Method for evaluating breeding capacity of submerged fishing ground

Abstract

The invention discloses a method for evaluating the culture capacity of a submerged fishing ground. The method comprises the following steps of: S1, calibrating a reference of the number of pixel points of cultured fishes; s2, a fish quantity data layering measurement step; s3, training a fish number estimation model; s4, estimating the total fish quantity; according to the method, the total fish quantity in the target submerging type fishery culture net cage can be estimated by using the omnidirectional sonar in a simple and rapid data acquisition mode, so that the culture capacity of the target submerging type fishery culture net cage is evaluated, and the method has the advantage that the estimation result of the total fish quantity is accurate; moreover, all data required by the implementation method can be obtained only by using the omnidirectional sonar according to the step S2, so that the method has the advantages of simplicity, convenience, high efficiency and rapidness, and can solve the problems of high data monitoring and acquisition requirements, high monitoring equipment price, complex operation and relatively high cost of the target submerged fishing ground aquaculture net cage in the prior art.

Description

Technical Field

[0001] The invention relates to fishery resource estimation and measurement, in particular to a method for evaluating the aquaculture capacity of a submerged and floating fishery. Background Art

[0002] At present, more than 10 countries in the world have developed deep-sea fish farm aquaculture equipment, and have successively developed more than a dozen deep-sea fish farm equipment with excellent performance, which are widely used in Norway, Finland, Sweden, the United States, Chile, Japan, Australia, Canada, the United Kingdom, Denmark and other countries. Shandong, Guangdong, Fujian, Zhejiang and other places in China have successively landed large-scale aquaculture fish farms and started a series of aquaculture verifications. my country has a vast space in the sea area with a depth of more than 25m, and the development of large-scale aquaculture fish farms has broad prospects.

[0003] With the release of the national strategic document on the active development of the marine economy, the aquaculture industry is facing the problem of leaving the harbor and going to the deep sea. The requirements for the safety, automation and intelligence of equipment have been raised to an unprecedented level, especially in the South China Sea, where there are a large number of floating HDPE cages. This type of cage has weak wind and wave resistance, requires a large number of manual labor, and has low production efficiency. The large-scale deep-sea aquaculture equipment of submerged and floating fisheries in the open waters of the South China Sea can withstand a 17-level typhoon, achieving the level of deep-sea aquaculture equipment to withstand super typhoons, and can effectively solve the shortcomings of traditional HDPE cage aquaculture with poor wind and wave resistance. By equipping with water quality monitoring, fish school monitoring and other system equipment, the aquaculture capacity of submerged and floating fisheries can be intelligently monitored, reducing manual work, improving aquaculture efficiency, and realizing unmanned aquaculture.

[0004] The aquaculture capacity information data (such as quantity, distribution, etc.) of submerged and floating fisheries are difficult to count in detail, and it is not easy to form a specific, systematic and scientific aquaculture and management plan, which cannot provide sufficient technical support for the development of deep-sea aquaculture. With the rapid development of deep-sea aquaculture, the aquaculture water body and scale are constantly increasing, and the accurate assessment of aquaculture capacity has become more and more difficult. Traditional manual statistical methods often cannot meet the accuracy requirements. Secondly, the aquaculture water body of submerged and floating fisheries is large, and the number of farmed fish is large, resulting in high manual statistical costs, and there may be omissions or misreporting. In addition, factors such as weather and water temperature in the marine environment will also affect the assessment of aquaculture capacity, further increasing the complexity of statistics. Therefore, accurately assessing the aquaculture capacity of submerged and floating fisheries is an important issue that needs to be urgently solved in the current deep-sea aquaculture industry. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for evaluating the aquaculture capacity of a submerged and floating fish farm.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0007] A method for evaluating the aquaculture capacity of a submerged and floating fish farm, comprising:

[0008] Step S1, the step of calibrating the pixel count of farmed fish, is:

[0009] The target submerged and floating fish farm aquaculture cages are calibrated using an omnidirectional sonar to obtain a single fish pixel point benchmark, which represents the average pixel point of a fish in the sonar scanning image scanned by the omnidirectional sonar;

[0010] Among them, the omnidirectional sonar can adopt a sonar that can realize 360° omnidirectional scanning at the same time, or it can adopt a sonar that realizes 360° omnidirectional scanning by driving the scanning beam to rotate in the horizontal plane. When using the latter, since the scanning beam will form a spiral scanning path, the vertical movement speed of the omnidirectional sonar cannot be too fast, so as to avoid failure to fully scan the overall space of the target submerged and floating fishery farming cages.

[0011] Step S2, fish quantity data layered measurement step, comprises:

[0012] Step S2-1, placing the omnidirectional sonar at the center of the target submerged and floating fish farm aquaculture cage, and driving the omnidirectional sonar to move vertically from the bottom of the target submerged and floating fish farm aquaculture cage to the water surface at a uniform speed, or driving the omnidirectional sonar to move vertically from the water surface to the bottom of the target submerged and floating fish farm aquaculture cage at a uniform speed to complete an overall scan;

[0013] Step S2-2: The data outputted by the omnidirectional sonar after an overall scan is played back through its supporting software to be converted into a single sonar scan video, and a video frame is intercepted from the single sonar scan video every N seconds as a sonar scan image (see Figure 4 ) to obtain M / N sonar scanning images arranged by interception time, which are recorded as the sonar scanning images of the first layer to the M / Nth layer, wherein M is the duration of a single sonar scanning video, that is, the time to complete an overall scan, N is the time required for the omnidirectional sonar to complete a 360° scan in the horizontal direction, and M / N is an integer obtained by directly removing the decimal places and rounding;

[0014] Step S2-3, binarizing the sonar scanning image and removing noise points to identify the effective pixel points representing the farmed fish in the target submerged and floating fish farm aquaculture cages in each layer of the sonar scanning image, and calculating the number of farmed fish contained in each layer to the M / Nth layer of the sonar scanning image based on the single fish pixel point benchmark calibrated in step S1, and recording it as a layered fish number sequence;

[0015] Step S3, fish population estimation model training step, comprises:

[0016] Step S3-1, executing step S2 multiple times to obtain multiple sets of training data, each set of training data includes: a layered fish quantity sequence and a corresponding actual fish quantity, wherein the actual fish quantity is the actual quantity of cultured fish in the target submerged and floating fish farm culture cages when executing step S2;

[0017] Step S3-2, taking the stratified fish quantity sequence as input and the actual fish quantity as output, the AutoGluon model is trained with the training data obtained in step S3-1 to obtain a fish quantity estimation model, thereby being able to solve the nonlinear mapping problem between the stratified fish quantity sequence, that is, the number of farmed fish from the first layer to the M / Nth layer identified in step S2 and the actual number of farmed fish in the target submerged and floating fish farm cages;

[0018] Step S4, total fish quantity estimation step, comprises:

[0019] Step S4-1, at the target evaluation time, obtain the corresponding stratified fish quantity sequence according to step S2;

[0020] Step S4-2: Input the stratified fish quantity sequence obtained in step S4-1 into the fish quantity estimation model to output an estimated value of the total fish quantity in the target submerged and floating fish farm aquaculture cages at the target assessment time.

[0021] Therefore, the present invention can estimate the total amount of fish in the target submerged and floating fish farm aquaculture cages by using omnidirectional sonar in a relatively simple and fast data collection method through steps S1 to S4, so as to evaluate the aquaculture capacity of the target submerged and floating fish farm aquaculture cages, and has the advantage of accurate total fish amount estimation result;

[0022] Moreover, since the present invention only needs to use omnidirectional sonar to obtain all the data required for the implementation method according to step S2, the requirements for scanning the omnidirectional sonar are relatively low. It only needs to scan the complete target submerged and floating fishery breeding cages by moving in the vertical direction at a uniform speed. It has the advantages of being simple, convenient, efficient and fast, and can solve the problems of high requirements for data monitoring and collection of target submerged and floating fishery breeding cages in the prior art, expensive monitoring equipment, complex operation and high cost.

[0023] Preferably, in step S1, the method for calibrating the pixel number benchmark of a single fish is:

[0024] Step S1-1, see Figure 2 , dividing the target submerged and floating fish farm aquaculture cages into a plurality of annular areas with the center position as the axis and equal radial width;

[0025] Step S1-2, placing the omnidirectional sonar at the center of the target submersible fish farm aquaculture cage, and continuously tracking and scanning the calibration fish in the target submersible fish farm aquaculture cage until the calibration fish swims through each of the annular areas;

[0026] Step S1-3, extracting the total number of pixels of the calibration fish in each annular area from all the sonar scanning images obtained by the omnidirectional sonar scanning in step S1-2;

[0027] Since the calibration fish will swim through the same annular area in multiple scans, the total number of pixels in the area is the total number of pixels scanned by the calibration fish when it swims through the same annular area each time.

[0028] Among them, the calibration fish can be a single fish and its body size is the average value of farmed fish in the target submerged and floating fishery cages, so that the total number of pixels in the area can be directly scanned; or, the calibration fish can also be multiple fish, and the total number of pixels in the area is the total number of pixels scanned when each fish swims through the same annular area the same number of times divided by the number of fish.

[0029] Step S1-4, perform mean processing on the total number of pixels in each annular area, that is: the total number of pixels in the area divided by the number of times the calibration fish is scanned in the annular area, so as to calibrate the pixel number benchmark of a single fish corresponding to each annular area; thus, using the mean processing method of each annular area, the problem that fish swimming in different environmental areas with different postures will present different pixel numbers in the sonar scanning image is solved.

[0030] In step S2-3, the method for calculating the number sequence of stratified fish is as follows:

[0031] Step S2-3-1, counting the number of effective pixels in each annular area of ​​each layer of the sonar scanning image from the identified effective pixels;

[0032] Step S2-3-2, for each layer of the sonar scanning image: calculate the number of farmed fish in each annular area based on the number of valid pixels in the same annular area and the number of pixels of a single fish, and record the total number of farmed fish in all annular areas as the number of farmed fish contained in the sonar scanning image of this layer, thereby obtaining the layered fish number sequence.

[0033] Thus, the present invention realizes the partitioning of the annular area of ​​the target submerged and floating fish farm aquaculture cages on the horizontal plane through steps S1-1 to S1-4, and calibrates the single fish pixel point benchmark corresponding to each annular area, and then calculates the number of farmed fish in each annular area one by one according to the annular area through steps S2-3-1 and S2-3-2, so as to obtain the number of farmed fish contained in the sonar scanning image of each layer, that is, the layered fish number sequence, thereby reducing the influence of the mutual occlusion of farmed fish in the target submerged and floating fish farm aquaculture cages on the scanning results of the omnidirectional sonar, and further improves the accuracy of the present invention in estimating the total amount of fish in the target submerged and floating fish farm aquaculture cages.

[0034] Preferably, in step S1-1, the radial width δd of each annular area is taken as the average length of the cultured fish in the culture cages of the target submerged and floating fish farm.

[0035] Preferred: See Figure 3 The omnidirectional sonar is loaded on the UAV through a vertical lifting drive device. The UAV hoists the omnidirectional sonar to the center of the target submerged fish farming cage, and the vertical lifting drive device drives the omnidirectional sonar to move up and down in the vertical direction.

[0036] Preferably, in step S3-1, performing step S2 multiple times satisfies the following conditions:

[0037] Condition 1: When step S2 is performed each time, the actual number of fish in the aquaculture cages of the target submerged and floating fish farm is different from each other;

[0038] Condition 2: The calibration fish used for the single fish pixel point benchmark calibration in step S1 is of the same species and has a similar growth cycle as the farmed fish in the target submerged and floating fish farm cages when executing steps S2 and S4.

[0039] Preferably: in step S3-2, the AutoGluon model is trained using the following settings:

[0040] The AutoGluon model uses a two-layer stack; the algorithm model for training is determined by AutoGluon, and all algorithms used are trained using default hyperparameters. The training time is not set, and the trained models and results are automatically saved by the algorithm. The model will autonomously try all possible algorithm sub-models that can be used to invert the quantity sequence to obtain the actual quantity, and continuously adjust the parameter scale and value of the sub-model, and finally use the sub-model with the highest score or the superposition of two sub-models to perform the task of estimating the number of fish.

[0041] The AutoGluon model selects the best_quality gear for training to obtain a better model through a longer training time. The experimental data is divided into training data and test data, and the fish quantity table data of each water layer is converted into a csv format file. Actual quantity represents the actual number of fish, which is the target data that needs to be inverted, and Depth represents the number of fish at the corresponding depth.

[0042] After the AutoGluon model uses TabularDataset() to read the training data, it sets the label label to Actual quantity and the num_gpus parameter to 1, which allows the GPU to be used to train the model, thereby speeding up the training. Then, TabularPredictor.fit() is used to start training.

[0043] The AutoGluon model can train two models: the default gear and the best quality gear, and the best one can be used.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] First, the present invention can use the omnidirectional sonar 2 to estimate the total amount of fish in the target submerged and floating fish farm aquaculture cage 1 in a relatively simple and fast data collection method through steps S1 to S4, so as to evaluate the aquaculture capacity of the target submerged and floating fish farm aquaculture cage 1, and has the advantage of accurate total fish amount estimation result;

[0046] Moreover, since the present invention only needs to use the omnidirectional sonar 2 to obtain all the data required for the implementation method according to step S2, the requirements for scanning the omnidirectional sonar 2 are relatively low. It only needs to scan the complete target submerged fishery breeding cage 1 in the vertical direction at a uniform speed. It has the advantages of being simple, convenient, efficient and fast, and can solve the problems of the prior art in which the data monitoring and collection requirements for the target submerged fishery breeding cage 1 are high, the monitoring equipment is expensive, the operation is complicated and the cost is high.

[0047] Second, the present invention realizes the partitioning of the annular area of ​​the target submerged floating fishery breeding cage 1 on the horizontal plane through steps S1-1 to S1-4, and calibrates the single fish pixel point benchmark corresponding to each annular area, and then calculates the number of farmed fish in each annular area one by one according to the annular area through steps S2-3-1 and S2-3-2, so as to obtain the number of farmed fish contained in the sonar scanning image of each layer, that is, the layered fish number sequence, thereby reducing the influence of the mutual occlusion of the farmed fish in the target submerged floating fishery breeding cage 1 on the scanning result of the omnidirectional sonar 2, and further improving the accuracy of the present invention in estimating the total amount of fish in the target submerged floating fishery breeding cage 1. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments:

[0049] Figure 1 It is a flowchart of the present invention;

[0050] Figure 2 It is a schematic diagram of dividing the annular area in step S1-1 of the present invention;

[0051] Figure 3 It is a schematic diagram of the arrangement of the omnidirectional sonar in the target submerged and floating fishery aquaculture cages in the present invention;

[0052] Figure 4 It is a schematic diagram of the sonar scanning image in the present invention. DETAILED DESCRIPTION

[0053] The present invention is described in detail below in conjunction with the embodiments and the accompanying drawings to help those skilled in the art better understand the inventive concept of the present invention. However, the protection scope of the claims of the present invention is not limited to the following embodiments. For those skilled in the art, all other embodiments obtained without creative work without departing from the inventive concept of the present invention belong to the protection scope of the present invention.

[0054] like Figures 1 to 4 As shown, the present invention discloses a method for evaluating the aquaculture capacity of a submerged and floating fish farm, comprising:

[0055] Step S1, the step of calibrating the pixel count of farmed fish, is:

[0056] The target submerged and floating fish farm aquaculture cage 1 is calibrated using the omnidirectional sonar 2 to obtain a single fish pixel point benchmark, which represents the average pixel point of a fish in the sonar scanning image scanned by the omnidirectional sonar 2;

[0057] Among them, the omnidirectional sonar 2 can adopt a sonar that can simultaneously realize 360° omnidirectional scanning, or it can adopt a sonar that realizes 360° omnidirectional scanning by driving the scanning beam to rotate in the horizontal plane. When the latter is adopted, since the scanning beam will form a spiral scanning path, the vertical movement speed of the omnidirectional sonar cannot be too fast, so as to avoid failure to fully scan the overall space of the target submerged fishery breeding cage 1.

[0058] In step S1, the method for calibrating the pixel number benchmark of a single fish is:

[0059] Step S1-1, see Figure 2 , dividing the target submerged and floating fish farm aquaculture cage 1 into a plurality of annular areas with the center position as the axis and equal radial width;

[0060] Step S1-2, placing the omnidirectional sonar 2 at the center of the target submerged fish farming cage 1, and continuously tracking and scanning the calibration fish in the target submerged fish farming cage 1 until the calibration fish swims through each of the annular areas;

[0061] Step S1-3, extracting the total number of pixels of the calibration fish in each annular area from all the sonar scanning images obtained by scanning the omnidirectional sonar 2 in step S1-2;

[0062] Since the calibration fish will swim through the same annular area in multiple scans, the total number of pixels in the area is the total number of pixels scanned by the calibration fish when it swims through the same annular area each time.

[0063] Among them, the calibration fish can be a single fish and its body size is the average value of the farmed fish in the target submerged and floating fish farm aquaculture cage 1, thereby the total number of pixels in the area can be directly scanned; or, the calibration fish can also be multiple fish, and the total number of pixels in the area is the total number of pixels scanned when each fish swims through the same annular area the same number of times divided by the number of fish.

[0064] Step S1-4, perform mean processing on the total number of pixels in each annular area, that is: the total number of pixels in the area divided by the number of times the calibration fish is scanned in the annular area, so as to calibrate the pixel number benchmark of a single fish corresponding to each annular area; thus, using the mean processing method of each annular area, the problem that fish swimming in different environmental areas with different postures will present different pixel numbers in the sonar scanning image is solved.

[0065] Preferably, in step S1 - 1 , the radial width δd of each annular area is taken as the average length of the cultured fish in the target submerged and floating fish farm culture cage 1 .

[0066] Step S2, fish quantity data layered measurement step, comprises:

[0067] Step S2-1, placing the omnidirectional sonar 2 at the center of the target submerged fish farming cage 1, and driving the omnidirectional sonar 2 to move at a uniform speed along the vertical direction from the cage bottom surface 1-1 of the target submerged fish farming cage 1 to the water surface, or driving the omnidirectional sonar 2 to move at a uniform speed along the vertical direction from the water surface to the cage bottom surface 1-1 of the target submerged fish farming cage 1, so as to complete an overall scan;

[0068] Step S2-2: The data outputted by the omnidirectional sonar 2 after an overall scan is played back through its supporting software to be converted into a single sonar scan video, and a video frame is intercepted from the single sonar scan video every N seconds as a sonar scan image (see Figure 4 ) to obtain M / N sonar scanning images arranged by interception time, which are recorded as the sonar scanning images of the first layer to the M / Nth layer, wherein M is the duration of a single sonar scanning video, that is, the time to complete an overall scan, N is the time required for the omnidirectional sonar 2 to complete a 360° scan in the horizontal direction, and M / N is an integer obtained by directly removing the decimal places and rounding;

[0069] Step S2-3, binarizing the sonar scanning image and removing noise points to identify the effective pixel points representing the farmed fish in the target submerged and floating fish farm aquaculture cage 1 in each layer of the sonar scanning image, and according to the single fish pixel point benchmark calibrated in step S1, calculate the number of farmed fish contained in each layer to the M / Nth layer of the sonar scanning image, and record it as a layered fish number sequence;

[0070] In step S2-3, the method for calculating the number sequence of stratified fish is as follows:

[0071] Step S2-3-1, counting the number of effective pixels in each annular area of ​​each layer of the sonar scanning image from the identified effective pixels;

[0072] Step S2-3-2, for each layer of the sonar scanning image: calculate the number of farmed fish in each annular area based on the number of valid pixels in the same annular area and the number of pixels of a single fish, and record the total number of farmed fish in all annular areas as the number of farmed fish contained in the sonar scanning image of this layer, thereby obtaining the layered fish number sequence.

[0073] Thus, the present invention realizes the partitioning of the annular area of ​​the target submerged and floating fish farming cage 1 on the horizontal plane through steps S1-1 to S1-4, and calibrates the single fish pixel point benchmark corresponding to each annular area, and then calculates the number of farmed fish in each annular area one by one according to the annular area through steps S2-3-1 and S2-3-2, so as to obtain the number of farmed fish contained in the sonar scanning image of each layer, that is, the layered fish number sequence, thereby reducing the influence of the mutual occlusion of the farmed fish in the target submerged and floating fish farming cage 1 on the scanning result of the omnidirectional sonar 2, and further improving the accuracy of the present invention in estimating the total amount of fish in the target submerged and floating fish farming cage 1.

[0074] Preferred: See Figure 3 The omnidirectional sonar 2 is loaded on the drone 4 through the vertical lifting drive device 3. The drone 4 lifts the omnidirectional sonar 2 to the center position of the target submerged fish farming cage 1, and the vertical lifting drive device 3 drives the omnidirectional sonar 2 to move up and down in the vertical direction.

[0075] Step S3, fish population estimation model training step, comprises:

[0076] Step S3-1, executing step S2 multiple times to obtain multiple sets of training data, each set of training data includes: a layered fish quantity sequence and a corresponding actual fish quantity, wherein the actual fish quantity is the actual quantity of cultured fish in the target submerged and floating fish farm culture cage 1 when executing step S2;

[0077] Preferably, in step S3-1, performing step S2 multiple times satisfies the following conditions:

[0078] Condition 1: When step S2 is executed each time, the actual number of fish in the target submerged and floating fish farm culture cage 1 is different;

[0079] Condition 2: The calibration fish used for the single fish pixel point benchmark calibration in step S1 is of the same species and has a similar growth cycle as the farmed fish in the target submerged and floating fish farm aquaculture cage 1 when executing steps S2 and S4.

[0080] Step S3-2, taking the stratified fish quantity sequence as input and the actual fish quantity as output, the AutoGluon model is trained with the training data obtained in step S3-1 to obtain a fish quantity estimation model, thereby being able to solve the nonlinear mapping problem between the stratified fish quantity sequence, that is, the number of farmed fish from the first layer to the M / Nth layer identified in step S2 and the actual number of farmed fish in the target submerged and floating fish farm farming cage 1;

[0081] Preferably: in step S3-2, the AutoGluon model is trained using the following settings:

[0082] The AutoGluon model uses a two-layer stack; the algorithm model for training is determined by AutoGluon, and all algorithms used are trained using default hyperparameters. The training time is not set, and the trained models and results are automatically saved by the algorithm. The model will autonomously try all possible algorithm sub-models that can be used to invert the quantity sequence to obtain the actual quantity, and continuously adjust the parameter scale and value of the sub-model, and finally use the sub-model with the highest score or the superposition of two sub-models to perform the task of estimating the number of fish.

[0083] The AutoGluon model selects the best_quality gear for training to obtain a better model through a longer training time. The experimental data is divided into training data and test data, and the fish quantity table data of each water layer is converted into a csv format file. Actual quantity represents the actual number of fish, which is the target data that needs to be inverted, and Depth represents the number of fish at the corresponding depth.

[0084] After the AutoGluon model uses TabularDataset() to read the training data, it sets the label label to Actual quantity and the num_gpus parameter to 1, which allows the GPU to be used to train the model, thereby speeding up the training. Then, TabularPredictor.fit() is used to start training.

[0085] The AutoGluon model can train two models: the default gear and the best quality gear, and the best one can be used.

[0086] Step S4, total fish quantity estimation step, comprises:

[0087] Step S4-1, at the target evaluation time, obtain the corresponding stratified fish quantity sequence according to step S2;

[0088] Step S4-2, input the hierarchical fish quantity sequence obtained in step S4-1 into the fish quantity estimation model to output an estimated value of the total fish quantity of the target submerged and floating fish farm culture cage 1 at the target evaluation time.

[0089] Therefore, the present invention can use the omnidirectional sonar 2 to estimate the total amount of fish in the target submerged and floating fish farm aquaculture cage 1 in a relatively simple and fast data collection method through steps S1 to S4, so as to evaluate the aquaculture capacity of the target submerged and floating fish farm aquaculture cage 1, and has the advantage of accurate total fish amount estimation result;

[0090] Moreover, since the present invention only needs to use the omnidirectional sonar 2 to obtain all the data required for the implementation method according to step S2, the requirements for scanning the omnidirectional sonar 2 are relatively low. It only needs to scan the complete target submerged fishery breeding cage 1 in the vertical direction at a uniform speed. It has the advantages of being simple, convenient, efficient and fast, and can solve the problems of the prior art in that the data monitoring and collection requirements for the target submerged fishery breeding cage 1 are high, the monitoring equipment is expensive, the operation is complicated and the cost is high.

[0091] The present invention is not limited to the above-mentioned specific implementation modes. According to the above-mentioned contents, in accordance with the common technical knowledge and customary means in the field, without departing from the above-mentioned basic technical ideas of the present invention, the present invention can also make other various forms of equivalent modifications, replacements or changes, all of which fall within the protection scope of the present invention.

Claims

1. A method for evaluating the aquaculture capacity of a submerged and floating fish farm, characterized in that: include: Step S1, the step of calibrating the pixel count of farmed fish, is: The target submerged and floating fish farm aquaculture cage (1) is calibrated using an omnidirectional sonar (2) to obtain a single fish pixel point benchmark, which represents the average pixel point of a fish in a sonar scanning image scanned by the omnidirectional sonar (2); Step S2, fish quantity data layered measurement step, comprises: Step S2-1, placing the omnidirectional sonar (2) at the center of the target submerged and floating fish farm aquaculture cage (1), and driving the omnidirectional sonar (2) to move at a uniform speed along the vertical direction from the cage bottom surface (1-1) of the target submerged and floating fish farm aquaculture cage (1) to the water surface, or driving the omnidirectional sonar (2) to move at a uniform speed along the vertical direction from the water surface to the cage bottom surface (1-1) of the target submerged and floating fish farm aquaculture cage (1), so as to complete an overall scan; Step S2-2, the data outputted by the omnidirectional sonar (2) after an overall scan is played back through its supporting software to be converted into a single sonar scan video, and a video frame is intercepted from the single sonar scan video every N seconds as a sonar scan image, so as to obtain M / N sonar scan images arranged according to the interception time, which are sequentially recorded as the first layer to the M / Nth layer of sonar scan images, wherein M is the duration of the single sonar scan video, that is, the time to complete an overall scan, and N is the time required for the omnidirectional sonar (2) to complete a 360° scan in the horizontal direction; Step S2-3, binarizing the sonar scanning image and removing noise points, so as to identify effective pixel points representing farmed fish in the target submerged and floating fish farm aquaculture cage (1) in each layer of the sonar scanning image, and, based on the single fish pixel point reference calibrated in step S1, calculating the number of farmed fish contained in each layer to the M / Nth layer of the sonar scanning image, and recording it as a layered fish number sequence; Step S3, fish population estimation model training step, comprises: Step S3-1, executing step S2 multiple times to obtain multiple sets of training data, each set of training data comprising: a layered fish number sequence and a corresponding actual fish number, wherein the actual fish number is the actual number of cultured fish in the target submerged and floating fish farm culture cage (1) when executing step S2; Step S3-2, taking the stratified fish quantity sequence as input and the actual fish quantity as output, and training the AutoGluon model with the training data obtained in step S3-1 to obtain a fish quantity estimation model; Step S4, total fish quantity estimation step, comprises: Step S4-1, at the target evaluation time, obtain the corresponding stratified fish quantity sequence according to step S2; Step S4-2, inputting the hierarchical fish quantity sequence obtained in step S4-1 into the fish quantity estimation model to output an estimated value of the total fish quantity in the target submerged and floating fish farm culture cage (1) at the target evaluation time.

2. The method for evaluating the aquaculture capacity of a submerged and floating fish farm according to claim 1, characterized in that: In step S1, the method for calibrating the pixel number benchmark of a single fish is: Step S1-1, dividing the target submerged and floating fish farm aquaculture cage (1) into a plurality of annular areas with the center position as the axis and equal radial width; Step S1-2, placing the omnidirectional sonar (2) at the center of the target submerged and floating fish farm aquaculture cage (1), and continuously tracking and scanning the calibration fish in the target submerged and floating fish farm aquaculture cage (1) until the calibration fish swims through each of the annular areas; Step S1-3, extracting the total number of pixels of the calibration fish in each annular area from all the sonar scanning images obtained by scanning the omnidirectional sonar (2) in step S1-2; Step S1-4, performing mean processing on the total number of pixels in each annular area to calibrate the single fish pixel number benchmark corresponding to each annular area; In step S2-3, the method for calculating the number sequence of stratified fish is as follows: Step S2-3-1, counting the number of effective pixels in each annular area of ​​each layer of the sonar scanning image from the identified effective pixels; Step S2-3-2, for each layer of the sonar scanning image: calculate the number of farmed fish in each annular area based on the number of valid pixels in the same annular area and the number of pixels of a single fish, and record the total number of farmed fish in all annular areas as the number of farmed fish contained in the sonar scanning image of this layer, thereby obtaining the layered fish number sequence.

3. The method for evaluating the aquaculture capacity of a submerged and floating fish farm according to claim 2, characterized in that: In the step S1-1, the radial width δd of each annular area is taken as the average length of the cultured fish in the target submerged and floating fish farm culture cage (1).

4. The method for evaluating the aquaculture capacity of a submerged and floating fish farm according to any one of claims 1 to 3, characterized in that: The omnidirectional sonar (2) is loaded onto a drone (4) via a vertical lifting drive device (3); the drone (4) lifts the omnidirectional sonar (2) to the center of a target submerged and floating fish farming cage (1); and the vertical lifting drive device (3) drives the omnidirectional sonar (2) to move in a vertical direction.

5. The method for evaluating the aquaculture capacity of a submerged and floating fish farm according to any one of claims 1 to 3, characterized in that: In step S3-1, performing step S2 multiple times meets the following conditions: Condition 1: When step S2 is performed each time, the actual number of fish in the target submerged and floating fish farm culture cage (1) is different from each other; Condition 2: The calibration fish used for the single fish pixel point benchmark calibration in step S1 is of the same species and has a similar growth cycle as the farmed fish in the target submerged and floating fish farm aquaculture cage (1) when executing steps S2 and S4.

6. The method for evaluating the aquaculture capacity of a submerged and floating fish farm according to any one of claims 1 to 3, characterized in that: In step S3-2, the AutoGluon model is trained using the following settings: The AutoGluon model uses a two-layer stack; The AutoGluon model selects the best_quality gear for training; After the AutoGluon model uses TabularDataset() to read the training data, it sets the label label to Actual quantity, sets the num_gpus parameter to 1, and then uses TabularPredictor.fit() to start training.

Images