A fish recognition method and device based on deep learning

Abstract

The embodiment of the present disclosure provides a fish recognition method and device based on deep learning. The device comprises: a collection table provided with a mounting port; a placing plate which can pass through the mounting port up and down, when the first surface of the placing plate is lowered below the collection table, it is used for obtaining fish to be collected, when the first surface of the placing plate is raised to be flush with the first surface of the collection table, it is used for placing the fish to be collected on the collection area of the collection unit; a driving assembly which is rotatably arranged on the side wall of the mounting port of the collection table and rotates along the circumference of the mounting port; a collection assembly which is arranged on the driving assembly, the collection assembly can rotate along the circumference of the mounting port with the driving assembly, and the collection unit of the collection assembly is above the first surface of the placing plate at any preset position, and multi-angle image collection of fish is performed. The device of the present disclosure has simple structure, convenient collection and higher accuracy in collecting features of each part of fish.

Description

Technical Field

[0001] This disclosure relates to the field of fish identification technology, and in particular to a fish identification method and apparatus based on deep learning. Background Technology

[0002] Existing fish resource surveys mainly employ field survey methods. After fish are caught in wild waters, the species of fish are identified primarily through visual inspection and experience by field survey personnel, followed by verbal counting and registration. The length of fish is mainly measured by visual estimation.

[0003] Traditional fish resource survey methods rely on visual inspection and experience to identify fish species, demanding extremely high levels of expertise from surveyors. Inexperienced surveyors often struggle to identify rare or endangered fish, leading to frequent errors. Existing technologies utilize pre-established fish information databases, comparing images of newly collected fish by their characteristic features against a fish model database. While this improves accuracy, image-based identification still carries inherent errors, and measuring fish length and taking photographs require separate operations, resulting in low efficiency. Currently, a data acquisition device capable of both acquiring fish images and measuring fish length is lacking. Summary of the Invention

[0004] The purpose of this disclosure is to provide a fish identification method and apparatus based on deep learning, thereby solving the aforementioned problems existing in the prior art.

[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of this disclosure are as follows:

[0006] This disclosure provides, in one aspect, a fish identification device based on deep learning, the device comprising:

[0007] The data acquisition station is equipped with an installation port;

[0008] The placement plate can be raised and lowered through the mounting port. When the first surface of the placement plate is lowered to below the acquisition platform, it is used to acquire the fish image to be acquired. When the first surface of the placement plate is raised to be flush with the first surface of the acquisition platform, it is used to place the fish image to be acquired in the acquisition area of ​​the acquisition unit.

[0009] A drive component is disposed at the mounting port of the acquisition station and can rotate circumferentially along the mounting port;

[0010] A data acquisition component is disposed on the driving component. The data acquisition component can rotate circumferentially along the mounting opening with the driving component. The data acquisition unit of the data acquisition component is above the first surface of the placement plate. With the center of the placement plate as the center, the data acquisition unit rotates in a semi-circular manner with the distance between the data acquisition unit and the center as the radius, so that the data acquisition unit can acquire images of fish on the placement plate from multiple angles.

[0011] Optionally, the first surface or sidewall of the mounting port is recessed to form a mounting groove;

[0012] The driving component includes:

[0013] The first motor is located on one side of the data acquisition platform;

[0014] The drive gear has its shaft connected to the output end of the first motor;

[0015] The driven gear ring is rotatably sleeved on the outside of the bearing and meshes with the driving gear. The bearing and the driven gear ring are disposed in the mounting groove.

[0016] Optionally, the acquisition component includes:

[0017] The annular slide rail is semi-circular in shape, with its two ends spanning the radial sides of the driven toothed ring, and has a hollowed-out slide track in the middle.

[0018] A slider is slidably disposed in the slide rail, and a mounting hole is provided in the central area thereof;

[0019] An insert plate is movably inserted through the mounting hole, and the acquisition unit is provided at one end facing the placement plate.

[0020] A rotating component is disposed on the driven gear ring and connected to one end of the insert plate via a connector. When the rotating component rotates, it drives the acquisition unit to rotate around the center of the placement plate as the center, with the distance between the acquisition unit and the center as the radius.

[0021] Optionally, the rotating assembly includes:

[0022] A vertical plate, one end of which is disposed on the driven toothed ring and forms a connection point, wherein the line connecting the connection point and the center of the placement plate is perpendicular to the line connecting the two ends of the annular slide rail;

[0023] The mounting bracket is attached to the side of the vertical plate facing the annular slide rail, with its two ends close to the two ends of the vertical plate respectively;

[0024] The screw is rotatably connected at both ends to the two ends of the mounting bracket.

[0025] A meshing block is fitted onto the outer surface of the screw and threadedly connected to the screw. One side of the meshing block is in contact with the side of the vertical plate facing the annular slide rail, and the other side is connected to one end of the insert plate through a connector.

[0026] The second motor is located on the side of the vertical plate away from the annular slide rail, near the end of the acquisition platform, and its output end passes through the vertical plate and is connected to the end of the mounting frame near the acquisition platform.

[0027] The third motor is located at the other end of the mounting frame away from the acquisition platform, and its output shaft is connected to one end of the screw through the other end of the mounting frame.

[0028] Optionally, the device further includes two ranging components, respectively disposed at the bottom of both ends of the annular slide rail, used to measure the size of the fish on the placing plate when the placing plate rises to the level of its surface with the surface of the collection platform.

[0029] Optionally, the ranging component includes:

[0030] The second electric push rod is located at the bottom of one end of the annular slide rail, and its telescopic end passes through the bottom of the annular slide rail and moves telescopically toward the acquisition platform.

[0031] A clamp is provided at the telescopic end of the second electric push rod;

[0032] One of the ranging components has a rangefinder at both ends of its clamping plate. The clamping plates of the two ranging components are used to hold fish, and the rangefinder measures the size of the fish between the two clamping plates.

[0033] Optionally, the device further includes:

[0034] A pushing component, located on one side of the collection platform, is used to push the fish on the placement plate out of the collection platform.

[0035] Optionally, the pushing component includes:

[0036] A base is provided on the acquisition station;

[0037] The third electric push rod has one end fixed to the base and the other retractable end facing the acquisition platform;

[0038] A push plate is located at the telescopic end of the third electric push rod.

[0039] Optionally, the device further includes: a control unit and an identification unit, wherein the control unit is connected to the identification unit, the acquisition unit, the drive assembly, and each motor of the acquisition assembly, respectively;

[0040] The recognition unit is used to identify the species of fish in the acquired image, wherein the recognition unit stores a trained fish recognition model;

[0041] The control unit is used to drive the start and stop of each motor, and to control the recognition unit to acquire and recognize the fish images collected by the acquisition unit.

[0042] Another aspect of this disclosure provides a fish identification method based on deep learning, the method comprising:

[0043] Fish images are acquired using the deep learning-based fish recognition device described above;

[0044] The fish image is input into a pre-built fish recognition model to obtain the fish recognition result; wherein, the fish recognition model includes models trained based on Fast RCNN;

[0045] The training process for the fish recognition model includes:

[0046] Acquire fish images;

[0047] The data set includes the types of fish images and the various parts of the fish. The various parts of the fish include one or more of the following: gill openings, gill covers, scales, dorsal fin, pectoral fin, and caudal fin.

[0048] The dataset was divided into training, validation, and test sets to train Fast R-CNN and build a fish recognition model.

[0049] Another aspect of this disclosure provides a fish identification device based on deep learning, the device comprising:

[0050] The acquisition module is used to acquire fish images using the aforementioned device;

[0051] The recognition module is used to input the fish image into a pre-built fish recognition model to obtain the fish recognition result; wherein, the fish recognition model includes one trained on Fast RCNN;

[0052] The training process for the fish recognition model includes:

[0053] Acquire fish images;

[0054] The data set includes the types of fish images and the various parts of the fish. The various parts of the fish include one or more of the following: gill openings, gill covers, scales, dorsal fin, pectoral fin, and caudal fin.

[0055] The dataset was divided into training, validation, and test sets to train Fast R-CNN and build a fish recognition model.

[0056] The beneficial effects of the embodiments disclosed herein are:

[0057] Compared to existing technologies, the apparatus of this disclosure uses a feeding stage to feed fish samples onto a placement plate. Then, through the cooperation of a driving component, a collection component, and a collection unit, the fish samples are photographed from multiple angles, capturing detailed features of various parts of the fish and measuring the sample length. After transmitting the collected data to a database, the sample is sent out via a feeding stage, improving the efficiency of fish sample collection. Compared to manually classifying and collecting feature information from samples one by one, this method is more convenient, faster, and easier for database entry. Attached Figure Description

[0058] Figure 1 A schematic diagram of the structure of a deep learning-based fish identification device provided in an embodiment of this disclosure;

[0059] Figure 2 This is a schematic diagram of the bottom structure of the data acquisition platform of the deep learning-based fish identification device according to an embodiment of this disclosure;

[0060] Figure 3 This is a schematic diagram of the driving component structure of a deep learning-based fish identification device according to an embodiment of the present disclosure;

[0061] Figure 4 This is a schematic diagram of the acquisition component structure of the deep learning-based fish identification device according to an embodiment of the present disclosure;

[0062] Figure 5 This is a top view of a deep learning-based fish identification device according to an embodiment of the present disclosure.

[0063] In the picture:

[0064] 1. Acquisition platform; 11. Feeding platform; 12. Feeding platform; 2. Placement plate; 21. First electric push rod; 3. Drive assembly; 31. First motor; 32. Drive gear; 33. Driven gear ring; 4. Acquisition assembly; 41. Circular slide rail; 42. Second electric push rod; 43. Clamping plate; 44. Slider; 45. Insertion plate; 46. Acquisition unit; 47. Vertical plate; 48. Second motor; 49. Mounting bracket; 410. Screw; 411. Third motor; 412. Meshing block; 413. Rangefinder; 5. Third electric push rod; 51. Push plate. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the embodiments of this disclosure will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of this disclosure and are not intended to limit the embodiments of this disclosure.

[0066] like Figure 1As shown in the figure, this disclosure proposes a fish identification device based on deep learning, the device comprising:

[0067] The acquisition platform 1 is provided with an installation port; the placement plate 2 can be raised and lowered through the installation port. When the first surface of the placement plate 2 is lowered to below the second surface of the acquisition platform 1, it is used to acquire the fish image to be acquired. When the first surface of the placement plate 2 is raised to be flush with the first surface of the acquisition platform 1, it is used to place the fish image to be acquired in the acquisition area of ​​the acquisition unit 46; the driving component 3 is rotatably disposed on the side wall of the installation port of the acquisition platform 1 and can rotate circumferentially along the installation port to achieve rotation around the placement plate 2; the acquisition component 4 is disposed on the driving component 3 and can rotate circumferentially along the installation port with the driving component 3. The acquisition unit 46 of the acquisition component 4 is above the first surface of the placement plate 2 and rotates semi-circularly with the center of the placement plate 2 as the center, so that the acquisition unit 46 can acquire the image of the fish on the placement plate 2 from multiple angles.

[0068] The acquisition platform 1 of this embodiment has a support portion. For ease of operation, the size of the acquisition platform 1 can be set according to actual needs, and the mounting opening can be provided in the central area of ​​the acquisition platform 1. The mounting opening can be circular, and the diameter of the placement plate 2 is slightly smaller than the diameter of the mounting opening. The two cooperate to allow the placement plate 2 to be raised and lowered through the mounting opening. Figure 2As shown, the second surface of the acquisition platform 1 is provided with first electric push rods 21 on both sides of the radial direction of the mounting opening. Depending on the thickness of the placement plate 2, the fixed end of the first electric push rod 21 can be fixed to the second surface of the acquisition platform 1, or it can be embedded in the second surface of the acquisition platform 1, ensuring that the first surface of the placement plate 2 is flush with the first surface of the acquisition platform 1 after it rises. The telescopic end of the first electric push rod 21 can be connected to the lower surface of the placement plate 2, or it can be connected to a connector on the lower surface of the placement plate 2. There can be two connectors, respectively located on the radial edges of the placement plate 2; the connector can also be a long plate located on the second surface of the placement plate 2 along its radial direction, with the portion extending beyond the second surface of the placement plate 2 for connection to the telescopic end of the first electric push rod 21. Thus, the first electric push rod 21 can drive the placement plate 2 to rise or fall. When the placement plate 2 falls below the second surface of the acquisition platform 1, the fish sample to be captured is obtained through the gap between the placement plate 2 and the acquisition platform 1. When the placement plate 2 rises until its first surface is flush with the first surface of the acquisition platform 1, the fish sample to be captured is placed in the acquisition area of ​​the acquisition unit 46. In this embodiment, the first surface or sidewall of the mounting port can be recessed inward to form a mounting groove, and the drive assembly 3 is disposed in the mounting groove. The center of the driving component 3 coincides with the center of the placement plate 2. When the driving component 3 rotates circumferentially along the mounting opening, it drives the acquisition component 4 to rotate horizontally around the placement plate 2 by 360°. When the acquisition component 4 rotates, it drives the acquisition unit 46 to rotate vertically in a 180° semicircle above the placement plate 2 with the center of the placement plate 2 as the center and the distance between the acquisition unit 46 and the center as the radius. This allows the acquisition unit 46 to perform multi-angle image acquisition of the fish on the placement plate 2. The acquisition unit 46 can be a camera or webcam or other acquisition device. The first surface is away from the ground, i.e., the upper surface; the second surface faces the ground, i.e., the lower surface.

[0069] Because the fish need to be transported to the collection location, and their length and images need to be captured, the fish need to be kept still. This can be achieved by chilling the fish to put them to rest, securing them with cable ties, or manually holding the fish to the placement plate 2 during data collection. The method of using the device includes: lowering the placement plate 2 below the collection platform 1, acquiring the image of the fish through the gap between the collection platform 1 and the placement plate 2; raising the placement plate 2 until its first surface is flush with the first surface of the collection platform 1; driving the drive assembly 3 to rotate circumferentially around the placement plate 2 to a suitable position; the acquisition assembly 4 rotating circumferentially around the placement plate 2 with the drive assembly 3 to a suitable horizontal position; adjusting the acquisition assembly 4 to move the acquisition unit 46 above the placement plate 2, rotating it in a circle with the center of the placement plate 2 as the center and the distance between the acquisition unit 46 and the center as the radius to a preset position, so that the acquisition unit 46 can capture images of the fish on the placement plate 2. By repeatedly adjusting the positions of the drive assembly 3 and the acquisition assembly 4, images of the fish can be captured from multiple angles.

[0070] As a specific example of the drive component 3, the first surface or sidewall of the mounting port is recessed inward to form a mounting groove;

[0071] like Figure 3 As shown, the driving component 3 includes:

[0072] The first motor 31 is located on one side of the acquisition station 1;

[0073] The drive gear 32 has its shaft connected to the output end of the first motor 31;

[0074] The driven gear ring 33 is rotatably sleeved on the outside of the bearing and meshes with the driving gear 32. The bearing and the driven gear ring 33 are disposed in the mounting groove.

[0075] The drive assembly 3 of this embodiment further includes a bearing, a driven gear ring 33 rotatably sleeved on the outside of the annular bearing, and the bearing fixed in the mounting groove; the driven gear ring 33 and the annular bearing may each be provided with a protective shell, the protective shell of the driven gear ring 33 being disposed on its first surface, and the protective shell of the annular bearing being disposed on its first surface and sidewall. The driven gear ring 33 being disposed in the mounting groove allows its first surface to be flush with the first surface of the collection platform 1, facilitating the pushing of the fish out of the placement plate 2.

[0076] Specifically, when the size of the acquisition platform 1 is large enough, a recess can be formed in one side of the mounting slot to form a groove. The groove is connected to the mounting slot and is used to install the drive gear 32. The first motor 31 is fixed to the second surface of the acquisition platform 1. The output end of the first motor 31 passes through the acquisition platform 1 and is fixedly connected to the rotating shaft of the drive gear 32. The drive gear 32 meshes with the driven gear ring 33. When the size of the acquisition platform 1 is relatively small, an opening can be provided in the mounting slot near the edge of the acquisition platform 1, and a mounting plate can be provided at the edge of the acquisition platform 1. The first motor 31 is located on the second surface of the mounting plate. The output end of the first motor 31 passes through the mounting plate and is fixedly connected to the rotating shaft of the drive gear 32. The drive gear 32 is located above the first surface of the mounting plate, and the drive gear 32 meshes with the driven gear ring 33 through the opening.

[0077] In this embodiment, the first motor 31 drives the driving gear 32 to rotate, thereby driving the driven gear ring 33 to rotate, and thus changing the acquisition position of the acquisition component 4.

[0078] like Figure 4 As shown, as an example of acquisition component 4, acquisition component 4 includes:

[0079] The annular slide rail 41 is semi-circular in shape, with its two ends spanning the radial sides of the driven toothed ring 33, and a hollowed-out slide rail in the middle.

[0080] Slider 44 is slidably disposed in the slide rail, and a mounting hole is provided in the central area thereof;

[0081] The insertion plate 45 is movably inserted through the mounting hole, and the acquisition unit 46 is provided at one end facing the placement plate 2.

[0082] A rotating assembly is disposed on the driven gear ring 33 and connected to one end of the insert plate 45 via a connector. When the rotating assembly rotates, it drives the acquisition unit 46 to rotate around the center of the placement plate 2 with the distance between the acquisition unit 46 and the center of the circle as the radius.

[0083] In this embodiment, the center of the annular slide rail 41 coincides with the center of the driven toothed ring 33. The annular slide rail 41 is hollow in the middle to form a slide channel, and both sides of the slide channel are provided with inwardly recessed grooves. The two ends of the slider 44 are respectively embedded in the grooves on both sides of the annular slide rail 41. The central area of ​​the slider 44 is provided with a mounting hole, which cooperates with the slide channel. The insert plate 45 can be movably inserted through the mounting hole to achieve radial movement along the annular slide rail 41. The annular slide rail 41 can also be hollow in the middle to form a slide channel, with both ends of the slider 44 respectively provided with inwardly recessed grooves, and the two sides of the slide channel respectively embedded in the corresponding grooves. Because the location of fish feature information is uncertain, and the acquisition unit 46 can move along the semi-circular track to achieve a certain tilt angle for shooting, it is more conducive to the acquisition of fish feature information.

[0084] like Figure 4 As shown, as an example of a rotating assembly, the rotating assembly includes:

[0085] The vertical plate 47 has one end set on the driven toothed ring 33 and forms a connection point. The line connecting the connection point and the center of the placement plate 2 is perpendicular to the line connecting the two ends of the annular slide rail 41.

[0086] Mounting bracket 49 is attached to the side of the vertical plate 47 facing the annular slide rail 41, with its two ends close to the two ends of the vertical plate 47 respectively;

[0087] The screw 410 is rotatably connected at both ends to the two ends of the mounting bracket 49;

[0088] The engagement block 412 is sleeved on the outer surface of the screw 410 and threadedly connected to the screw 410. One side of the engagement block 412 is in contact with the side of the vertical plate 47 facing the annular slide rail 41, and the other side is connected to one end of the insert plate 45 through a connector.

[0089] The second motor 48 is located on the side of the vertical plate 47 away from the annular slide rail 41 and close to the acquisition platform 1, and its output end passes through the vertical plate 47 and is connected to the end of the mounting frame 49 close to the acquisition platform 1.

[0090] The third motor 411 is located at the other end of the mounting frame 49 away from the acquisition platform 1, and its output shaft is connected to one end of the screw 410 through the other end of the mounting frame 49.

[0091] In this embodiment, one end of the vertical plate 47 is fixed to the driven toothed ring 33, forming a connection point. The line connecting the connection point and the center of the placement plate 2 is perpendicular to the line connecting the two ends of the annular slide rail 41, which is the diameter of its semicircle. The mounting frame 49 includes a frame body and connecting parts. The frame body of the mounting frame 49 is attached to the side of the vertical plate 47 facing the annular slide rail 41. Both ends of the mounting frame 49 are bent toward the annular slide rail 41 to form connecting parts. The screw 410 is rotatably connected to the two connecting parts. The mounting frame 49 may also include a connecting block, which is attached to the surface of the vertical plate 47. One end of the mounting frame 49 near the acquisition platform 1 or the connecting part is connected to the connecting block. The output end of the second motor 48 can be fixedly connected to the vertical plate 47 and the connecting block or connecting part of the mounting frame 49 near the acquisition platform 1.

[0092] The meshing block 412 of this embodiment can also mesh with the screw 410. The meshing block 412 has a block structure, and one side of the meshing block 412 is in contact with the vertical plate 47. The axis of the second motor 48 is on the same straight line as the center of the annular slide rail 41, and the line connecting the two is perpendicular to the line connecting the two ends of the annular slide rail 41. The third motor 411 drives the screw 410 to rotate. The block-shaped engagement block 412 is blocked by the vertical plate 47 and will not rotate on its own, so it can slide along the surface of the mounting frame 49, thereby driving the acquisition unit 46 under the insertion plate 45 to move vertically. After the acquisition unit 46 moves to the predetermined position, the second motor 48 drives one end of the mounting frame 49 to rotate. The frame of the mounting frame 49 separates from the vertical plate 47 and rotates around the intersection of the second motor 48 and the vertical plate 47. The engagement block 412 drives the insertion plate 45 to slide in the annular slide rail 41, thereby driving the acquisition unit 46 at the end of the insertion plate 45 to rotate around the center of the annular slide rail 41. The radius is the distance from the acquisition unit 46 to the center of the annular slide rail 41. Adjusting the distance between the engagement block 412 and the acquisition stage 1 is equivalent to adjusting the distance between the acquisition unit 46 and the center of the annular slide rail 41. When acquiring fish images, the position of the acquisition unit 46 relative to the fish and at various angles can be adjusted, and multi-angle image acquisition of fish samples can be performed, especially the acquisition of information on feature parts.

[0093] The third motor 411 in this embodiment drives the screw 410 to rotate. The meshing block 412 meshes with the screw 410, causing the insertion plate 45 to move in the vertical direction, changing the height of the acquisition unit 46. This can be applied to fish samples of different sizes, thereby making the image clarity and size similar.

[0094] like Figure 3 and Figure 5As shown, as a specific example of the device, the device also includes two ranging components, which are respectively disposed at the bottom of both ends of the annular slide rail 41. When the placement plate 2 rises to the point where its surface is flush with the surface of the collection platform 1, it is used to measure the size of the fish on the placement plate 2.

[0095] The ranging component includes:

[0096] The second electric push rod 42 is located at the bottom of one end of the annular slide rail 41, and its telescopic end passes through the bottom of the annular slide rail 41 to extend and retract toward the acquisition platform 1.

[0097] Clamping plate 43 is disposed at the telescopic end of the second electric push rod 42;

[0098] One of the ranging components has a rangefinder 413 at both ends of the clamping plate 43. The clamping plates 43 of the two ranging components are used to hold fish. The rangefinder 413 measures the size of the fish between the two clamping plates, thus realizing the measurement of the fish size.

[0099] In this embodiment of the present disclosure, the bottom ends of both ends of the annular slide rail 41 are fixedly mounted with second electric push rods 42, and the extended ends of the second electric push rods 42 are fixed with clamping plates 43, wherein, as Figure 5 As shown, a rangefinder 413 is fixed at both ends of the clamping plate 43 on one side of the collection platform 1. The second electric push rod 42 drives the clamping plate 43 to clamp and limit the fish sample. The rangefinder 413 measures the distance between the clamping plates 43 on both sides and uploads it to the database. The drive component 3 drives the rotation of the annular slide rail 41, which can measure the length of the fish sample at multiple positions. This can collect fish information and help identify fish species.

[0100] As a specific example of the device, the device further includes: a feeding platform 12 and a delivery platform 11, which are arranged opposite to each other on both sides of the collection platform 1. The feeding platform 12 is located at the bottom of one side of the collection platform 1; the delivery platform 11 is located at the edge of the other side of the collection platform 1. A conveying mechanism is provided between the feeding platform 12 and the placement plate 2. First, the placement plate 2 is lowered below the collection platform 1, and the obtained fish sample is placed into the feeding platform 12. The fish sample from the feeding platform 12 is then conveyed to the placement plate 2 through the conveying mechanism, which includes a conveyor belt and rollers. After the fish is fed onto the placement plate 2 through the feeding platform 12, the fish can be kept still by manually pressing it down, provided that it does not affect the collection of fish characteristic information. Alternatively, before conveying, the fish can be tied up to prevent its body from bending. The fish can also be put into a dormant state without being killed.

[0101] The bottom of the tail end of the collection platform 1 in this embodiment is fixed with a feeding platform 12, and the top of the front end of the collection platform 1 is fixed with a delivery platform 11. Both the feeding platform 12 and the delivery platform 11 are conveyor platforms composed of conveyor belts. When using the device, the feeding platform 12 feeds the fish sample onto the placement plate 2. Then, the first electric push rod 21 drives the placement plate 2 to rise to be level with the collection platform 1. The collection unit 46 performs length measurement, image acquisition, and feature acquisition on the fish sample, and cooperates with the drive component 3 and the collection component 4 to perform multi-directional acquisition. After the acquisition is completed, the sample is fed into the delivery platform 11 and sent out to collect the next set of samples.

[0102] like Figure 5 As shown, as a specific example of the device, the device further includes:

[0103] A pushing component is disposed on one side of the collection platform 1, used to push the fish on the placement plate 2 out of the collection platform 1.

[0104] In this embodiment, the pushing component can be positioned on one side of the acquisition platform 1 and maintained at a certain distance from the acquisition component 4 to avoid obstructing the rotation of the acquisition component 4 when acquiring images. When pushing the fish on the placement plate 2 out of the acquisition platform 1, the pushing component is on the same side as the vertical plate 47 of the acquisition component 4, and the vertical plate 47 is located in the central area of ​​the pushing component. This avoids collisions with other parts of the acquisition component 4 when pushing the fish sample out of the acquisition platform 1. When this device has a delivery platform 11, the pushing component pushes the fish out of the placement plate 2 and into the delivery platform 11.

[0105] like Figure 5 As shown, as a specific example of a pushing component, the pushing component includes:

[0106] The base is disposed on the acquisition station 1;

[0107] The third electric push rod 5 has one end fixed to the base and the other retractable end facing the acquisition platform 1;

[0108] Push plate 51 is disposed at the telescopic end of the third electric push rod 5.

[0109] In this embodiment of the device, the base can be disposed on the collection platform 1. If the collection platform 1 is relatively small, a connecting plate can be disposed on one side of the collection platform 1, and the base can be disposed on the connecting plate. Two third electric push rods 5 can be disposed at intervals on both sides of the base. The pushing area of ​​the push plates 51 on the two third electric push rods 5 covers the collection platform 1. The distance between the two push plates 51 can pass through the third motor 411 and the vertical plate 47, but should not be too large to avoid missing fish when pushing the fish on the placement plate 2.

[0110] Two third electric push rods 5 are fixed at the tail end of the collection platform 1, and a push plate 51 is fixed at the extended end of the third electric push rod 5. When the device includes the feeding platform 12 and the delivery platform 11, the pushing component and the feeding platform 12 are located on the same side of the collection platform 1. When the pushing component of this embodiment is not working, the push plate 51 is located at the rear end of the vertical plate 47 and will not interfere with the rotation of the vertical plate 47. After the fish sample collection is completed, the third electric push rod 5 drives the push plate 51 to push the sample onto the delivery platform 11 for delivery.

[0111] When using the device, the feeding platform 12 receives fish samples and, via a conveying mechanism, feeds them onto the placement plate 2. Then, the first electric push rod 21 raises the placement plate 2 to be level with the collection platform 1. The first motor 31 drives the drive gear 32 to rotate and mesh with the driven gear ring 33, thereby changing the collection position of the collection unit 46. The second motor 48 drives the mounting bracket 49 to rotate, causing the slider 44 to slide along the annular slide rail 41, adjusting the position of the collection unit 46. This allows for multi-angle image acquisition of the fish sample, especially for capturing information on characteristic areas. The third motor 411 drives the screw 410 to rotate, meshing... The engagement of the block 412 and the screw 410 drives the insert plate 45 to move vertically, changing the height of the acquisition unit 46. This allows it to be used for fish samples of different sizes, resulting in images with similar clarity and size. The second electric push rod 42 drives the clamping plate 43 to clamp and limit the fish sample, and the distance between the clamping sides is measured by the rangefinder 413 and uploaded to the database. The drive component 3 drives the rotation of the annular slide rail 41, which can measure the length of the fish sample at multiple positions. After the fish sample is collected, the third electric push rod 5 drives the push plate 51 to push the sample onto the delivery platform 11 for delivery.

[0112] The device further includes a control unit and a recognition unit. The control unit is connected to the recognition unit, the acquisition unit 46, the drive assembly 3, and each motor of the acquisition assembly 4. The recognition unit is used to identify the species of fish in the acquired images. The recognition unit stores a trained fish recognition model. The control unit is used to drive the start and stop of each motor and control the recognition unit to acquire and recognize the fish images acquired by the acquisition unit 46. When the device includes a first push rod, a second push rod, and a third push rod, the control unit is electrically connected to the first push rod, the second push rod, and the third push rod respectively to control the start and stop of each push rod.

[0113] By adopting the above-described technical solutions disclosed in the embodiments of this disclosure, the following beneficial effects are obtained:

[0114] The device in this embodiment delivers fish samples onto the placement plate 2. The placement plate 2 rises to meet the collection platform 1, and the collection unit 46 photographs the fish samples to obtain fish images. Alternatively, the fish samples can be delivered onto the placement plate 2 via the delivery platform 12. Then, the driving component 3 cooperates with the collection component 4 to take multi-angle photographs of the fish samples, collecting detailed features of various parts of the fish and measuring the sample length. After transmitting the collected data to the database, the sample is sent out via the delivery platform 11, improving the efficiency of fish sample collection. The collected fish images are input into the recognition unit to achieve fish identification. Compared to manually classifying samples one by one and collecting feature information item by item, this method is more convenient and faster, and facilitates database entry.

[0115] Another aspect of this disclosure provides a deep learning-based fish identification method, which includes the following steps:

[0116] Step S100: Obtain fish images using the above-described device.

[0117] Specifically, the fish sample is fed onto the placement plate 2 via the feeding platform 12 and the conveying mechanism, and the fish sample is photographed by the acquisition unit 46 to obtain fish images; the collected fish sample is then pushed out of the placement plate 2 and sent out by the delivery platform 11.

[0118] Step S200: Input the fish image into a pre-built fish recognition model to obtain the fish recognition result; wherein, the fish recognition model includes models trained based on Fast RCNN.

[0119] The recognition method of this disclosure performs convolution calculations on the fish image using the VGG network of Fast RCNN to extract the feature information of the fish image.

[0120] For example, the fish recognition model training process includes:

[0121] Step S210: Obtain fish images;

[0122] Step S220: Label the types of fish images and the various parts of the fish as a dataset. The various parts of the fish include one or more of the following: gill openings, gill covers, scales, dorsal fin, pectoral fin, and caudal fin.

[0123] Step S230: Divide the dataset into training set, validation set and test set to train Fast RCNN and build a fish recognition model.

[0124] To train the fish recognition model, a large number of image samples of different types of fish need to be acquired. The types and body parts of different fish need to be labeled. The location, shape, and size of body parts vary among different types of fish; some fish may lack scales, so manual labeling is necessary. Based on the fish images, the gill openings, gill covers, scales, dorsal fin, pectoral fin, and caudal fin areas of different fish species are identified. The fish images include various morphologies of the fish samples, such as front, back, and side views, and whether the fins are spread or not.

[0125] For example, dividing the dataset into training, validation, and test sets to train Fast R-CNN and construct a fish recognition model includes:

[0126] Step S231: The database is divided into a training set, a validation set, and a test set; a benchmark model is trained based on the training set, wherein the benchmark model is a FastRCNN model;

[0127] Step S232: Input the validation set into the benchmark model to generate a first prediction result, and adjust the parameters of the benchmark model according to the first prediction result to generate a fish recognition model;

[0128] Step S233: Input the test set into the fish recognition model to generate a second prediction result, and generate a generalization performance score of the fish recognition model based on the second prediction result;

[0129] Step S234: Determine the final fish recognition model based on the generalization performance score, and perform fish recognition based on the final fish recognition model.

[0130] Another aspect of this disclosure provides a deep learning-based fish identification device, the device comprising:

[0131] The acquisition module is used to acquire fish images using the aforementioned device;

[0132] The recognition module is used to input the fish image into a pre-built fish recognition model to obtain the fish recognition result; wherein, the fish recognition model includes one trained on Fast RCNN.

[0133] The fish identification module, also known as the identification unit, is an embodiment of this disclosure.

[0134] The training process for the fish recognition model includes:

[0135] The acquisition module is used to acquire images of fish.

[0136] The labeling module is used to label the types of fish images and the various parts of the fish as a dataset. The various parts of the fish include one or more of the following: gill openings, gill covers, scales, dorsal fin, pectoral fin, and caudal fin.

[0137] The calibration module disclosed herein calibrates the fish-containing areas based on the fish image, generates a description file corresponding to the fish-containing areas, and associates the description file with the fish image.

[0138] The building block is used to divide the dataset into training, validation, and test sets to train Fast R-CNN and build a fish recognition model.

[0139] The building module includes:

[0140] The data processing module is used to build a dataset based on each fish image and its corresponding description file, wherein the dataset is divided into a training set, a validation set, and a test set.

[0141] The model training module is used to train a baseline model based on the training set, wherein the baseline model is a FastRCNN model.

[0142] The model validation module is used to input the validation set into the benchmark model to generate a first prediction result, and adjust the parameters of the benchmark model according to the first prediction result to generate a fish recognition model.

[0143] The model testing module is used to input the test set into the fish recognition model to generate a second prediction result, and generate a generalization performance score of the fish recognition model based on the second prediction result.

[0144] The identification module is used to determine the final fish identification model based on the generalization performance score, and to identify fish based on the final fish identification model.

[0145] Another aspect of this disclosure provides a computer-readable storage medium for fish identification based on deep learning, wherein the storage medium stores a fish identification program based on deep learning, which, when executed by a processor, implements the fish identification method as described above.

[0146] Another aspect of this disclosure provides a terminal device for fish identification based on deep learning, including a memory, a processor, and a fish identification program based on deep learning stored in the memory and executable on the processor. When the processor executes the fish identification program based on deep learning as described above, it implements a fish identification method.

[0147] The foregoing has shown and described the basic principles and main features of the embodiments of this disclosure and the advantages of the invention. It will be apparent to those skilled in the art that the embodiments of this disclosure are not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the embodiments of this disclosure. Therefore, the embodiments should be considered exemplary and non-limiting in all respects. The scope of the embodiments of this disclosure is defined by the appended claims rather than the foregoing description, and therefore all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the embodiments of this disclosure. No reference numerals in the claims should be construed as limiting the scope of the claims. The above descriptions are merely preferred embodiments of the embodiments of this disclosure. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principles of the embodiments of this disclosure, and these improvements and modifications should also be considered within the protection scope of the embodiments of this disclosure.

Claims

1. A fish identification device based on deep learning, characterized in that, The device includes: The data acquisition station is equipped with an installation port; The placement plate can be raised and lowered through the mounting port. When the first surface of the placement plate is lowered to below the acquisition platform, it is used to acquire the fish image to be acquired. When the first surface of the placement plate is raised to be flush with the first surface of the acquisition platform, it is used to place the fish image to be acquired in the acquisition area of ​​the acquisition unit. A drive component is disposed at the mounting port of the acquisition station and can rotate circumferentially along the mounting port; A data acquisition component is disposed on the driving component. The data acquisition component can rotate circumferentially along the mounting opening with the driving component. The data acquisition unit of the data acquisition component is above the first surface of the placement plate. With the center of the placement plate as the center, the data acquisition unit rotates in a semi-circular manner with the distance between the data acquisition unit and the center as the radius, so that the data acquisition unit can acquire images of fish on the placement plate from multiple angles. The first surface or sidewall of the mounting port is recessed to form a mounting groove; The driving component includes: The first motor is located on one side of the data acquisition station; The drive gear has its shaft connected to the output end of the first motor; The driven gear ring is rotatably sleeved on the outside of the bearing and meshes with the driving gear. The bearing and the driven gear ring are disposed in the mounting groove. The acquisition component includes: The annular slide rail is semi-circular in shape, with its two ends spanning the radial sides of the driven toothed ring, and has a hollowed-out slide track in the middle. A slider is slidably disposed in the slide rail, and a mounting hole is provided in the central area thereof; An insert plate is movably inserted through the mounting hole, and the acquisition unit is provided at one end facing the placement plate. A rotating component is disposed on the driven gear ring and connected to one end of the insert plate via a connector. When the rotating component rotates, it drives the acquisition unit to rotate around the center of the placement plate as the center, with the distance between the acquisition unit and the center as the radius. The rotating assembly includes: A vertical plate, one end of which is disposed on the driven toothed ring and forms a connection point, wherein the line connecting the connection point and the center of the placement plate is perpendicular to the line connecting the two ends of the annular slide rail; The mounting bracket is attached to the side of the vertical plate facing the annular slide rail, with its two ends close to the two ends of the vertical plate respectively; The screw is rotatably connected at both ends to the two ends of the mounting bracket. A meshing block is fitted onto the outer surface of the screw and threadedly connected to the screw. One side of the meshing block is in contact with the side of the vertical plate facing the annular slide rail, and the other side is connected to one end of the insert plate through a connector. The second motor is located on the side of the vertical plate away from the annular slide rail, near the end of the acquisition platform, and its output end passes through the vertical plate and is connected to the end of the mounting frame near the acquisition platform. The third motor is located at the other end of the mounting frame away from the acquisition platform, and its output shaft is connected to one end of the screw through the other end of the mounting frame.

2. The apparatus according to claim 1, characterized in that, The device also includes two ranging components, which are respectively disposed at the bottom of both ends of the annular slide rail. When the placement plate rises to the point where its surface is flush with the surface of the collection platform, it is used to measure the size of the fish on the placement plate.

3. The apparatus according to claim 2, characterized in that, The ranging component includes: The second electric push rod is located at the bottom of one end of the annular slide rail, and its telescopic end passes through the bottom of the annular slide rail and moves telescopically toward the acquisition platform. A clamp is provided at the telescopic end of the second electric push rod; One of the ranging components has a rangefinder at both ends of its clamping plate. The clamping plates of the two ranging components are used to hold fish, and the rangefinder measures the size of the fish between the two clamping plates.

4. The apparatus according to any one of claims 1 to 3, characterized in that, The device further includes: A pushing component, located on one side of the collection platform, is used to push the fish on the placement plate out of the collection platform.

5. The apparatus according to claim 4, characterized in that, The actuating component includes: A base is provided on the acquisition station; The third electric push rod has one end fixed to the base and the other retractable end facing the acquisition platform; A push plate is located at the telescopic end of the third electric push rod.

6. The apparatus according to any one of claims 1 to 3, characterized in that, The device further includes a control unit and an identification unit, wherein the control unit is connected to the identification unit, the acquisition unit, the drive assembly, and each motor of the acquisition assembly. The recognition unit is used to identify the species of fish in the acquired image, wherein the recognition unit stores a trained fish recognition model; The control unit is used to drive the start and stop of each motor and control the recognition unit to acquire and recognize the fish images collected by the acquisition unit.

7. A fish identification method based on deep learning, characterized in that, The method includes: Fish images are acquired using any of the deep learning-based fish recognition devices according to claims 1 to 6; The fish image is input into a pre-built fish recognition model to obtain the fish recognition result; wherein, the fish recognition model includes models trained based on Fast RCNN; The training process for the fish recognition model includes: Acquire fish images; The data set includes the types of fish images and the various parts of the fish. The various parts of the fish include one or more of the following: gill openings, gill covers, scales, dorsal fin, pectoral fin, and caudal fin. The dataset was divided into training, validation, and test sets to train Fast R-CNN and build a fish recognition model.

Images