A system and method for counting the amount of organisms passing through underwater passages

Abstract

This invention discloses an underwater passage biological throughput counting system and method, comprising a camera unit, a counting processing unit, and a pressure-resistant sealed shell. The camera unit and the counting processing unit are built into the pressure-resistant sealed shell and are connected. The camera unit is used to acquire a video stream of the underwater passage over a period of time. The counting processing unit obtains the number of targets to be measured passing through the underwater passage during the time period based on the video stream. The targets to be measured include, but are not limited to, underwater organisms such as fish and shrimp that can be clearly captured by the camera. The counting processing unit realizes the statistics of passing targets by setting up a virtual channel, and further achieves high counting accuracy and precision even under high biological throughput by identifying duplicate targets.

Description

Technical Field

[0001] This invention relates to the field of underwater equipment technology, specifically to an underwater channel biological throughput counting system and method. Background Technology

[0002] Underwater channel counting is a method for calculating the number of underwater targets passing through. This method first allows the targets to pass through a channel; then, counting devices installed in the channel count the number of targets passing through, often used for calculation. Existing underwater channel counting systems generally consist of three main parts: a counting channel, sensing devices, and a processing module. The counting channel is the effective counting space of the system, located in the waterway, and targets passing through this channel are included in the count. The sensing devices are information acquisition devices for targets within the channel, generally installed in the counting channel section, collecting information about passing targets; these typically include infrared sensors, optical cameras, etc. The processing module receives the target information collected by the sensing devices, processes it, and performs the counting. The following are existing underwater biological counting methods:

[0003] Patent CN115170535A discloses a fish counting method and system for fish passages in hydropower engineering based on image recognition. It simplifies target contour extraction through binarization and performs filtering based on the inherent errors in traditional counting methods, ensuring that only fish contours that can be counted are those that will not be counted repeatedly. Furthermore, by tracking the swimming direction of the fish, it only counts fish that have completely entered and exited the fish passage. This invention uses simple and low-cost image processing technology to achieve fish counting, improves counting accuracy, has strong scalability, and is suitable for various counting scenarios.

[0004] Patent CN116363494A discloses a method and system for monitoring fish populations and tracking migration. The method includes: training a deep learning model using an underwater fish image dataset to obtain a fish detection model; predicting a first fish detection box and its corresponding confidence level using the fish detection model; calculating a probability map of fish in continuously acquired images of fish to be tested using a Gaussian mixture model and extracting a second fish detection box; calculating the total number of the first and second fish detection boxes as a preliminary statistical value of fish populations; comparing the depth features of fish within the detection boxes at different times, tracking the fish trajectories, removing duplicate fish counts, and obtaining a fish population monitoring value; and tracking fish migration based on the fish trajectories.

[0005] Patent CN116145626A discloses a fish passage monitoring device, which includes a fence with an inner wall forming a channel suitable for migratory fish to pass through, and the fence being fixed inside the fish passage; several grid bars arranged inside the fence to divide the internal channel of the fence into multiple fish passage units; a camera connected to the grid bars to capture images of the fish passage units; a data acquisition module communicating with the camera to acquire video information captured by the camera; and a control module communicating with the camera and the data acquisition module to control the operating status of the camera and the data acquisition module. This improves the image quality acquired by the camera and facilitates subsequent identification and statistics of fish.

[0006] This invention provides another underwater channel biological throughput counting system and method based on neural network image target recognition, virtual decision line algorithm and same target detection algorithm. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies, this invention provides an underwater channel biological throughput counting system and method, which solves the problems that existing underwater channel counting devices are difficult to operate in deeper waters and have low accuracy in processing repetitive large amounts of data.

[0008] To achieve the above objectives, the present invention relates to an underwater passage biological throughput counting system, comprising a camera unit, a counting processing unit, and a pressure-resistant sealed housing. The camera unit and the counting processing unit are built into the pressure-resistant sealed housing and are connected. The camera unit is used to acquire a video stream of the underwater passage over a period of time, and the counting processing unit obtains the number of targets to be measured passing through the underwater passage during the time period based on the video stream. The targets to be measured include, but are not limited to, fish, shrimp, and other underwater organisms that can be clearly captured by the camera.

[0009] The counting processing unit includes a virtual channel drawing module, a target recognition module, a virtual recognition box drawing and judgment point setting module, a distance judgment module, a same target recognition module, and a target counting module;

[0010] The virtual channel drawing module is used to symmetrically set two virtual detection lines on both sides of the vertical center line of each frame of the image to form a virtual channel. The horizontal coordinates of the two virtual detection lines in the image are X1 and X2, respectively. The total number of the targets to be tested is obtained by counting the targets entering the virtual channel in all video frames.

[0011] The target recognition module identifies the target in each frame of the image based on existing target recognition methods;

[0012] The virtual recognition frame drawing and decision point setting module is used to draw a virtual recognition frame around the target to be measured, set decision points based on the recognition frame. The decision point can be a certain point on or inside the recognition frame, and the coordinates of the decision point in the image are (X, Y).

[0013] The distance determination module is used to determine whether the target to be measured recognized in each frame of the image is within the virtual channel. The determination method is as follows: If X1 + k(X2 - X1) ≤ X ≤ X2 - k(X2 - X1), 0 < k < 0.25, it is considered that the target to be measured enters the virtual channel; if not, that is, the target to be measured does not enter the virtual channel.

[0014] The same target recognition module is used to determine whether the targets to be measured that enter the virtual channel in two adjacent frames of the image are the same target. If the height difference between a certain target to be measured in the next frame of the image and all targets to be measured in the current frame of the image is the smallest and this minimum value is within the set range, it is determined that the two are the same target.

[0015] The target counting module counts based on the results of distance determination and the discrimination of the same target. If the distance determination confirms that the target to be measured enters the virtual channel and the discrimination of the same target determines that the target to be measured is not the same target, then counting is performed.

[0016] As one implementation, the pressure-resistant sealed housing is an integral structure. As another implementation, the pressure-resistant sealed housing is divided into a counting processing unit sealed housing and a camera sealed housing. The counting processing unit sealed housing includes a first sealed cylinder, a first front end cover, and a first rear end cover. The first front end cover and the first rear end cover are respectively sealed and fixed at the openings at the front and rear ends of the first sealed cylinder. The counting processing unit is built inside the counting processing unit sealed housing; the camera sealed housing includes a second sealed cylinder, a front lens, and a second rear end cover. The front lens and the second rear end cover are respectively sealed and fixed at the openings at the front and rear ends of the second sealed cylinder. The camera unit is built inside the camera sealed housing and is close to the front lens; One end of the transmission cable is sealed through the opening on the first front end cover and connected to the counting processing unit, and the other end is sealed through the opening on the second rear end cover and connected to the camera unit.

[0017] Furthermore, the underwater channel biological throughput counting system further includes a heat sink, which is used to dissipate heat from the counting processing unit. The heat sink is placed between the counting processing unit and the counting processing unit sealed housing.

[0018] Furthermore, the underwater channel biological throughput counting system further includes an interface. An interface is provided on the first rear end cover, which facilitates the counting processing unit to be externally connected to a power supply device.

[0019] Correspondingly, an underwater channel biological throughput counting method specifically includes the following steps:

[0020] (1) The camera unit acquires the video stream of the underwater channel over a period of time;

[0021] (2) The virtual channel drawing module sets two virtual detection lines symmetrically on both sides of the vertical center line of each frame to form a virtual channel. The horizontal coordinates of the two virtual detection lines in the image are X1 and X2, respectively. The total number of the targets to be tested is obtained by counting the targets entering the virtual channel in all video frames.

[0022] (3) The target recognition module identifies the target in each frame of the image based on existing target recognition methods;

[0023] (4) The virtual recognition box drawing and decision point setting module draws a virtual recognition box around the target to be tested and sets decision points based on the recognition box. The decision point can be a point on or inside the recognition box, and the coordinates of the decision point in the image are (X,Y).

[0024] (5) The distance determination module determines whether the target identified in each frame of the image is within the virtual channel. The determination method is as follows: if X1+k(X2-X1)≤X≤X2-k(X2-X1), then it is considered that the target has not entered the virtual channel; if the distance condition is not met, that is, the target has not entered the virtual channel,

[0025] (6) The same target recognition module determines whether the target to be tested entering the virtual channel in two adjacent frames is the same target. The determination method is as follows: the target to be tested in the underwater channel generally swims in the same direction. If the height difference between the two targets to be tested in two adjacent frames is the smallest and the minimum value is within the set range, then the two are identified as the same target.

[0026] (601) Record the height coordinates of the judgment points corresponding to all the targets to be tested that entered the virtual channel in the previous frame image;

[0027] (602) Calculate the distance between the height coordinates of the judgment point corresponding to each target entering the virtual channel in the current frame image and the height coordinates of the judgment points corresponding to all targets in the previous frame image, and determine the minimum distance among them.

[0028] (603) When the minimum distance is less than or equal to the set value, the two corresponding targets to be measured are identified as the same target. The set value is equal to (X2-X1) / 32k.

[0029] (7) The target counting module counts based on the results of distance determination and same target identification. If the distance determination confirms that the target to be tested has entered the virtual channel, and the same target identification determines that the target to be tested is not the same target, then counting is performed.

[0030] Compared with the prior art, the present invention has the following advantages: (1) By placing the camera unit and the counting processing unit in two smaller pressure-resistant containers, the pressure borne by a single pressure-resistant container is reduced, making it usable in deeper water; (2) The counting processing unit realizes the statistics of passing targets by setting up virtual channels, and then achieves high counting accuracy and precision even under high biological throughput by identifying repeated targets. Attached Figure Description

[0031] Figure 1 This invention relates to a schematic diagram of an underwater channel biological throughput counting system.

[0032] Figure 2 This is a schematic diagram of the counting processing unit structure involved in the present invention.

[0033] Figure 3 This is a schematic diagram showing the position of a target under test in three adjacent frames of images, as per the present invention.

[0034] Figure 4 This invention relates to a flowchart of a method for counting the amount of organisms passing through underwater channels. Detailed Implementation

[0035] To more clearly illustrate the content of this invention, the invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0036] Example 1

[0037] like Figure 1-2 As shown in the figure, this embodiment relates to an underwater passage biological passage counting system, which includes a camera unit 1, a counting processing unit 2, and a pressure-resistant sealed shell. The camera unit 1 and the counting processing unit 2 are built into the pressure-resistant sealed shell and are connected. The camera unit 1 is used to acquire a video stream of the underwater passage over a period of time. The counting processing unit 2 obtains the number of targets to be measured passing through the underwater passage during the time period based on the video stream. The targets to be measured include, but are not limited to, underwater organisms such as fish and shrimp that can be clearly captured by the camera.

[0038] The counting processing unit 2 includes a virtual channel drawing module 201, a target recognition module 202, a virtual recognition box drawing and judgment point setting module 203, a distance judgment module 204, a same target recognition module 205, and a target counting module 206;

[0039] The virtual channel drawing module 201 is used to symmetrically set two virtual detection lines on both sides of the vertical center line of each frame of the image to form a virtual channel. The horizontal coordinates of the two virtual detection lines in the image are X1 and X2, respectively. The total number of the targets to be tested is obtained by counting the targets entering the virtual channel in all video frames.

[0040] The target to be measured recognition module 202 recognizes the target to be measured in each frame of image based on existing target recognition methods;

[0041] The virtual recognition frame drawing and decision point setting module 203 is used to draw a virtual recognition frame around the target to be measured, set decision points based on the recognition frame, the decision point can be a certain point on or inside the recognition frame, and the coordinates of the decision point in the image are (X, Y).

[0042] The distance determination module 204 is used to determine whether the recognized target to be measured in each frame of image is within the virtual channel. The determination method is: if X1 + k(X2 - X1) ≤ X ≤ X2 - k(X2 - X1) (k is the proportionality coefficient), it is considered that the target to be measured enters the virtual channel; if not, that is, the target to be measured does not enter the virtual channel. It is found during the experiment that 0 < k < 0.25, and the calculation result is inaccurate when exceeding this value.

[0043] As Figure 3 shown, the targets to be measured in the underwater channel generally swim in the same direction (swim to the left in the figure). If point A in the figure is used as the decision point, the same target to be measured is recognized as entering the virtual channel in two adjacent frames (the current frame and the next frame) of images. If directly counting, it will inevitably cause duplication. Therefore, the system has the same target recognition module ******** for recognizing duplicate targets. The same target recognition module ******** is used to determine whether the targets to be measured entering the virtual channel in two adjacent frames of images are the same target. If the height difference between a certain target to be measured in the next frame of image and all targets to be measured in the current frame of image is the smallest and this minimum value is within the set range, it is recognized that the two are the same target. The set range is determined during the experiment and is generally related to the distance between two virtual detection lines, the proportionality coefficient k in the distance determination module 204, etc.

[0044] The target counting module 206 performs counting based on the results of distance determination and same target discrimination. If the distance determination confirms that the target to be measured enters the virtual channel and the same target discrimination determines that the target to be measured is not the same target, counting is performed.

[0045] For the convenience of calculation, the coordinates involved in the counting processing unit 2 can directly be the pixel coordinates in the image.

[0046] As an implementation manner, the pressure-resistant sealed housing is divided into a counting processing unit sealed housing 4 and a camera sealed housing 5.

[0047] The counting processing unit sealed housing 4 includes a first sealed cylinder 401, a first front end cover 402 and a first rear end cover 403. The first front end cover 402 and the first rear end cover 403 are respectively sealed and fixed at the openings at the front and rear ends of the first sealed cylinder 401. The counting processing unit 2 is内置 in the counting processing unit sealed housing 4. It should be noted that there is an incorrect expression "内置" in the original Chinese text. It should be "内置" which is not a standard Chinese character. I have translated it as "内置" as it is the closest possible meaning. If this is a specific technical term with a different correct translation, please adjust accordingly. Also, there is an unclear "********" in the translation of item

[13] , which should be corrected to the correct content according to the actual situation.

[0048] The camera sealing housing 5 includes a second sealing cylinder 501, a front lens 502, and a second rear cover 503. The front lens 502 and the second rear cover 503 are respectively sealed and fixed at the front and rear openings of the second sealing cylinder 501. The camera unit 1 is built into the camera sealing housing 5 and is close to the front lens 502.

[0049] One end of the transmission cable 7 is sealed and passes through the opening on the first front cover 402 to connect with the counting processing unit 2, and the other end is sealed and passes through the opening on the second rear cover 503 to connect with the camera unit 1.

[0050] By placing the camera unit 1 and the counting processing unit 2 in two sealed housings respectively, the immersion depth of the equipment can be increased to 250m.

[0051] Furthermore, the underwater passage biological throughput counting system also includes a heat sink 3, which is used to dissipate heat from the counting processing unit 2. The heat sink 3 is placed between the counting processing unit 2 and the sealing housing 4 of the counting processing unit.

[0052] Furthermore, the underwater channel biological passage counting system also includes an interface 6, with an interface provided on the first rear end cover 403, which facilitates the connection of the counting processing unit 2 to an external power supply device.

[0053] like Figure 4 As shown, a method for counting the number of organisms passing through underwater passages specifically includes the following steps:

[0054] (1) Camera unit 1 acquires the video stream of the underwater channel over a period of time;

[0055] (2) The virtual channel drawing module 201 sets two virtual detection lines symmetrically on both sides of the vertical center line of each frame image to form a virtual channel. The horizontal coordinates of the two virtual detection lines in the image are X1 and X2, respectively. The total number of the targets to be tested is obtained by counting the targets entering the virtual channel in all video frames.

[0056] (3) The target identification module 202 identifies the target in each frame of the image based on existing target identification methods (such as the methods disclosed in patent CN2021108655020, a marine fish identification method based on TPPTCCNN, and CN2020113193614, an underwater video fish identification method based on neural networks).

[0057] (4) The virtual recognition box drawing and decision point setting module 203 draws a virtual recognition box around the target to be tested and sets decision points based on the recognition box. The decision point can be a point on or inside the recognition box, and the coordinates of the decision point in the image are (X,Y).

[0058] (5) The distance determination module 204 determines whether the target identified in each frame of the image is within the virtual channel. The determination method is as follows: if X1+k(X2-X1)≤X≤X2-k(X2-X1) (k is a proportionality coefficient), then it is considered that the target has not entered the virtual channel; if the distance condition is not met, that is, the target has not entered the virtual channel,

[0059] (6) The same target recognition module 205 determines whether the target to be tested entering the virtual channel in two adjacent frames is the same target. The determination method is as follows: the target to be tested in the underwater channel generally swims in the same direction. If the height difference between the two targets to be tested in two adjacent frames is the smallest and the minimum value is within the set range, then the two are identified as the same target. Specifically:

[0060] (601) Record the height coordinates of the detection boxes corresponding to all targets entering the virtual channel in the previous frame image;

[0061] (602) Calculate the distance between the height coordinates of the detection box corresponding to each target entering the virtual channel in the current frame image and all height coordinates in the previous frame, and determine the minimum distance among them;

[0062] (603) When the minimum distance is less than or equal to the set value, the two corresponding targets to be measured are identified as the same target. The set value is equal to (X2-X1) / 32k (k is the proportional coefficient mentioned in the distance determination module 204).

[0063] (7) The target counting module 206 counts based on the results of distance determination and same target identification. If the distance determination confirms that the target to be tested has entered the virtual channel, and the same target identification determines that the target to be tested is not the same target, then counting is performed.

Claims

1. A system for counting the amount of organisms passing through an underwater passage, characterized in that, It includes a camera unit, a counting processing unit, and a pressure-resistant sealed housing. The camera unit and the counting processing unit are built into the pressure-resistant sealed housing. The camera unit and the counting processing unit are connected. The camera unit is used to obtain the video stream of the underwater channel for a period of time, and the counting processing unit obtains the number of待测 targets passing through the underwater channel during the period based on the video stream; the待测 target is an underwater creature that can be clearly photographed by the underwater camera. The counting processing unit includes a virtual channel drawing module, a待测 target recognition module, a virtual recognition frame drawing and decision point setting module, a distance determination module, the same target recognition module, and a target counting module. The virtual channel drawing module is used to symmetrically set two virtual detection lines on both sides of the vertical midline of each frame of image to form a virtual channel. The abscissas of the two virtual detection lines in the image are X1 and X2 respectively. By counting the待测 targets entering the virtual channel in all video frames, the total amount of the待测 targets is obtained. The待测 target recognition module recognizes the待测 targets in each frame of image based on the existing target recognition method. The virtual recognition frame drawing and decision point setting module is used to draw a virtual recognition frame around the待测 target and set decision points based on the recognition frame. The decision point can be a certain point on or inside the recognition frame, and the coordinates of the decision point in the image are (X, Y). The distance determination module is used to determine whether the recognized待测 target in each frame of image is in the virtual channel. The determination method is: if X1 + k(X2 - X1) ≤ X ≤ X2 - k(X2 - X1), 0 < k < 0.25, it is considered that the待测 target enters the virtual channel; if not, that is, the待测 target does not enter the virtual channel. The same target recognition module is used to determine whether the待测 targets entering the virtual channel in two adjacent frames of images are the same target. If the height difference between a certain待测 target in the next frame of image and all待测 targets in the current frame of image is the smallest and this minimum value is within the set range, it is determined that the two are the same target. The target counting module counts based on the results of distance determination and the discrimination of the same target. If the distance determination confirms that the待测 target enters the virtual channel and the discrimination of the same target determines that the待测 target is not the same target, then counting is performed. The pressure-resistant sealed housing is divided into a counting processing unit sealed housing and a camera sealed housing. The counting processing unit sealed housing includes a first sealed cylinder, a first front end cover, and a first rear end cover. The first front end cover and the first rear end cover are respectively sealed and fixed at the openings at the front and rear ends of the first sealed cylinder. The counting processing unit is built into the counting processing unit sealed housing. The camera sealed housing includes a second sealed cylinder, a front lens, and a second rear end cover. The front lens and the second rear end cover are respectively sealed and fixed at the openings at the front and rear ends of the second sealed cylinder. The camera unit is built into the camera sealed housing and is close to the front lens. One end of the transmission cable is sealed through the opening on the first front end cover and connected to the counting processing unit, and the other end is sealed through the opening on the second rear end cover and connected to the camera unit.

2. The underwater passage biological throughput counting system according to claim 1, characterized in that, The underwater passage biological throughput counting system also includes a heat sink, which is used to dissipate heat from the counting processing unit and is placed between the counting processing unit and the sealed housing of the counting processing unit.

3. The underwater passage biological throughput counting system according to claim 1, characterized in that, The underwater channel biological throughput counting system also includes an interface, with an interface provided on the first rear end cover, which facilitates the connection of the counting processing unit to an external power supply device.

4. A method for counting the amount of organisms passing through an underwater passage based on the underwater passage organism throughput counting system of claim 1, characterized in that, Includes the following steps: (1) The camera unit acquires the video stream of the underwater channel over a period of time; (2) The virtual channel drawing module sets up two virtual detection lines symmetrically on both sides of the vertical center line of each frame to form a virtual channel. The horizontal coordinates of the two virtual detection lines in the image are X1 and X2, respectively. The total number of the targets to be tested is obtained by counting the targets entering the virtual channel in all video frames. (3) The target recognition module identifies the target in each frame of the image based on existing target recognition methods; (4) The virtual recognition box drawing and decision point setting module draws a virtual recognition box around the target to be tested and sets decision points based on the recognition box. The decision point can be a point on or inside the recognition box, and the coordinates of the decision point in the image are (X,Y). (5) The distance determination module determines whether the target identified in each frame of the image is within the virtual channel. The determination method is as follows: if X1+k( X2- X1)≤X≤X2-k( X2- X1), then it is considered that the target has not entered the virtual channel; if the distance condition is not met, that is, the target has not entered the virtual channel, (6) The same target recognition module determines whether the target to be tested entering the virtual channel in two adjacent frames is the same target. The determination method is as follows: the target to be tested in the underwater channel is swimming in the same direction. If the height difference between the two targets to be tested in two adjacent frames is the smallest and the minimum value is within the set range, then the two are considered to be the same target. (7) The target counting module counts based on the results of distance determination and same target identification. If the distance determination confirms that the target to be tested has entered the virtual channel, and the same target identification determines that the target to be tested is not the same target, then counting is performed.

5. The underwater passage biological throughput counting method according to claim 4, characterized in that, Step (6) specifically involves: (601) Record the height coordinates of the judgment points corresponding to all the targets to be tested that entered the virtual channel in the previous frame image; (602) Calculate the distance between the height coordinates of the judgment point corresponding to each target entering the virtual channel in the current frame image and the height coordinates of the judgment points corresponding to all targets in the previous frame image, and determine the minimum distance among them. (603) When the minimum distance is less than or equal to the set value, the two corresponding targets are identified as the same target. The set value is equal to (X2-X1) / 32k.

Images