A machine vision-based fish head sawing device and method

Abstract

The application provides a sawing fish head device and method based on machine vision, which comprises a band sawing machine, a six-axis mechanical arm, a gantry and a fish box, the fish to be cut is placed in the fish box, the fish box is provided with the six-axis mechanical arm on the side, the fish box is provided with a 3D industrial camera and a 2D industrial camera above, the 3D industrial camera and the 2D industrial camera are arranged on the gantry, and the fish box is provided with the band sawing machine at the rear; the 3D industrial camera identifies the fish on the uppermost part of the fish box and transmits position and angle information to the six-axis mechanical arm, and the six-axis mechanical arm grasps the fish on the uppermost part; the 2D industrial camera identifies the curve at the joint of the fish head and the fish body and converts the cutting track to the six-axis mechanical arm; and the six-axis mechanical arm sends the fish to the band sawing machine and cuts off the fish head according to the cutting track. The application can not only eliminate safety hazards, but also can perform high-precision cutting to prevent under-cutting from causing fish meat to mix with gill impurities or over-cutting from causing fish meat to be wasted.

Description

Technical Field

[0001] This invention relates to the field of fish processing technology, and more particularly to a machine vision-based device and method for sawing fish heads. Background Technology

[0002] In the general process of fish processing, removing the head and tail is a key step in the pre-processing of fish.

[0003] Currently, fish head cutting is mostly done manually or with fish head cutting equipment. Manual cutting means that the hands are close to the cutting device, which is not only labor-intensive but also poses a significant safety hazard. Fish head cutting equipment usually cuts in a uniform position, which can lead to undercutting of larger fish, resulting in some fish heads not being cut off and leaving gills and other impurities behind. Or it can overcut smaller fish, resulting in the removal of part of the fish body and waste of fish meat. Summary of the Invention

[0004] To address the safety hazards and waste inherent in existing fish head cutting methods, this invention provides a machine vision-based fish head sawing device and method. This invention primarily provides a fully automated, precise fish head cutting device and method based on machine vision, which not only eliminates safety hazards but also enables high-precision cutting to prevent undercutting that could result in gill contamination or overcutting that could lead to waste.

[0005] The technical means employed in this invention are as follows:

[0006] A machine vision-based fish head sawing device includes a band saw, a six-axis robotic arm, a gantry frame, and a fish box. The fish box contains fish to be cut. The six-axis robotic arm is mounted on the side of the fish box. A 3D industrial camera and a 2D industrial camera are mounted on the top of the fish box. The 3D industrial camera and the 2D industrial camera are mounted on the gantry frame. The band saw is mounted at the rear of the fish box.

[0007] The 3D industrial camera identifies the fish at the top of the fish tank and transmits its position and angle information to the six-axis robotic arm, which then grabs the fish at the top. The 2D industrial camera identifies the curve at the connection between the fish head and body and converts it into a cutting trajectory, which is then transmitted to the six-axis robotic arm. The six-axis robotic arm then delivers the fish to the band saw, which cuts off the fish head according to the cutting trajectory.

[0008] Furthermore, the fish box includes a fish head box, a fish body box, and a whole fish box arranged in sequence. The fish head box is in contact with the cutting table of the band saw. The cutting table is provided with a fish head slide that is inclined towards the fish head box. The fish head slide is provided with a slit for placing the saw blade. A saw blade protective cover is provided on one side of the saw blade. A handle is provided on the saw blade protective cover. A limit block is provided on the other side of the saw blade. A slit is opened on the limit block for the saw blade to pass through.

[0009] Furthermore, the fish box is mounted on a support frame equipped with two wheels. The front of the support frame is connected to the fish box via a locking device, and the middle of the support frame is connected to the fish box via a fixed support.

[0010] Furthermore, the locking device includes two clamping plates, which are mounted on the mounting frame. A rotating handle and a locking piece are provided between the two clamping plates. The notch of the locking piece contacts the central shaft, which is located at the front of the fish box.

[0011] The present invention also provides a method for sawing fish heads based on machine vision, implemented based on any of the above-mentioned machine vision-based fish head sawing devices, comprising the following steps:

[0012] S1. The point cloud data of the fish stacked in the fish box is acquired by a 3D industrial camera and analyzed to determine the optimal grasping target. The position and pose parameters of the fish are then transmitted to the six-axis robotic arm to guide the six-axis robotic arm to grasp the whole fish.

[0013] The S2 and 2D industrial cameras identify the fish head of the whole fish captured by the S1 and plan the cutting path. The coordinate information of the cutting line and the cutting path is uploaded to the PC. The PC then sends the coordinate information of the cutting path to the six-axis robotic arm.

[0014] S3. The six-axis robotic arm drives the robotic hand to grab the whole fish to the position of the saw blade, and moves according to the coordinate information of the cutting path planned in S2 to perform the fish sawing operation.

[0015] S4. After the operation is completed, the fish body is placed into the fish body box by the robotic arm, and the fish head slides down the slide into the fish head box.

[0016] S5. Perform S1-S4 sequentially on all fish in the fish tank until all fish in the fish tank have been cut.

[0017] Furthermore, the specific steps of S1 are as follows:

[0018] S11. Obtain whole fish point cloud data through a 3D industrial camera, and preprocess the point cloud data using a pass-through filtering and uniform downsampling algorithm to reduce the amount of computation, and obtain the preprocessed whole fish point cloud.

[0019] S12. Use a clustering segmentation algorithm to segment the preprocessed whole fish point cloud into instances and filter out the fish body point cloud that is incomplete due to occlusion.

[0020] S13. For the fish body point cloud processed in S12, combine the spatial feature information of each fish body point cloud and calculate the score according to the formula. The target with the higher score is the priority target to be captured.

[0021]

[0022] Where d represents the distance from the centroid of the fish body point cloud to the origin of the 3D industrial camera coordinate system, θ is the angle between the normal vector of the fish body point cloud and the Z-axis of the camera coordinate system, ε is a small number used to reduce the excessive weight when θ is very small and to prevent division by zero error; ω1 and ω2 are weight coefficients.

[0023] S15. Calculate the optimal position and pose parameters of the target fish body;

[0024] S16. Transmit the optimal target fish body position coordinates and attitude parameters to the robot controller to guide the six-axis robotic arm to grasp the target fish body.

[0025] Furthermore, the specific steps of S2 are as follows:

[0026] S21. Calculate the transformation relationship between image coordinates and robot coordinates. Use a calibration board to establish the transformation relationship between image coordinates and robot coordinates. Manually obtain the teaching point in the robot coordinate system. The camera captures images to obtain the image coordinates corresponding to the teaching point. Calculate the transformation matrix M between image coordinates and robot coordinates.

[0027] S22. Train the fish head cutting region network to obtain the trained network;

[0028] S23. Use a six-axis robotic arm to grab the whole fish and send it to a 2D camera for shooting. After training the network in S22, identify the cutting area of ​​the fish head, extract the edge line of the fish head, and refine the coordinates to form the cutting path. Then, convert it into the robot's motion trajectory through the calibrated transformation matrix M and send it to the six-axis robotic arm for cutting.

[0029] Furthermore, the specific steps of S22 are as follows:

[0030] S221. Prepare several whole fish image samples and mark the edge area of ​​the fish head. After processing, export two file formats: jpg and json.

[0031] S222. Use a program in PyCharm to convert the annotated JSON file into a txt file required for instance splitting;

[0032] S223. Divide the txt file into a training set of 70% and a validation set of 30%. Add an image directory and a label directory to the training set and the validation set respectively. Put the jpg file into the image directory and the txt file into the label directory.

[0033] S224. Put the entire dataset into the YOLO algorithm in PyCharm, use the YOLO-v8 seg instance segmentation algorithm to train the training set, and filter out the part of the segmentation result that belongs to the fish head cutting area according to the category and bounding box information of the target object. By continuously optimizing the structure, an optimized training model is obtained.

[0034] S225. Test the model's performance using test set images to obtain the trained network.

[0035] Compared with the prior art, the present invention has the following advantages:

[0036] Compared to the common method of cutting fish heads at a fixed position, this invention can automatically identify the boundaries of the fish head and cut precisely along an arc, which can prevent undercutting from causing fish meat to mix with gill impurities or overcutting from wasting fish meat.

[0037] The use of robotic arms to automatically catch fish and complete the entire cutting process greatly reduces the need for manpower and is safe and convenient.

[0038] The slide device allows fish to slide directly into the fish head box, and it is easy to disassemble and clean. The fish box makes it easy to process the fish body and head, effectively reducing labor intensity. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of the assembly structure of the present invention.

[0041] Figure 2 This is a structural diagram of the band saw of the present invention.

[0042] Figure 3 This is a structural diagram of the six-axis robotic arm of the present invention.

[0043] Figure 4 This is a structural diagram of the gantry and industrial camera of the present invention.

[0044] Figure 5 This is a structural diagram of the fish tank of the present invention.

[0045] Figure 6 for Figure 5 Enlarged view of the locking device at point A.

[0046] Figure 7 for Figure 2Enlarged view of the saw blade teeth at point B.

[0047] Figure 8 This is a structural diagram of a robotic arm mounted on a six-axis robotic arm.

[0048] Figure 9 A diagram showing the cut area of ​​the fish head.

[0049] Figure 10 This is a flowchart of the method of the present invention.

[0050] In the diagram: 1. Band saw, 2. Six-axis robotic arm, 3. Gantry, 4. Fish box, 11. Saw blade, 12. Protective cover, 13. Slide rail, 14. Control box, 15. Cutting table, 16. Limit block, 17. Fastening screw, 21. Table, 22. Robotic arm, 31. 2D industrial camera, 32. 3D industrial camera, 33. Gantry rail, 34. Sliding device, 41. Whole fish box, 42. Fish body box, 43. Fish head box, 44. Fixed support, 45. Mounting frame, 46. Locking device, 461. Rotating handle, 462. Set screw, 463. Clamping plate, 464. Central shaft, 465. Clamping plate, 221. Mechanical claw, 222. Anti-slip pad, 223. Slide bar, 224. Fixed block. Detailed Implementation

[0051] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0054] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0055] In the description of this invention, it should be understood that the orientation or positional relationship indicated by directional terms such as "front, back, up, down, left, right", "horizontal, vertical, horizontal" and "top, bottom" is generally based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing this invention and simplifying the description. Unless otherwise stated, these directional terms do not indicate or imply that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the scope of protection of this invention. The directional terms "inner" and "outer" refer to the inner and outer contours relative to the outline of each component itself.

[0056] like Figure 1-8 As shown, this invention provides a fully automatic fish head-cutting device based on machine vision, consisting of four parts: a band saw 1, a six-axis robotic arm 2, a gantry 3 equipped with a 3D industrial camera 32 and a 2D industrial camera 31, and a fish box 4. The principle is as follows: multiple fish to be cut are placed in a whole fish box 41 within the fish box 4. The 3D industrial camera 32 identifies the topmost fish and transmits its position and angle information to the six-axis robotic arm 2 for grasping. The 2D industrial camera 31 identifies the curve at the connection between the fish head and body and converts it into a cutting trajectory, which is then transmitted to the six-axis robotic arm 2. The six-axis robotic arm delivers the fish to the band saw to cut off the fish head according to the trajectory. Then, the six-axis robotic arm 2 places the fish body into the fish body box 42, while the fish head falls into the fish head box 43 along the slide 13.

[0057] The band saw 1 is used to cut fish heads, and the protective cover 12 is used to cover the saw blade 11 to prevent accidental cutting. A limiting block 16 for controlling the position of the saw blade is installed on the other side of the band saw frame. The limiting block 16 has a slit that allows the saw blade to pass through and is installed on one side of the band saw body by fastening screws 17. Depending on the size of the fish, this limiting block can be moved up and down to adjust the effective size of the blade to ensure safety.

[0058] The fish box 4 is divided into three parts by a partition: a fish head box 43 connected to the cutting table 15, a middle fish body box 42, and an outer whole fish box 41. The three parts respectively hold the fish head, fish body, and whole fish. The fish head box 43 of the fish box 4 is connected to the cutting table 15 of the band saw machine 1, and the slide rail 13 is installed on the cutting table 15 and fixed.

[0059] Optionally, the fish box 4 is fixed by a mounting frame 45 with two wheels. A locking device 46 is installed on one side of the mounting frame 45, which can fix the fish box 4 horizontally on the mounting frame 45. When the fish head box is full of fish heads, the fish box can be pushed to tilt the fish heads by unlocking the locking device 46.

[0060] The six-axis robotic arm 2 is installed on one side of the fish box 4 for grabbing the fish, cutting off the fish head, and sending the fish body back to the fish body box 42.

[0061] The gantry frame 3 is installed above the whole fish tank 41, and a 3D industrial camera 31 and a 2D industrial camera 32 are mounted on it. The 3D industrial camera is used to identify the position and state of the uppermost fish so that it can be grasped by a six-axis robotic arm; the 2D industrial camera is used to identify the position of the fish head and plan the cutting path to cut precisely along the edge of the fish head.

[0062] The specific work process is as follows:

[0063] S1: A six-axis robotic arm based on 3D vision for fish catching. Point cloud data of stacked fish in a fish tank is acquired using an RGB-D camera and preprocessed. Clustering segmentation algorithms are used to segment the preprocessed fish point cloud into instances. Then, local concavity and convexity analysis is performed on the segmented, adhered fish point cloud to determine the optimal catching target. The position and pose parameters of the fish are detected to guide the six-axis robotic arm to catch the whole fish and send it to a 2D industrial camera for cutting path planning.

[0064] The point cloud data of the fish stacked in the fish box is acquired by an RGB-D camera and analyzed to determine the optimal target for grasping. The position and pose parameters of the fish are then transmitted to a six-axis robotic arm, which is guided to grasp the whole fish and send it to a 2D industrial camera for cutting path planning.

[0065] S2: Cutting path planning based on a 2D industrial camera. Before production begins, the transformation relationship M between image coordinates and robot coordinates is calculated, and a fully convolutional neural network is trained for recognizing the boundary line between the fish head and body. During production, the trained neural network is used to identify the edge line of the boundary line between the fish head and body, and the transformation is converted to the robot coordinate system according to the above transformation relationship M to form the cutting path. Finally, the fish head is cut off by a band saw fed by a robot arm according to the cutting path.

[0066] The 2D camera 31 identifies the head of the target fish and plans the cutting path by identifying the cutting area through an algorithm. The coordinate information of the cutting line and the cutting path are uploaded to the PC, and the PC transmits the coordinate information to the six-axis robotic arm 2.

[0067] S3: The six-axis robotic arm 2 drives the robotic hand 22 to grab the fish to the position of the saw blade 11, and moves according to the planned path coordinates to perform the fish sawing operation.

[0068] S4: After the operation is completed, the fish body is placed into the fish body box 42 by the robotic arm 2, and the fish head slides down the slide 13 into the fish head box 43.

[0069] S5: After all the fish in the whole fish box 41 have been processed, the support frame 45 is equipped with wheels, which can push the fish box 4 to a designated position to process the fish head and body. Before processing, the rotating handle 461 is operated to drive the clamping piece 465 to rotate 90 degrees clockwise and disengage it from the central shaft 464. At this time, the fish box 4 and the support frame 45 are in a semi-separated state, and the fish box 4 can be lifted to rotate around the fixed support 44. The fish body box 42, the fixed support 44, and the fish head box 43 together form a set of lever mechanisms to reduce the workload of the workers.

[0070] The specific process of the S1 six-axis robotic arm catching fish based on 3D vision is as follows:

[0071] S11: The point cloud data of the whole fish is acquired through an RGB-D camera, and the point cloud data is preprocessed using pass-through filtering and uniform downsampling algorithms to reduce the amount of computation.

[0072] S12: Use a clustering segmentation algorithm to segment the preprocessed whole fish point cloud into instances, and filter out the fish body point cloud that is incomplete due to occlusion.

[0073] S13: Combine the spatial feature information of each fish body point cloud, calculate the score according to the formula, and prioritize the target with the highest score.

[0074]

[0075] Where d represents the distance from the centroid of the fish body point cloud to the origin of the camera coordinate system, θ is the angle between the normal vector of the fish body point cloud and the Z-axis of the camera coordinate system, ε is a small number used to reduce the excessive weight when θ is very small and to prevent division by zero errors; ω1 and ω2 are weight coefficients.

[0076] S15: Calculate the optimal position and pose parameters of the target fish body.

[0077] S16: The acquired grasping coordinates and posture parameters are transmitted to the robot controller to guide the six-axis robotic arm to grasp the fish body target.

[0078] The specific process of S2 cutting path planning based on a 2D industrial camera is as follows:

[0079] S21: Calculate the transformation relationship between image coordinates and robot coordinates. Establish the transformation relationship between image coordinates and robot coordinates using a calibration board. Manually acquire the teaching point in the robot coordinate system, and use the camera to acquire the image coordinates corresponding to the teaching point. Calculate the transformation matrix M from image coordinates to robot coordinates.

[0080] S22: Network Training

[0081] S221: Prepare over 200 whole fish image samples and label the edge area of ​​the fish head, such as... Figure 9 As shown, after processing, it will export two file formats: JPG and JSON.

[0082] S222: Use a program in PyCharm to convert the annotated JSON file into a TXT file needed for instance splitting.

[0083] S223: Divide the txt file into a training set of 70% and a validation set of 30%. Add an image directory and a label directory to the training set and validation set respectively. Put the jpg file into the image directory and the txt file into the label directory.

[0084] S224: Put the entire dataset into PyCharm's YOLO algorithm and train the training set using the YOLO-v8 seg instance segmentation algorithm.

[0085] S23: Cutting path planning.

[0086] A six-axis robotic arm grabs a fish and sends it to a 2D camera for shooting. The network trained by S2 identifies the cutting area of ​​the fish head, extracts the edge line of the fish head, and refines the coordinates to form a cutting path. The path is then converted into a robot motion trajectory through a calibrated transformation matrix M and transmitted to the six-axis robotic arm for cutting.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A machine vision-based fish head sawing device, characterized in that: The system includes a band saw (1), a six-axis robotic arm (2), a gantry (3), and a fish box (4). The fish box (4) contains fish to be cut. The six-axis robotic arm (2) is located on the side of the fish box (4). A 3D industrial camera (32) and a 2D industrial camera (31) are located above the fish box (4). The 3D industrial camera (32) and the 2D industrial camera (31) are located on the gantry (3). The band saw (1) is located at the rear of the fish box (4). The 3D industrial camera (32) identifies the fish at the top of the fish box (4) and transmits the position and angle information to the six-axis robotic arm (2). The six-axis robotic arm (2) grabs the fish at the top. The 2D industrial camera (31) identifies the curve at the connection between the fish head and the body and converts it into a cutting trajectory, which is then transmitted to the six-axis robotic arm (2). The six-axis robotic arm (2) sends the fish to the band saw (1) and cuts off the fish head according to the cutting trajectory. The working method of the machine vision-based fish head sawing device is as follows: S1. The point cloud data of the fish stacked in the fish box (4) is obtained by the 3D industrial camera (32) and analyzed to determine the optimal grasping target. The position and pose parameters of the fish are transmitted to the six-axis robotic arm (2) to guide the six-axis robotic arm (2) to grasp the whole fish. S11. Obtain whole fish point cloud data through 3D industrial camera (32), and preprocess the point cloud data using pass-through filtering and uniform downsampling algorithm to reduce the amount of computation, and obtain the preprocessed whole fish point cloud; S12. Use a clustering segmentation algorithm to segment the preprocessed whole fish point cloud into instances and filter out the fish body point cloud that is incomplete due to occlusion. S13. For the fish body point cloud processed in S12, combine the spatial feature information of each fish body point cloud and calculate the score according to the formula. The target with the higher score is the priority target to be captured. in, This represents the distance from the centroid of the fish body point cloud to the origin of the coordinate system of the 3D industrial camera (32). The angle between the normal vector of the fish's body point cloud and the Z-axis of the camera coordinate system is denoted as . It is a small number used to reduce when When the value is very small, it can lead to excessive weighting, and it also helps prevent division by zero errors. and These are the weighting coefficients; S15. Calculate the optimal position and pose parameters of the target fish body; S16. Transmit the optimal target fish body position coordinates and attitude parameters to the robot controller to guide the six-axis robotic arm (2) to grab the fish body target; S2, 2D industrial camera (31) identifies the head of the whole fish captured by S1 and plans the cutting path. The coordinate information of the cutting line and the cutting path is uploaded to the PC. The PC sends the coordinate information of the cutting path to the six-axis robotic arm (2). S21. Calculate the transformation relationship between image coordinates and robot coordinates. Use a calibration board to establish the transformation relationship between image coordinates and robot coordinates. Manually obtain the teaching point in the robot coordinate system. The camera captures images to obtain the image coordinates corresponding to the teaching point. Calculate the transformation matrix M between image coordinates and robot coordinates. S22. Train the fish head cutting region network to obtain the trained network; S23. Use a six-axis robotic arm (2) to grab the whole fish and send it to a 2D camera for shooting. After training the network in S22, identify the cutting area of ​​the fish head, extract the edge line of the fish head, and refine the coordinates to form the cutting path. Then, convert it into the robot motion trajectory through the calibrated transformation matrix M and send it to the six-axis robotic arm (2) for cutting. S3. The six-axis robotic arm (2) drives the robotic hand to grab the whole fish to the position of the saw blade (11), and moves according to the coordinate information of the cutting path planned in S2 to perform the fish sawing operation. S4. After the operation is completed, the fish body is placed into the fish body box (42) by the robotic arm, and the fish head slides down the slide (13) into the fish head box (43); S5. Perform S1-S4 sequentially on all fish in the fish box (4) until all fish in the fish box (4) have been cut.

2. The machine vision-based fish head sawing device according to claim 1, characterized in that, The fish box (4) includes a fish head box (43), a fish body box (42) and a whole fish box (41) arranged in sequence. The fish head box (43) is in contact with the cutting table (15) of the band saw (1). The cutting table (15) is provided with a fish head slide (13) that is inclined toward the fish head box (43). The fish head slide (13) is provided with a slit for placing the saw blade (11). A saw blade guard (12) is provided on one side of the saw blade (11). A handle is provided on the saw blade guard (12). A limit block (16) is provided on the other side of the saw blade (11). A slit is opened on the limit block (16) for the saw blade (11) to pass through.

3. The machine vision-based fish head sawing device according to claim 2, characterized in that, The fish box (4) is mounted on a two-wheeled frame (45). The front of the frame (45) is connected to the fish box (4) via a locking device (46), and the middle of the frame (45) is connected to the fish box (4) via a fixed support (44).

4. The machine vision-based fish head sawing device according to claim 3, characterized in that, The locking device (46) includes two clamps (463), which are mounted on the mounting frame (45). A rotating handle (461) and a locking piece (465) are provided between the two clamps (463). The notch of the locking piece (465) is in contact with the central shaft (464), which is located at the front of the fish box (4).

5. The machine vision-based fish head sawing device according to claim 1, characterized in that, The specific steps of S22 are as follows: S221. Prepare several whole fish image samples and mark the edge area of ​​the fish head. After processing, export two file formats: jpg and json. S222. Use a program in PyCharm to convert the annotated JSON file into a txt file required for instance splitting; S223. Divide the txt file into a training set of 70% and a validation set of 30%. Add an image directory and a label directory to the training set and the validation set respectively. Put the jpg file into the image directory and the txt file into the label directory. S224. Put the entire dataset into the YOLO algorithm in PyCharm and train the training set using the YOLO-v8 seg instance segmentation algorithm.

Images