A method for estimating the number of fish schools in a cage based on fish school density

Abstract

The present invention discloses a method for estimating the number of fish in a cage based on fish density. First, sonar images corresponding to each circle of rotational scanning of the target cage are obtained through steps S1 and S2, and the number of fish corresponding to each sonar image extracted based on image recognition is obtained. Then, a linear correlation curve between the effective area of the net in water and the sampling depth, and a non-linear correlation curve between fish density and sampling depth are obtained through step S3. Finally, the two curves are integrated by intervals through step S4, the number of fish in each interval is calculated layer by layer and accumulated to obtain an estimated value of the number of fish in the target cage. Therefore, the present invention fully considers the non-uniformity of fish density and the area of the cage net changing with depth, so as to improve the measurement accuracy of the number of fish in the cage, and has the advantages of high accuracy, high efficiency, fast implementation, convenience and rapidity.

Description

Technical Field

[0001] The present invention relates to the estimation and measurement of fishery resources, and specifically to a method for estimating the number of fish schools in a cage based on fish school density. Background Art

[0002] In modern aquaculture, cage culture has become an important aquaculture method due to its efficient space utilization and good water circulation. However, how to efficiently and accurately monitor the number of fish schools in the cage remains a key issue in aquaculture management. Traditional manual observation or monitoring methods are limited by water turbidity, light conditions, and labor costs, and often cannot provide accurate fish school number information, affecting aquaculture yield assessment and feeding strategy optimization. As an advanced underwater detection technology, sonar technology has been widely used in fishery monitoring because it can penetrate water bodies and provide real-time fish school distribution information. However, existing sonar monitoring technologies mostly stay in the two-dimensional distribution recognition stage and cannot accurately estimate the overall number of fish schools in the cage. Especially in complex environments with uneven fish school density distribution or net deformation, the measurement error is relatively large. Summary of the Invention

[0003] The technical problem to be solved by the present invention is: to provide a method for estimating the number of fish schools in a cage based on fish school density.

[0004] To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0005] A method for estimating the number of fish schools in a cage based on fish school density, characterized by comprising:

[0006] Step S1, refer to Figure 2 , use sonar to scan the target cage in the following manner: the sonar is arranged at the center position of the target cage and descends from the water surface to the bottom of the target cage or ascends from the bottom of the target cage to the water surface at a fixed speed v in the vertical direction. At the same time, the sonar rotates and scans in the horizontal direction at a constant angular velocity ω; thus, the vertical movement of the sonar is synchronized with the horizontal rotational scan to perform an overall scan of the target cage to obtain complete sonar image data of the cage; among them, the fixed speed v at which the sonar rises and falls needs to be appropriate, that is, the overlapping part of each scan should not be too large, nor should there be a large area that is not scanned.

[0007] Step S2, refer to Figure 3 , from the complete sonar image data of the cage, based on each complete rotation scan of the sonar, extract the sonar image corresponding to each rotation scan; among them, the complete sonar image data of the cage can be saved in video format, and the sonar image can be extracted based on the time T required for the sonar to rotate and scan one circle, or other parameters can be used as the basis for extraction;

[0008] And, referring to Figure 4 and Figure 5 , each sonar image extracted is recognized by an image recognition model to obtain the corresponding fish school quantity; wherein, the fish school quantity and the sampling depth corresponding to the sonar image of the th circle of rotational scanning are respectively denoted as and , , is a positive integer, and , k is the total number of rotational circles of the sonar during the scanning process in step S1, is the single-circle depth that the sonar moves in the vertical direction corresponding to one circle of rotational scanning in step S1;

[0009] Step S3. Obtain the linear correlation curve and the non-linear correlation curve , the specific steps include:

[0010] Step S3-1. Referring to Figure 6 , according to the following formula, calculate the effective area of the netting in water of the part of the target cage netting corresponding to each circle of rotational scanning of the sonar during the time period of this circle of rotational scanning, so as to approximately represent the deformation change of the target cage netting in the water body:

[0011] ;

[0012] In the formula, represents the effective area of the netting in water of the target cage netting corresponding to the th circle of rotational scanning, represents the sum of the pixel points in the internal area of the target cage netting in the sonar image corresponding to the th circle of rotational scanning, see the square area in Figure 6 ; represents the sum of the pixel points of the sonar image corresponding to the th circle of rotational scanning; represents the rotational scanning area of the sonar, which is determined by the scanning radius r of the sonar and is calculated according to ;

[0013] Step S3-2. Referring to Figure 7 , based on the sampling depth and the effective area of the netting in water corresponding to each circle of rotational scanning, obtain the linear correlation curve of the effective area of the netting in water and the sampling depth by linear fitting;

[0014] Step S3-3: Convert the number of fish schools obtained in Step S2 into fish school density according to the following formula, so as to obtain a fish school density data point based on each sonar image, and denote the fish school density data point corresponding to the sonar image of the th circle of rotational scanning as :

[0015] ;

[0016] In the formula, represents the fish school density corresponding to the sonar image of the th circle of rotational scanning;

[0017] Moreover, for two adjacent fish school density data points corresponding to each sampling depth interval and , interpolation is performed using the same interpolation method to obtain a batch of interpolated fish school density data points corresponding to each sampling depth interval;

[0018] Thus, based on all the fish school density data points and the interpolated fish school density data points, a non-linear correlation curve of fish school density versus sampling depth is fitted;

[0019] Step S4: Calculate and accumulate the estimated value of the number of fish schools in the target net cage according to the following formula:

[0020] .

[0021] Therefore, the present invention first obtains the sonar images corresponding to each circle of rotational scanning of the target net cage by the sonar through Step S1 and Step S2, and based on the number of fish schools corresponding to each sonar image extracted by image recognition, then obtains the linear correlation curve of the effective area of the net in water versus sampling depth , as well as the non-linear correlation curve of fish school density versus sampling depth through Step S3, and finally integrates the two curves by intervals in Step S4, calculates the number of fish schools in each interval layer by layer and accumulates to obtain the estimated value of the number of fish schools in the target net cage. Therefore, the present invention fully considers the non-uniformity of fish school density and the area of the net cage net changing with depth, so as to improve the measurement accuracy of the number of fish schools in the net cage, and has the advantages of high accuracy, high efficiency, fast implementation and convenience.

[0022] Preferably: In the said Step S1, the fixed speed v is calculated according to the following formula:

[0023] ;

[0024] In the formula, H is the total depth of the target net cage, T is the time required for the sonar to rotate and scan one circle at the angular velocity, and L is the scanable length of the sonar for the netting of the target net cage. See Figure 2 , and the scanable length L is the length of the scan range of the sonar projected onto the netting of the target net cage, which is calculated according to the opening angle α of the sonar.

[0025] Thus, it can be ensured that the water area scanned when the sonar moves in the vertical direction is as non-repetitive as possible, and at the same time, the situation of not being scanned is avoided.

[0026] Preferably: In step S1, the sonar is moved in the vertical direction by a mechanical lifting device driven by a motor or by manual driving; and, the sonar is an omnidirectional sonar so as to be able to perform 360° rotational scanning in the horizontal direction by driving the scanning beam.

[0027] Preferably: In step S2, the image recognition model adopts the YOLOv8s model, and its training method is as follows:

[0028] Step S2-1: Randomly extract 500 from the existing sonar images, divide them into a training set and a test set according to a ratio of 7:3, and use YoloLabel software to label the fish positions in the sonar images;

[0029] Step S2-2: Based on the labeled sonar images, perform model training using ultralytics, and its training parameters are: the number of training rounds is 300, the initial learning rate is 0.001, the picture size is 800×600, and finally the image recognition model is obtained so as to be able to identify the fish positions in the sonar images using the image recognition model. See Figure 4 , and count the fish positions to obtain the corresponding number of fish.

[0030] Thus, in step S2 of the present invention, by adopting the YOLOv8s model, the number of fish can be efficiently and accurately counted, and compared with manual observation or traditional two-dimensional sonar analysis, the data acquisition efficiency and accuracy are greatly improved.

[0031] Preferably: In step S3-2, the linear fitting adopts unary linear regression analysis, and the regression model is optimized by minimizing the mean square error.

[0032] Its formula is:

[0033] ;

[0034] where m is the number of samples, is the predicted value of the model, is the true value.

[0035] Preferably, in the step S3-3, the interpolation method adopts cubic spline interpolation to avoid overfitting and underfitting and obtain a non-linearly correlated curve with the best fitting effect.

[0036] That is, for each sampling depth interval construct a cubic polynomial , and use the cubic polynomial function of the sub-intervals to approximate the relationship between data points to obtain the final non-linearly correlated curve . Each cubic polynomial of each segment has the following form:

[0037] ;

[0038] where , , , are undetermined coefficients.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] First, the present invention first obtains the sonar images corresponding to each circle of rotational scanning of the target cage by the sonar through steps S1 and S2, and based on the number of fish schools corresponding to each sonar image extracted by image recognition, and then obtains the effective area of the net in water and the linear correlation curve with the sampling depth , as well as the non-linear correlation curve between the fish school density and the sampling depth . Finally, through step S4, the two curves are integrated by intervals, and the number of fish schools in each interval is calculated layer by layer and accumulated to obtain the estimated value of the number of fish schools in the target cage . Therefore, the present invention fully considers the non-uniformity of the fish school density and the net area of the cage changing with the depth, so as to improve the measurement accuracy of the number of fish schools in the cage, and has the advantages of high accuracy, high efficiency, fast implementation and convenience.

[0041] Second, the fixed speed v for driving the sonar to move up and down in the vertical direction in step S1 of the present invention is calculated according to the formula based on the total depth H, time T and scan length L, which can ensure that the water body scanned by the sonar when moving in the vertical direction is as non-repetitive as possible, and at the same time avoid the situation of not being scanned.

[0042] Thirdly, in step S2 of the present invention, the YOLOv8s model is adopted, which can efficiently and accurately count the number of fish schools, greatly improving the data acquisition efficiency and accuracy compared with manual observation or traditional two-dimensional sonar analysis.

[0043] Fourthly, in step S3-3 of the present invention, cubic spline interpolation is adopted, which can avoid overfitting and underfitting and obtain the optimal fitting effect of a non-linearly related curve. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention will be further described in detail below with reference to the drawings and specific embodiments:

[0045] Figure 1 is a flowchart of the present invention;

[0046] Figure 2 is a schematic diagram of the sonar scanning in water in step S1 of the present invention;

[0047] Figure 3 is the sonar image corresponding to the i-th circle of rotational scanning in step S2 of the present invention;

[0048] Figure 4 is a schematic diagram of the image data obtained by the image recognition model for recognizing the sonar image in step S2 of the present invention;

[0049] Figure 5 is the sampling depth of all sonar images obtained in step S2 of the present invention and the number of fish schools relation curve graph;

[0050] Figure 6 is a schematic diagram of the total number of pixel points calibrating the internal area of the target cage net in step S3-1 of the present invention diagram;

[0051] Figure 7 is the sampling depth in step S3-3 of the present invention and the effective area of the net in water and the linearly related curve obtained by fitting schematic diagram. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] The present invention will be described in detail below in conjunction with the embodiments and their 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 labor on the premise of not departing from the inventive concept of the present invention belong to the protection scope of the present invention.

[0053] As Figure 1 ​As shown in the figure, the present invention discloses a method for estimating the number of fish in a net cage based on fish density, including:

[0054] Step S1, referring to Figure 2 , use a sonar to scan the target net cage in the following manner: The sonar is arranged at the center of the target net cage and dives from the water surface to the bottom of the target net cage or rises from the bottom of the target net cage to the water surface at a fixed speed v in the vertical direction. At the same time, the sonar rotates and scans in the horizontal direction at a constant angular velocity ω; thus, the vertical movement of the sonar and the horizontal rotational scan are synchronized to perform an overall scan of the target net cage to obtain complete sonar image data of the net cage; among them, the fixed speed v at which the sonar moves up and down needs to be appropriate, that is, the overlapping part of each scan should not be too large, nor should there be a large area that is not scanned.

[0055] Preferably: In the step S1, the fixed speed v is calculated according to the following formula:

[0056] ;

[0057] In the formula, H is the total depth of the target net cage, T is the time required for the sonar to rotate and scan one circle at the angular velocity, and L is the scanable length of the sonar for the net cage netting 1 of the target net cage. See Figure 2 , the scanable length L is the length of the scan range of the sonar projected onto the net cage netting 1 of the target net cage, and it is calculated according to the opening angle α of the sonar.

[0058] Thus, it can be ensured that the water area scanned when the sonar moves in the vertical direction is as non-repetitive as possible, and at the same time, the situation of not being scanned is avoided.

[0059] Preferably: In the step S1, the sonar is moved in the vertical direction by a mechanical lifting device driven by a motor or by manual driving; and, the sonar is an omnidirectional sonar so as to be able to perform a 360° rotational scan in the horizontal direction by driving the scanning beam.

[0060] Step S2, referring to Figure 3 , from the complete sonar image data of the net cage, based on each time the sonar completes one circle of rotational scan, extract the sonar image corresponding to each circle of rotational scan; among them, the complete sonar image data of the net cage can be saved in video format, and the sonar image is extracted based on the time T required for the sonar to rotate and scan one circle, or it can be extracted based on other parameters;

[0061] And, referring to Figure 4 and Figure 5 , through an image recognition model, identify each extracted sonar image to obtain the corresponding number of fish schools; among them, the The number of fish schools and the sampling depth corresponding to the circular rotation scan of the sonar image are respectively denoted as and , , is a positive integer, and , where k is the total number of rotation circles during the scan in step S1, is the single - circle depth of the sonar moving in the vertical direction corresponding to one - circle rotation scan in step S1;

[0062] Preferably: in the said step S2, the image recognition model adopts the YOLOv8s model, and its training method is as follows:

[0063] Step S2 - 1: Randomly select 500 sonar images from the existing sonar images, divide them into a training set and a test set according to a ratio of 7:3, and use YoloLabel software to mark the positions of fish schools in the sonar images;

[0064] Step S2 - 2: Use the marked sonar images to train the model based on ultralytics, and its training parameters are: the number of training rounds is 300, the initial learning rate is 0.001, the picture size is 800×600, and finally obtain the said image recognition model to be able to identify the positions of fish schools in the sonar images, see Figure 4 , and count the fish school positions to obtain the corresponding number of fish schools.

[0065] Thus, in step S2 of the present invention, by adopting the YOLOv8s model, the number of fish schools can be efficiently and accurately counted, which greatly improves the data acquisition efficiency and accuracy compared with manual observation or traditional two - dimensional sonar analysis.

[0066] Step S3: Obtain the linear correlation curve and the non - linear correlation curve , and the specific steps include:

[0067] Step S3 - 1: See Figure 6 , according to the following formula, calculate the effective area of the netting of the target cage net 1 in the water during the time period of each - circle rotation scan of the sonar, to approximately represent the deformation change of the target cage net 1 in the water body:

[0068] ;

[0069] In the formula, represents the effective area of the target cage net 1 in the water corresponding to the i - th circle rotation scan, represents the sum of the pixel points in the internal area of the target cage net 1 in the sonar image corresponding to the -th circle rotation scan, see Figure 6the square area in; indicating the sum of the pixel points of the sonar image corresponding to the circular rotation scan; indicating the rotational scan area of the sonar, which is determined by the scan radius r of the sonar, and calculated according to calculate;

[0070] Step S3-2, referring to Figure 7 , based on the sampling depth corresponding to each circular rotation scan and the effective area of the net in water , obtain the linear correlation curve of the effective area of the net in water and the sampling depth through linear fitting;

[0071] Preferably: in the step S3-2, the linear fitting adopts unary linear regression analysis and optimizes the regression model by minimizing the mean square error.

[0072] Its formula is:

[0073] ;

[0074] where m is the number of samples, is the predicted value of the model, is the true value.

[0075] Step S3-3, convert the fish school quantity obtained in step S2 into fish school density according to the following formula, so as to obtain a fish school density data point based on each sonar image, and denote the fish school density data point of the sonar image corresponding to the i-th circular rotation scan as :

[0076] ;

[0077] In the formula, represents the fish school density of the sonar image corresponding to the i-th circular rotation scan;

[0078] And, for each adjacent two fish school density data points corresponding to each sampling depth interval and , perform interpolation with the same interpolation method to obtain a batch of interpolated fish school density data points corresponding to each sampling depth interval;

[0079] Thus, based on all the fish school density data points and the interpolated fish school density data points, fit to obtain the non-linear correlation curve of the fish school density and the sampling depth ;

[0080] Preferably, in the step S3-3, the interpolation method adopts cubic spline interpolation to avoid overfitting and underfitting and obtain a non-linearly correlated curve with the best fitting effect.

[0081] That is, for each sampling depth interval construct a cubic polynomial , and use the cubic polynomial function of the sub-intervals to approximate the relationship between data points to obtain the final non-linearly correlated curve . Each cubic polynomial of each segment has the following form:

[0082] ;

[0083] where , , , are undetermined coefficients.

[0084] Step S4: Calculate the estimated value of the number of fish in the target net cage by integral accumulation according to the following formula :

[0085] .

[0086] Therefore, the present invention first obtains the sonar images corresponding to each rotation scan of the target net cage by the sonar through steps S1 and S2, and the number of fish corresponding to each sonar image extracted based on image recognition. Then, through step S3, the effective area of the net in water and the sampling depth linear correlation curve , as well as the fish density and the sampling depth non-linear correlation curve . Finally, through step S4, the two curves are integrated by intervals, and the number of fish in each interval is calculated layer by layer and accumulated to obtain the estimated value of the number of fish in the target net cage . Therefore, the present invention fully considers the non-uniformity of the fish density and the net cage net area changing with depth, so as to improve the measurement accuracy of the number of fish in the net cage, and has the advantages of high accuracy, high efficiency, fast implementation and convenience.

[0087] The present invention is not limited to the above specific embodiments. According to the above content, according to the common technical knowledge and conventional means in the art, without departing from the above basic technical idea of the present invention, the present invention can also make various other forms of equivalent modifications, substitutions or changes, all of which fall within the protection scope of the present invention.

Claims

1. A method for estimating the number of fish in a cage based on fish density, characterized in that: include: Step S1, scanning the target cage with sonar in the following manner: the sonar is arranged at the center of the target cage, and dives from the water surface to the bottom of the target cage or rises from the bottom of the target cage to the water surface at a fixed speed v in the vertical direction, and at the same time, the sonar performs rotational scanning in the horizontal direction at a constant angular speed ω; thereby, obtaining complete sonar image data of the cage; Step S2, from the complete sonar image data of the cage, based on each time the sonar completes a rotation scan, extracting the sonar image corresponding to each rotation scan; In addition, each extracted sonar image is identified through the image recognition model to obtain the corresponding number of fish schools; the number of fish schools and the sampling depth of the sonar image corresponding to the i-th rotation scan are recorded as N i and h i ,h i+1 -h i =△h, i is a positive integer, and 1≤i≤k, k is the total number of rotations of the sonar in the scanning process of step S1, and △h is the single-circle depth of the sonar moving in the vertical direction corresponding to one rotation scan completed in step S1; Step S3, obtaining the linear correlation curve S i (h i ) and the nonlinear correlation curve ρ i (h i ), the specific steps include: Step S3-1, when the sonar performs each rotation scan, the effective area of ​​the net (1) in the water corresponding to the rotation scan in the rotation scan period is calculated according to the following formula: In the formula, S i M represents the effective area of ​​the target cage net (1) in the water corresponding to the i-th rotation scan, i represents the total number of pixels in the internal area of ​​the target cage net (1) in the sonar image corresponding to the i-th rotation scan; T represents the total number of pixels in the sonar image corresponding to the i-th rotation scan; S T represents the rotational scanning area of ​​the sonar; Step S3-2: based on the sampling depth h corresponding to each rotation scan i and the effective area S of the net in water i , the effective area S of the net in water is obtained by linear fitting i With sampling depth h i The linear correlation curve S i (h i ); Step S3-3: Convert the number of fish obtained in step S2 into fish density according to the following formula, so as to obtain a fish density data point based on each sonar image, and record the fish density data point of the sonar image corresponding to the i-th rotation scan as (h i ,ρ i ): In the formula, ρ i It represents the fish density of the sonar image corresponding to the i-th rotation scan; And, for each sampling depth interval [h i ,h i+1 ] corresponding to two adjacent fish density data points (h i ,ρ i ) and (h i+1 ,ρ i+1 ), interpolating using the same interpolation method to obtain a batch of interpolated fish density data points corresponding to each sampling depth interval; Therefore, based on all the fish density data points and the interpolated fish density data points, the fish density ρ is fitted. i With sampling depth h i The nonlinear correlation curve ρ i (h i ); Step S4: Calculate the estimated value P of the number of fish in the target cage according to the following formula:

2. The method for estimating the number of fish in a cage based on fish density according to claim 1, characterized in that: In step S1, the sonar is moved in the vertical direction by a mechanical lifting device driven by a motor or by human power; and the sonar is an omnidirectional sonar, so that it can perform 360° rotation scanning in the horizontal direction by driving the scanning beam.

3. The method for estimating the number of fish in a cage based on fish density according to claim 1 or 2, characterized in that: In step S2, the image recognition model adopts the YOLOv8s model, and its training method is as follows: Step S2-1, randomly select 500 sonar images from the existing ones, divide them into training set and test set in a ratio of 7:3, and use YoloLabel software to mark the fish positions on the sonar images; Step S2-2, using the annotated sonar images, model training is performed based on ultralytics, and the training parameters are: 300 training rounds, an initial learning rate of 0.001, and finally the image recognition model is obtained, so that the image recognition model can be used to identify the position of fish schools in the sonar image, and the fish school positions are counted to obtain the corresponding number of fish schools.

4. The method for estimating the number of fish in a cage based on fish density according to claim 1 or 2, characterized in that: In step S3-2, the linear fitting adopts univariate linear regression analysis, and optimizes the regression model by minimizing the mean square error.

5. The method for estimating the number of fish in a cage based on fish density according to claim 1 or 2, characterized in that: In the step S3-3, the interpolation method adopts cubic spline interpolation.

Images