the target flow rate and safe overload lower limit corresponding to the transportation batches; The diversion control module is used to control the fish to enter the corresponding holding tank according to the transport batch, and to control the jet nozzle array on the side wall of the holding tank to generate a circumferential swirling water flow according to the target flow rate. The airdrop optimization module is used to control the drone to descend the spherical fish transport water tank and dock with the temporary holding box to suck up the fish. After the drone takes off, it performs airdrop optimization calculations and outputs the optimal airdrop plan. The airdrop execution module is used to control the UAV to fly to the drop point, unfold the bottom hatch to form a water bag, calculate the jet launch delay time based on the optimal airdrop scheme, and execute the airdrop.

2. The reservoir dam fish transport system for cross-river cableway fish collection and drone fish transport as described in claim 1, characterized in that, The fish feature extraction model includes a main feature branch, a feature pyramid fusion network, a classification prediction detection head, and a regression prediction detection head; The parameters of the rotation frame are the principal axis pixel length, secondary axis pixel width, deflection angle, and center point pixel coordinates of the smallest bounding rectangle of the fish target. The process of calculating the estimated volume and estimated mass of a single fish specifically includes: Based on a preset size conversion factor, the main axis pixel length and the secondary axis pixel width are converted into physical projection length and physical projection width; The corresponding body proportion correction factor is queried from the fish parameter database according to the category number, and the product of the body proportion correction factor and the physical projection width is used as the fish body thickness. The fish body shape is abstracted as a triaxial ellipsoid, with the physical projection length as the major axis, the physical projection width and the fish body thickness as the two minor axes, and the estimated volume of a single fish is calculated. The preset average fish density is used, and the product of the estimated volume and the average fish density is used as the estimated mass of a single fish.

3. The reservoir dam fish-passing system for cross-river cableway fish collection and drone fish transport as described in claim 2, characterized in that, The batching algorithm process specifically includes: When the first fish feature record is obtained, the first batch is created and stored in that fish feature record; When a subsequent fish feature record is obtained, it is compared with the fish feature records in the existing batch to check whether the merged batch meets the batch constraints. If all constraints are met, the fish feature record is stored in the first existing batch that meets all constraints, and the cumulative mass and cumulative volume of the batch are updated. If none of the existing batches meet the constraints, a new batch is created and stored if the total number of batches has not reached the upper limit. The batch constraints include a first mass constraint, a second volume constraint, a third ecological compatibility constraint, and a fourth velocity intersection constraint.

4. The reservoir dam fish-passing system for cross-river cableway fish collection and drone fish transport as described in claim 3, characterized in that, The batching algorithm process also includes: When the number of batches already created reaches the maximum limit for the total number of batches that can be created and no new batches can be created, the batch division module performs a judgment on the current fish feature record according to the following logic: Check whether the fish characteristic record meets the first quality constraint and the second volume constraint of any existing batch. If none of the existing batches meet the constraints, then classify the record into the warning batch. For candidate batches that meet the first quality constraint and the second volume constraint, further check whether they meet the third eco-compatibility constraint. If all candidate batches do not meet the constraint, the record is classified as a warning batch. If there are candidate batches that satisfy the first quality constraint, the second volume constraint, and the third ecological compatibility constraint, but are unable to be included in the batch only because they do not meet the fourth velocity intersection constraint, then a downgrade process is triggered. The downgrade process specifically includes: The candidate batches that meet the first three constraints are selected as fusion candidate batches. The average estimated quality of the fish already contained in them is calculated. The current fish feature records are forcibly assigned to the fusion candidate batch with the smallest average estimated quality and updated as the fusion batch.

5. A reservoir dam fish-passing system for cross-river cableway fish collection and drone fish transport as described in claim 3, characterized in that, The process of obtaining the target flow rate and safe overload lower limit corresponding to the batch specifically includes: For a fusion batch that has undergone degradation processing, the average value of the suitable flow velocity range of each record within it is extracted, and a weighted average is performed with the estimated quality as the weight to obtain the target flow velocity of the fusion batch. The minimum value of the critical acceleration of fish impact damage recorded in the fusion batch is obtained by multiplying each of the recorded critical accelerations by a preset reduction factor and then multiplying it by a preset safety redundancy factor. The lower limit of the safe overload of the fusion batch is then calculated. For normal batches that have not undergone degradation processing, query the critical acceleration of fish impact damage in all fish characteristic records within the normal batch, take the minimum value and multiply it by the preset safety redundancy coefficient as the lower limit of safe overload for the normal batch, calculate the midpoint of the intersection interval of suitable swimming velocity for the normal batch, and use it as the target velocity for the normal batch.

6. A reservoir dam fish-passing system for cross-river cableway fish collection and drone fish transport as described in claim 5, characterized in that, The process of controlling the jet nozzle array on the side wall of the temporary holding tank to generate a circumferential swirling water flow according to the target flow velocity specifically includes: Send a switching command to the electric gate of the diversion pool to open the corresponding target holding tank channel and close the other channels, so that the water carries the same batch of fish into the designated holding tank. The target flow velocity is substituted into a preset linear conversion formula to calculate the target driving frequency of the water pump. The safe driving frequency is output, and the frequency converter is controlled to drive the water pump so that the tangential jet nozzle array on the side wall of the temporary holding tank generates a circumferential swirling water flow with a linear velocity equal to the target flow velocity.

7. A reservoir dam fish-passing system for cross-river cableway fish collection and drone fish transport as described in claim 6, characterized in that, The process of performing the airdrop optimization calculation specifically includes: Construct an airdrop kinematic sequence and calculate the free fall time of the water bag based on the drop height; Calculate the vertical water-cutting velocity based on the free fall time; The ground speed of the drone's horizontal flight is used as the initial horizontal absolute speed at the moment of water bag release. The horizontal water cutting speed at the moment of water impact is calculated by combining the horizontal wind speed and the preset wind deflection response coefficient. The combined cutting angle is calculated based on the vertical cutting velocity and the horizontal cutting velocity.

8. A reservoir dam fish-passing system for cross-river cableway fish collection and drone fish transport as described in claim 7, characterized in that, The process of performing the airdrop optimization calculation also includes: A dynamic model of water impact is constructed, and the radius of the equivalent sphere is inversely calculated based on the volume of the spherical water tank. The moment when the bottom of the water tank touches the water is set as the starting point of time iteration, and the initial time, initial immersion depth, initial horizontal absolute velocity component and initial vertical absolute velocity component are set as the initial integral values. In each time step of the iterative calculation, the immersion depth is calculated based on the current vertical velocity integral, and the instantaneous cross-sectional area is dynamically calculated. The instantaneous dynamic impact acceleration overload is calculated. The instantaneous dynamic impact acceleration overload is proportional to the square of the instantaneous cross-sectional area and the resultant velocity. The calculation incorporates the water momentum correction coefficient, the water density of the reservoir area, and the total mass of the fish-water mixture in the current batch. The direction of the drag acceleration is always opposite to the instantaneous velocity vector. Its horizontal and vertical components are distributed in real time according to the proportion of the current velocity component. The simultaneous differential equations of the horizontal and vertical velocity components are simulated and iterated using a first-order Runge-Kutta integrator with a preset time step until the instantaneous vertical velocity decays to zero. The immersion depth, instantaneous cross-sectional area, and instantaneous velocity components are updated synchronously within each time step. The absolute peak acceleration on the time axis of the entire water entry process is extracted and converted into a dimensionless overload factor. Perform a triple security check, which includes a first check, a second check, and a third check; The first determination is that the overload multiple is less than or equal to the lower limit of safe overload; The second determination limits the synthetic cutting angle to within a preset safety range; The third determination is that the horizontal offset distance between the main water bag and the jet contact point is less than the offset tolerance value.

9. A reservoir dam fish-passing system for cross-river cableway fish collection and drone fish transport as described in claim 8, characterized in that, The process of performing the airdrop optimization calculation also includes: Within the feasible solution space that satisfies the triple safety criteria, an energy consumption evaluation function is constructed with the goal of minimizing the UAV's flight energy consumption to perform optimization search, and the optimal horizontal flight speed and optimal drop height are output. If the triple safety determination under the current environmental parameters results in an empty set of feasible solutions, the second determination is abandoned, the throwing height is forcibly set to the preset safety limit lower limit, and the horizontal speed of the single target is re-searched with the minimum overload multiple under the premise of strictly maintaining the third determination. If the obtained minimum overload multiple exceeds the batch safe overload lower limit, it is determined that the current airdrop is not feasible, and the bottom hatch of the fish transport water tank is locked; If the obtained minimum dimensionless overload multiple does not exceed the batch safety overload lower limit, the horizontal speed and the preset safety limit lower limit of the drop height obtained by the re-search will be sent to the UAV control system.

10. A reservoir dam fish-passing system for cross-river cableway fish collection and drone fish transport as described in claim 9, characterized in that, The process of calculating the jet emission delay time specifically includes: Calculate the free fall time of the main water tank based on the optimal throwing height; The jet flight time is calculated based on the jet exit velocity and the optimal throwing height. The difference between the free fall time of the main water bag and the flight time of the jet is taken as the jet launch delay time.

Images