Cultured fish classification device and classification method

Through the improved YOLOv8 algorithm and visual processing unit combined with a rotary partition design fish classification device, the problem of difficult to identify and classify sick fish in the prior art is solved, the accurate identification and automatic classification of farmed fish is realized, and the improvement of the improvement of the Classification counting efficiency and device reliability.

CN118968161APending Publication Date: 2024-11-15JIANGSU UNIV OF SCI & TECH
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411026161.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing breeding fish classification technology is difficult to accurately identify and classify different types of diseased fish, resulting in low disease detection and counting efficiency and affecting the production efficiency of the farm.

Method used

The improved YOLOv8 algorithm is used to combine visual processing units and rotary partition design fish classification device to identify fish types through the camera and automatically sort healthy fish and different types of sick fish. The pneumatic acceleration components are used to prevent fish from being trapped, realizing the Fine classification and counting of fish.

Benefits of technology

Accurate identification and automatic classification of different types of diseased fish are achieved, classification and counting efficiency is improved, mechanical structure complexity is reduced, the reliability of the device is ensured, and fish crowds are congested and water retention is prevented.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118968161A_ABST
    Figure CN118968161A_ABST
Patent Text Reader

Abstract

The invention discloses a cultured fish classification device and a classification method. The classification device comprises a to-be-classified fish pond, a healthy fish pond, a plurality of sick fish ponds, a fish school channel assembly and a camera, wherein the inlet end of the fish school channel assembly communicates with the to-be-classified fish pond, and the outlet ends of the fish school channel assembly communicate with the healthy fish pond and the sick fish ponds correspondingly; the camera is used for shooting to-be-classified fishes in the fish school channel assembly; the visual processing unit is used for receiving images shot by the camera and recognizing the types of healthy fishes and diseased fishes according to the images, and the fish sorting assembly is located in the fish school channel assembly and used for sorting the healthy fishes and the diseased fishes to different channels. According to the invention, the improved YOLOv8 algorithm is adopted, accurate identification of different types of sick fishes can be realized, automatic classification of healthy fishes and sick fishes can be realized by the fish sorting assembly, automatic fine classification of various sick fishes can be realized by the sick fish classification assembly, automatic counting of healthy fishes and various sick fishes can be realized, and the classification and counting efficiency is high.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a classification device and a classification method, and in particular to a farmed fish classification device and a classification method. Background Art

[0002] Farmed fish is one of the most important food sources in the world. To meet the growing market demand, farmers have gradually increased the density of farmed fish. However, the high-density farm environment provides favorable conditions for the spread of pathogens. Once a disease breaks out, it will spread rapidly and cause large-scale fish deaths. This will not only cause economic losses to farmers, but also worsen the ecological environment. In addition, the presence of sick fish will also affect the growth environment and quality of other healthy fish, thereby affecting the production efficiency of the entire farm.

[0003] In traditional aquaculture management, farmers mainly rely on manual observation and empirical judgment to detect and identify sick fish. This method is not only time-consuming and laborious, but also often difficult to detect sick fish in a timely and accurate manner. However, the existing farmed fish classification technology is not mature. There are many types of fish diseases, and it is difficult to classify and count sick fish for each disease. The existing classification device only picks out all sick fish in a unified manner, and does not form a refined classification for different types of sick fish. Summary of the invention

[0004] Purpose of the invention: The first purpose of the present invention is to provide a farmed fish classification device that can realize the classification of healthy fish and diseased fish, and the fine classification of multiple diseased fish.

[0005] The second object of the present invention is to provide a method for classifying farmed fish.

[0006] Technical solution: The present invention discloses a farmed fish classification device, comprising a fish pond to be classified, a healthy fish pond, a plurality of sick fish ponds, a fish school channel component whose inlet end is connected to the fish pond to be classified and whose multiple outlet ends are respectively connected to the healthy fish pond and the plurality of sick fish ponds, a camera for photographing the fish to be classified in the fish school channel component, a visual processing unit for receiving images photographed by the camera and identifying the types of healthy fish and sick fish according to the images, a fish sorting component located in the fish school channel component and used for sorting healthy fish and sick fish into different channels, a sick fish classification component located in the fish school channel component and used for sorting different types of sick fish into different channels, and a controller for controlling the fish sorting component and the sick fish classification component according to the recognition result of the visual processing unit.

[0007] Furthermore, the fish channel assembly includes a fish inlet channel, a fish identification channel, and a fish sorting channel with two outlets, which are sequentially connected to the fish ponds to be classified, wherein one outlet of the fish sorting channel is connected to the healthy fish pond through the healthy fish channel, and the other outlet of the fish sorting channel is connected to multiple sick fish ponds through a sick fish channel with multiple outlets; the outlet end of the healthy fish channel and the multiple outlet ends of the sick fish channel are both equipped with fish outlet channels.

[0008] Furthermore, a pneumatic acceleration component is installed on the fish outlet channel to prevent fish from being trapped and blocked at the end of the fish outlet channel. The pneumatic acceleration component includes an air pump and a delivery hose connected between the air pump and the fish outlet channel. A counter connected to a controller is installed at the end of the fish outlet channel, and the controller controls the action of the air pump according to the count of the counter.

[0009] Furthermore, the diameter of the fish inlet channel gradually decreases along the direction of the water flow, and the first fish limiting component and the second fish limiting component for limiting the number of fish entering are arranged inside the fish inlet channel at the front and back, the first fish limiting component includes a first upper fish baffle and a first lower fish baffle symmetrically rotatably installed on the inner wall of the fish inlet channel, a first upper limit spring is installed between the first upper fish baffle and the inner wall of the fish inlet channel, a first lower limit spring is installed between the first lower fish baffle and the inner wall of the fish inlet channel, and two fish can be allowed to pass when the first upper fish baffle and the first lower fish baffle are opened to the limit position; the second fish limiting component includes a second upper fish baffle and a second lower fish baffle symmetrically rotatably installed on the inner wall of the fish inlet channel, a second upper limit spring is installed between the second upper fish baffle and the inner wall of the fish inlet channel, a second lower limit spring is installed between the second lower fish baffle and the inner wall of the fish inlet channel, and one fish can be allowed to pass when the second upper fish baffle and the second lower fish baffle are opened to the limit position.

[0010] Furthermore, the fish sorting component includes a fish sorting sealing box arranged in sequence outside the fish sorting channel, a first motor installed in the fish sorting sealing box and with its output shaft extending into the fish sorting channel, a first sealing ring used to seal the connection between the output shaft of the first motor and the fish sorting channel, a fish sorting partition connected to the output shaft of the first motor and located in the fish sorting channel, and a first magnet fixedly installed on the outside of the fish sorting partition and used to fix the fish sorting partition after the first motor is powered off. The fish sorting partition is located at the three-port intersection node of the fish sorting pipeline, and the first motor drives the fish sorting partition to rotate to realize the opening and closing of the two outlets of the fish sorting channel.

[0011] The sick fish classification component includes a sick fish sealing box arranged outside the sick fish channel, a second motor installed in the sick fish sealing box and with its output shaft extending into the sick fish channel, a second sealing ring used to seal the connection between the output shaft of the second motor and the sick fish channel, a sick fish classification partition connected to the output shaft of the second motor and located in the sick fish channel, and a second magnet fixedly installed on the outside of the sick fish classification partition and used to fix the sick fish classification partition after the second motor is powered off. The sick fish classification partition is located at a multi-port intersection node of the sick fish channel, and the second motor drives the sick fish classification partition to rotate to open and close multiple outlets of the sick fish channel.

[0012] Furthermore, auxiliary water inlet pipes are installed at the ends of the two outlets of the fish sorting channel, and the other ends of the auxiliary water inlet pipes extend into the fish pond to be sorted. Filters are installed at both ends of the auxiliary water inlet pipes.

[0013] Furthermore, the water levels of the healthy fish pond and the multiple diseased fish ponds are lower than the water level of the fish pond to be classified, the inner wall of the fish pond to be classified is installed with a fish repellent, and the inner walls of the healthy fish pond and the multiple diseased fish ponds are installed with fish attracting lights.

[0014] Based on the same inventive concept, the present invention also discloses a method for classifying farmed fish, comprising the following steps:

[0015] The fish enter the fish entry channel, and the first fish limiting component and the second fish limiting component allow the fish to enter the fish identification channel one by one;

[0016] The camera acquires images of the fish passing through the fish identification channel one by one, and transmits the images to the visual processing unit of the controller;

[0017] The detection model carried by the visual control unit recognizes and judges the input fish image, and transmits the recognition and judgment result to the central control unit of the controller, and the central control unit controls the first motor and the second motor to rotate to realize automatic classification;

[0018] The counter of each fish outlet channel records the number of the fish outlet channel and transmits the counting information to the counting unit of the controller. The counting unit classifies and summarizes the information, and the counting unit transmits the counting data to the central control unit, which controls the action of the pneumatic acceleration component (10).

[0019] Furthermore, the training steps of the detection model carried by the visual control unit are as follows:

[0020] Collect image data related to various types of diseased fish in farms, use Labelimg annotation tools to annotate them, perform data expansion, establish a diseased fish dataset, and divide the dataset into a training set and a validation set;

[0021] Improve the YOLOv8 algorithm model, build a CCE module, and replace all bottlenecks in C2f with the CCE module to form a C2f-CCE module, and replace all C2f in Backbone and Neck with C2f-CCE modules; the CCE module consists of a Conv convolution module and an ELA efficient local attention module; the Conv convolution module includes two, the first one is used to reduce the dimension of the input feature map and extract preliminary features, and the second one is used to perform group convolution operations to achieve dimensionality increase and deep-level feature extraction; the ELA efficient local attention module performs weighted processing on the output feature map of the second convolution to enhance important features and suppress irrelevant or redundant features;

[0022] The training set is imported into the improved YOLOv8 algorithm model. First, the features of the image are extracted using the Backbone feature extraction part. Then, in the Neck part, the extracted image features are upsampled, downsampled, and feature concatenated to obtain a fused feature map. Finally, a preliminary model is obtained in the Head part.

[0023] Import the validation set, use the preliminary model, calculate the loss value, and then perform backpropagation and parameter update until the loss function value is minimized and tends to be stable. The iteration stops and the final detection model is obtained.

[0024] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention adopts an improved YOLOv8 algorithm, which can realize accurate identification of different types of sick fish, the fish sorting component can realize automatic classification of healthy fish and sick fish, the sick fish classification component can realize automatic and fine classification of multiple sick fish, and can realize automatic counting of healthy fish and multiple sick fish, and the classification and counting efficiency is high; the fish sorting component and the sick fish classification component adopt a rotating baffle, which blocks other classification ports while opening the corresponding classification port, reduces the complexity of the mechanical structure, and ensures the reliability of the entire device; the setting of the first fish limiting component and the second fish limiting component can prevent fish from being crowded in the fish inlet channel and ensure that the fish enter the fish identification channel one by one; the setting of the pneumatic acceleration component can not only oxygenate the water body, but also effectively prevent fish from being trapped and blocked in the fish outlet channel. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a structural schematic diagram of the present invention;

[0026] Figure 2 It is a structural schematic diagram of the fish passage assembly and the frame of the present invention;

[0027] Figure 3 It is a structural schematic diagram of the fish passage assembly of the present invention;

[0028] Figure 4 is a cross-sectional view of a first fish limiting component and a second fish limiting component of the present invention;

[0029] Figure 5 For the present invention Figure 4 A schematic diagram of the first fish-limiting component and the second fish-limiting component being opened by the squeeze of the fish;

[0030] Figure 6 is a cross-sectional view of a fish separation assembly of the present invention;

[0031] Figure 7 It is a structural schematic diagram of the fish separation partition of the present invention;

[0032] Figure 8A schematic diagram of opening different classification ports of the fish sorting component of the present invention;

[0033] Fig. 9 It is a structural schematic diagram of the diseased fish classification component of the present invention;

[0034] Fig.10 It is a structural schematic diagram of a diseased fish classification partition of the present invention;

[0035] Fig.11 A schematic diagram of the structure of the sick fish classification component of the present invention with different classification ports opened;

[0036] Fig.12 It is a cross-sectional view of the fish outlet channel of the present invention;

[0037] Fig.13 It is a schematic diagram of each module of the controller of the present invention;

[0038] Fig.14 It is a classification counting flow chart of the present invention;

[0039] Fig.15 This is a flow chart of the detection model training of the present invention;

[0040] Fig.16 It is a schematic diagram of the improved YOLOv8 model structure of the present invention;

[0041] Fig.17 It is a C2f-CCE structure diagram of the present invention;

[0042] Fig.18 It is a schematic diagram of the CCE module and ELA structure of the present invention. DETAILED DESCRIPTION

[0043] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0044] Example 1

[0045] like Figure 1 and Figure 2As shown, a farmed fish classification device of the present invention comprises a fish pond to be classified 1, a healthy fish pond 2, a plurality of sick fish ponds, a fish drive 6, a fish attracting lamp 7, a frame 8, an auxiliary water inlet pipe 9, a fish channel component, a camera 11, a visual processing unit, a controller 12, a counter 23 and a filter 24, the plurality of sick fish ponds comprise a Class I sick fish pond 3, a Class II sick fish pond 4 and a Class III sick fish pond 5, the water levels of the healthy fish pond 2 and the plurality of sick fish ponds are lower than the water level of the fish pond to be classified 1, the height difference between the fish pond to be classified 1 and the healthy fish pond 2, the Class I sick fish pond 3, the Class II sick fish pond 4 and the Class III sick fish pond 5 is conducive to the fish entering the healthy fish pond 2, the Class I sick fish pond 3, the Class II sick fish pond 4 and the Class III sick fish pond 5 from the fish pond to be classified 1, A fish drive 6 is installed on the inner wall of the fish pond 1 to be classified, and the fish drive 6 is preferably an ultrasonic fish drive. The inner walls of the healthy fish pond 2 and multiple diseased fish ponds are installed with fish attracting lights 7, and the healthy fish pond 2 and multiple diseased fish ponds are provided with water outlets on the side walls above the fish attracting lights 7. The arrangement of the fish drive 6 and the fish attracting lights 7 allows fish to enter the healthy fish pond 2, the Class I diseased fish pond 3, the Class II diseased fish pond 4 and the Class III diseased fish pond 5 from the fish pond 1 to be classified; the camera 11, the counter 23 and the electrical components in the fish channel assembly are electrically connected to the controller 12 respectively; the frame 8 is fixedly installed inside the healthy fish pond 2, the Class I diseased fish pond 3, the Class II diseased fish pond 4 and the Class III diseased fish pond 5, and the fish channel assembly and the controller 12 are both installed on the frame 8.

[0046] like Figures 1 to 3As shown, the fish channel component includes a fish entry channel 13, a fish identification channel 14 and a fish sorting channel 16 which are connected in sequence. The camera 11 is installed in the fish identification channel 14. The camera 11 takes images of fish passing through the fish identification channel 14 and transmits the images to the visual processing unit. The visual processing unit identifies the types of healthy fish and sick fish according to the images, and transmits the identification results to the controller 12. Preferably, the visual processing unit includes a trained improved YOLOv8 algorithm model, and the improved YOLOv8 algorithm model includes a Backbone layer, a Neck layer and a Head layer. The backbone feature extraction part of the Backbone layer is used to extract image features, and the Neck layer is used to perform up- Sampling, downsampling and feature connection are performed to obtain a fused feature map; the Head layer is used to obtain a preliminary model; the Backbone layer and the Neck layer both include a CCE module, and the CCE module is composed of a Conv convolution module and an ELA efficient local attention module; the fish sorting channel 16 is set as a three-way pipeline, one of the outlets of the fish sorting channel 16 is connected to the healthy fish channel 15, and the other outlet of the fish sorting channel 16 is connected to the sick fish channel 17, and the sick fish channel 17 is set as a multi-connected pipeline, the inlet end of the fish inlet channel 13 is connected to the fish pond 1 to be classified, the outlet of the healthy fish channel 15 is connected to the healthy fish pond 2, and the multiple outlets of the sick fish channel 17 are respectively connected to the Class I sick fish pond 3, the Class II sick fish pond 4 and the Class III sick fish pond The fish pond 5 is connected, the outlet of the healthy fish channel 15 and the multiple outlets of the sick fish channel 17 are connected and installed with a fish outlet channel 18, the fish enters from the fish inlet channel 13 and swims out from the fish outlet channel 18, the diameter of the fish inlet channel 13 gradually decreases along the direction of the water flow, which is conducive to the fish slowly passing through the fish inlet channel 13 and preventing the fish from being congested in the fish inlet channel 13, the diameter of the fish outlet channel 18 gradually increases along the direction of the water flow, and it is convenient for the fish to swim out, the fish inlet channel 13 and the fish outlet channel 18 are preferably Venturi channels, which are conducive to achieving fish inlet deceleration and outlet acceleration, the healthy fish channel 15, the fish separation channel 16 and the common disease fish channel 17 are also preferably Venturi channels; a counter 23 is installed at the outlet of the fish outlet channel 18, and the counter 2 3 calculates the number of fishes passing through the fish outlet channel 18; the outlet of the healthy fish channel 15 is connected to the first fish outlet channel 181, which is located in the healthy fish pond 2; the multiple outlets of the sick fish channel 17 are respectively connected to the second fish outlet channel 182, the third fish outlet channel 183, and the fourth fish outlet channel 184, the second fish outlet channel 182 is located in the Class I sick fish pond 3, the third fish outlet channel 183 is located in the Class II sick fish pond 4, and the fourth fish outlet channel 184 is located in the Class III sick fish pond 5; the first fish outlet channel 181 is equipped with a counter 23-1, the second fish outlet channel 182 is equipped with a counter 23-2, the third fish outlet channel 183 is equipped with a counter 23-3, and the fourth fish outlet channel 184 is equipped with a counter 23-4.

[0047] like Figures 3 to 5As shown, the first fish limiting assembly 19 and the second fish limiting assembly 20 for limiting the entry of fish are arranged inside the fish inlet passage 13. The first fish limiting assembly 19 includes a first upper limit fish partition 191 and a first lower limit fish partition 192 rotatably mounted on the inner wall of the fish inlet passage 13. A first upper limit spring 193 is installed between the first upper limit fish partition 191 and the inner wall of the fish inlet passage 13. A first lower limit spring 194 is installed between the first lower limit fish partition 192 and the inner wall of the fish inlet passage 13. When the first upper limit fish partition 191 and the first lower limit fish partition 192 are opened to the limit position, two fish can be allowed to pass through. The second fish limiting component 20 includes a second upper fish partition 201 and a second lower fish partition 202 rotatably mounted on the inner wall of the fish inlet channel 13, a second upper limit spring 203 is installed between the second upper fish partition 201 and the inner wall of the fish inlet channel 13, and a second lower limit spring 204 is installed between the second lower fish partition 202 and the inner wall of the fish inlet channel 13. When the second upper fish partition 201 and the second lower fish partition 202 are opened to the limit position, one fish can be allowed to pass through; the setting of the first fish limiting component 19 and the second fish limiting component 20 can further prevent fish from being blocked in the fish inlet channel 13.

[0048] like Figure 2 and Figure 6 As shown, a fish sorting assembly 21 for screening healthy fish and sick fish is installed inside the fish sorting channel 16. The fish sorting assembly 21 includes a fish sorting sealing box 215 arranged below the fish sorting channel 16, a first motor 214 installed in the fish sorting sealing box 215, a fish sorting partition 211 connected to the output shaft of the first motor 214, and a first magnet 213 fixedly installed on the outside of the fish sorting partition 211 for fixing the fish sorting partition 211 after the first motor 214 is powered off. The inner wall of the fish sorting sealing box 215 is provided with an iron block adapted to the first magnet 213. The fish sorting partition 211 is provided with a plurality of small holes to maintain a certain amount of water inlet. The fish sorting partition 211 is located at the T-shaped node of the fish sorting pipeline. The first motor 214 drives the fish sorting partition 211 to rotate to realize the opening and closing of the two classification ports of the fish sorting channel 16. The fish sorting channel 16 is provided with a first through hole for the output shaft of the first motor 214 to pass through, and the inner wall of the first through hole is fixedly connected with a first sealing ring 212. Figure 7 As shown, the fish separation partition 211 is configured as a semi-cup-shaped structure, and the fish separation partition 211 is fitted with the inner wall of the fish separation pipe 16, as shown in FIG. Figure 8As shown, when the controller 12 determines that the fish is a healthy fish, the first motor 213 of the controller 12 drives the fish separating partition 211 to rotate, and the fish separating partition 211 blocks the classification port leading to the sick fish channel 17, and opens the classification port leading to the healthy fish channel 15, that is, the opening of the fish separating partition 211 faces the classification port leading to the healthy fish channel 15; when the controller 12 determines that the fish is a sick fish, the controller 12 controls the first motor 213 to drive the fish separating partition 211 to rotate, and the fish separating partition 211 opens the classification port leading to the sick fish channel 17, and blocks the classification port leading to the healthy fish channel 15, that is, the opening of the fish separating partition 211 faces the classification port leading to the sick fish channel 17.

[0049] like Figure 2 and Fig. 9 As shown, a sick fish classification component 22 for screening different types of sick fish is installed inside the sick fish channel 17. The sick fish classification component 22 includes a sick fish sealing box 225 arranged below the sick fish channel 17, a second motor 224 installed in the sick fish sealing box 225, a sick fish classification partition 221 connected to the output shaft of the second motor 224, and a second magnet 223 fixedly installed on the outside of the sick fish classification partition 221 for fixing the sick fish classification partition 221 after the second motor 224 is powered off. The inner wall of the sick fish sealing box 225 is provided with an iron block adapted to the second magnet 223. The sick fish classification partition 221 is provided with a plurality of small holes to maintain a certain amount of water inlet. The sick fish classification partition 221 is located at the node of the sick fish channel 17. The second motor 224 drives the sick fish classification partition 221 to rotate to realize the opening and closing of the multiple classification ports of the sick fish channel 17. The sick fish channel 17 is provided with a second through hole for the output shaft of the second motor 224 to pass through, and a second sealing ring 222 is fixedly connected to the inner wall of the second through hole. like Fig.10 As shown, the sick fish classification partition 221 is configured as a cup-shaped structure with a 120° notch, and the sick fish classification partition 221 is fitted with the inner wall of the sick fish passage 17, as shown in FIG. Fig.11 As shown, when the controller 12 determines that the fish is a sick fish with Class I disease, the controller 12 controls the second motor 224 to drive the sick fish classification baffle 221 to rotate, and the gap of the sick fish classification baffle 221 is facing the Class I classification port of the sick fish channel 17, and the Class II and Class III classification ports are blocked; when the controller 12 determines that the fish is a sick fish with Class II disease, the controller 12 controls the second motor 224 to drive the sick fish classification baffle 221 to rotate, and the gap of the sick fish classification baffle 221 is opened toward the Class II classification port of the sick fish channel 17, and the Class I classification port and the Class III classification port are blocked; when the controller 12 determines that the fish is a sick fish with Class III disease, the controller 12 controls the second motor 224 to drive the sick fish classification baffle 221 to rotate, and the gap of the sick fish classification baffle 221 is opened toward the Class III classification port of the sick fish channel 17, and the Class I classification port and the Class II classification port are blocked.

[0050] like Figure 2 and Fig.12 As shown, a pneumatic acceleration component 10 is installed on the fish outlet channel 18, and the pneumatic acceleration component 10 includes an air pump 101 and a delivery hose 102 connected between the air pump 101 and the fish outlet channel 18, the air pump 101 is installed on the frame 8, the air pump 101 includes a first air pump 1011, a second air pump 1012, a third air pump 1013 and a fourth air pump 1014, the delivery hose 102 includes a first delivery hose 1021, a second delivery hose 1022, a third delivery hose 1023 and a fourth delivery hose 1024, the first delivery hose 1021 is connected between the first air pump 1011 and the first fish outlet channel 181, and the second delivery hose 1023 is connected between the first air pump 1011 and the first fish outlet channel 181. 1022 is connected between the second air pump 1012 and the second fish outlet channel 182, the third delivery hose 1023 is connected between the third air pump 1013 and the third fish outlet channel 183, and the fourth delivery hose 1024 is connected between the fourth air pump 1014 and the fourth fish outlet channel 184. When the controller 12 receives the information sent by the counter 23 being blocked, the controller 12 controls the air pump 101 to work, generate airflow, and deliver the gas to the corresponding fish outlet channel 18 through the delivery hose 102. A large number of bubbles are generated by air injection, generating an upflow, which pushes the fish to swim faster, thereby avoiding the fish being stuck at the end of the fish outlet channel 18.

[0051] like Figure 1 and Figure 2 There are two auxiliary water inlet pipes 9, and the two are connected to the ends of the two outlets of the fish sorting channel 16 respectively. The other end of the auxiliary water inlet pipe 9 extends into the fish pond 1 to be classified. Filter nets 24 are installed at both ends of the auxiliary water inlet pipe 9. The auxiliary water inlet pipe 9 is located above the fish sorting channel 16. The setting of the filter net 24 can prevent fish from entering the auxiliary water inlet pipe 9. The auxiliary water inlet pipe 9 can maintain the water inlet volume of each classification port of the fish sorting channel 16 after the fish sorting component 21 is switched.

[0052] Example 2

[0053] A method for classifying cultured fish of the present invention comprises the following steps:

[0054] (1) The fish enters the fish entry channel 13, and the first fish limiting component 19 and the second fish limiting component 20 allow the fish to enter the fish identification channel 14 one by one; the fish enters the fish entry channel 13 under the combined effect of the water level difference, the fish driving device 6 and the fish attracting light 7, and the fish moves forward to push the first upper limit fish partition 191 and the first lower limit fish partition 192. When the first upper limit fish partition 191 and the first lower limit fish partition 192 are opened to the limit position, two fish can be allowed to pass through at the same time. After the fish passes through the first fish limiting component 19, the fish continues to move forward to push the second upper limit fish partition 201 and the second lower limit fish partition 202. When the second upper limit fish partition 201 and the second lower limit fish partition 202 are opened to the limit position, only one fish can be allowed to pass through. The first fish limiting component 19 and the second fish limiting component 20 allow the fish to enter the fish identification channel 14 one by one, thereby avoiding the situation of fish school congestion.

[0055] (2) The camera 11 acquires images of the fish passing through the fish identification channel 14 one by one, and transmits the images to the controller 12 (the schematic diagram of each module of the controller 12 is shown in FIG. Fig.13 A visual processing unit (shown).

[0056] (3) The detection model carried by the visual control unit recognizes and judges the input fish image, and transmits the recognition and judgment result to the central control unit of the controller 12, and the central control unit controls the first motor 214 and the second motor 224 to rotate to realize automatic classification; when the detection model recognizes that the fish is a healthy fish, it is judged as a healthy fish, and the visual processing unit sends the recognition result to the central control unit, and the central control unit controls the first motor 214 to rotate clockwise, and the first motor 214 drives the fish separation partition 211 to rotate. The fish separation partition 211 rotates to open the classification port leading to the healthy fish channel 15, and at the same time blocks the classification port leading to the sick fish channel 17, and the first motor 214 is powered off, and the fish separation partition 211 is turned off. The first magnet 213 on the outside of 211 is adsorbed onto the iron block on the inner wall of the fish separation sealing box 215, fixing the fish separation partition 211 at the current position, and the healthy fish passes through the healthy fish channel 15. When the healthy fish passes through the fish outlet channel 181, the counter 23 is triggered, and the counter 23 sends the counting information to the counting unit of the controller 12. The calculation unit sends a signal to the central control unit, and the central control unit controls the pneumatic acceleration device 10 to start, and the first air pump 1011 works. The gas enters the first fish outlet channel 181 through the first delivery hose 1021 to prevent the fish from stagnating and realize the smooth discharge of the fish; when the detection model recognizes that the fish is a Class I lesion site, it is judged as a Class I diseased fish, and the visual processing unit will recognize the result. The result is sent to the central control unit, which controls the first motor 214 to rotate counterclockwise, and the first motor 214 drives the fish sorting partition 211 to rotate. The fish sorting partition 211 rotates to block the classification port leading to the healthy fish channel 15, and at the same time opens the classification port leading to the sick fish channel 17. The first motor 214 is powered off, and the first magnet 213 on the outside of the fish sorting partition 211 is adsorbed onto the iron blocks distributed on the inner wall of the fish sorting sealing box 215, so that the fish sorting partition 211 is fixed at the current position, and the sick fish enters the sick fish channel 17. The central control unit controls the second motor 224 to rotate clockwise, and the second motor 224 drives the sick fish sorting partition 221 to rotate, and the sick fish sorting partition 221 rotates to the classification port I. At the same time, the classification ports II and III are blocked, the second motor 224 is powered off, the second magnet 223 installed on the outer side of the diseased fish classification partition 221 is adsorbed onto the iron block on the inner wall of the diseased fish sealing box 225, the diseased fish classification partition 221 is fixed at the current position, and the Class I diseased fish enter the Class I classification port. When the Class I diseased fish passes through the fish outlet channel 182, the counter 23 is triggered, the counter 23 sends the counting information to the counting unit of the controller 12, the calculation unit sends a signal to the central control unit, the central control unit controls the pneumatic acceleration device 10 to start, the second air pump 1012 works, and the gas enters the second fish outlet channel 182 through the second delivery hose 1022, so as to prevent the fish from stagnating and realize the smooth discharge of the fish;When the detection model identifies that the fish has a Class II diseased part, it is judged as a Class II diseased fish, and the aforementioned steps of controlling the operation of the first motor 214 for Class I diseased fish are repeated, and then the second motor 224 is controlled to drive the diseased fish classification partition 221 to rotate to the Class II classification port, and the Class I classification port and the Class III classification port are blocked at the same time. When the Class II diseased fish passes through the fish outlet channel 183, the counter 23 is triggered, and the counter 23 sends the counting information to the counting unit of the controller 12, and the calculation unit sends a signal to the central control unit, and the central control unit controls the pneumatic acceleration device 10 to start, and the third air pump 1013 works, and the gas enters the third fish outlet channel 183 through the third delivery hose 1023; When the detection model identifies that the fish has a Class III diseased part, it is judged as a Class III diseased fish, and the aforementioned steps of controlling the operation of the first motor 214 for Class I diseased fish are repeated, and then the second motor 224 is controlled to drive the diseased fish classification partition 221 to rotate to the Class III classification port, and at the same time block the Class I and Class II classification ports. When the Class III diseased fish passes through the fish outlet channel 184, the counter 23 is triggered, and the counter 23 sends the counting information to the counting unit of the controller 12, and the calculation unit sends a signal to the central control unit. The central control unit controls the pneumatic acceleration device 10 to start, the fourth air pump 1014 works, and the gas enters the fourth fish outlet channel 184 through the fourth delivery hose 1024. For example; Fig.15 As shown, the training steps of the detection model carried by the visual control unit are as follows:

[0057] (3-a) Collect image data related to various types of diseased fish in the farm, use Labelimg annotation tools to annotate, perform data expansion, establish a diseased fish dataset, and divide the dataset into a training set and a validation set. Preferably, the dataset is divided into a training set and a validation set according to a ratio of 8:2;

[0058] (3-b) Improve the YOLOv8 algorithm model and import the training set into the improved YOLOv8 algorithm model. First, use the Backbone feature extraction part to extract the image features; then in the Neck part, upsample, downsample and connect the extracted image features to obtain the fused feature map; finally, obtain the preliminary model in the Head part; Figures 16 to 18 As shown, the improvement steps of YOLOv8 are as follows:

[0059] Build the CCE module and replace all bottlenecks in C2f with the CCE module to form the C2f-CCE module. Replace all C2f in Backbone and Neck with the C2f-CCE module, such as Fig.17As shown in the figure, it is the overall structure and connection order of C2f-CCE. It is the structure after the bottleneck module in the original c2f structure is replaced with the CCE module, which is used to show the specific details of the replacement position; a super-fusion attention module (CCE module) combining double convolution and enhanced local attention is designed. With the help of double convolution operation, the input features are extracted at multiple levels to reduce the number of parameters. Then, the ELA module is used to dynamically adjust the weight distribution of each position in the feature map to reduce the interference of complex background in the image and improve the recognition ability of the model. Replace all the bottlenecks of C2f with CCE modules to form C2f-CCE modules, and then replace all C2fs of Backbone and Neck in the YOLOv8 structure with C2f-CCE;

[0060] The CCE module structure is as follows Fig.18 As shown in the left part, CCE consists of three sub-modules, two Conv convolution modules and an Efficient Local Attention module (ELA);

[0061] The CCE module consists of a Conv convolution module and an ELA efficient local attention module;

[0062] Conv convolution module consists of two, the first one is used to reduce the dimension of the input feature map and extract preliminary features, the second one is used to perform group convolution operation to achieve dimensionality increase and deep feature extraction, to the output feature map with global feature information, and this feature map is used as the input feature map of the next step;

[0063] The ELA efficient local attention module performs weighted processing on the output feature map of the second convolution, enhances important features, and suppresses irrelevant or redundant features. The processing process of the input feature map by the efficient local attention module is shown in the right part of the figure. Specifically: the initial feature map F of the second convolution input is divided into C×H×W, where C represents the number of input channels, H represents the height of the feature map, and W represents the width of the feature map. The input feature map is weighted in the horizontal and vertical directions respectively. In the horizontal direction, an adaptive average pooling operation is first used to pool the width dimension to 1 to obtain a C×H×1 feature map. The width dimension is removed to obtain a C×H feature map. Secondly, a 1D convolution operation is performed on the pooled height direction tensor to perform a linear transformation on the input features to enhance the features. Then, multiple groups of features are group normalized G. n The normalized feature map is weighted and the width dimension is added after the Sigmoid activation operation to generate the weighted features in the horizontal direction. Figure X h , the calculation formula in the horizontal direction is:

[0064] X h =σ(G n (F h (Z h (F))))

[0065] Among them, σ is the Sigmoid activation function, F h is a 1D convolution, Z h is the average pooling feature in the horizontal direction.

[0066] In the vertical direction, we first use an adaptive average pooling operation to pool the height dimension to 1 to obtain a C×1×W feature map. Then, we remove the height dimension to obtain a C×W feature map. Then, we perform a 1D convolution operation on the width direction tensor after pooling to perform a linear transformation on the input features and enhance the features. Then, we group normalize multiple groups of features. n The normalized feature map is weighted and the height dimension is added after the Sigmoid activation operation to generate the weighted features in the vertical direction. Figure X w , the calculation formula in the vertical direction is:

[0067] X w =σ(G n (F w (Z w (F))))

[0068] Among them, σ is the Sigmoid activation function, F w is a 1D convolution, Z w It is the average pooling feature in the vertical direction.

[0069] The original feature Figure X , horizontal characteristics Figure X h and vertical characteristics Figure X w Fusion is performed to output the final feature map Y:

[0070] Y=X×X h ×X w .

[0071] (3-c) Import the validation set, use the preliminary model, calculate the loss value, and then perform back propagation and parameter update until the loss function value is minimized and tends to be stable, the iteration stops, and the final detection model is obtained. The improved YOLOv8 algorithm can accurately identify different types of diseased fish and meet real-time requirements.

[0072] (4) The counter 23 of each fish outlet channel 18 records the number of the fish outlet channels 18 and transmits the counting information to the counting unit of the controller 12, which classifies and summarizes the information; Fig.14 As shown, the counting units are classified and summarized, counter 23-1 records the number of healthy fish, thereby obtaining the number of healthy fish A, counter 23-2 records the number of Class I diseased fish, the number of Class I diseased fish B, counter 23-3 records the number of Class II diseased fish, the number of Class II diseased fish C, counter 234 records the number of Class III diseased fish, the number of Class III diseased fish D. Specifically, the total number of fish = A+B+C+D, the total number of diseased fish = B+C+D.

Claims

1. A farmed fish classification device, Features: The invention comprises a fish pond to be classified (1), a healthy fish pond (2), a plurality of diseased fish ponds, a fish school channel component whose inlet end is connected to the fish pond to be classified (1) and whose outlet ends are respectively connected to the healthy fish pond (2) and the plurality of diseased fish ponds, a camera (11) for photographing the fish to be classified in the fish school channel component, a visual processing unit for receiving images photographed by the camera (11) and identifying the types of healthy fish and diseased fish according to the images, a fish sorting component (21) located in the fish school channel component and used for sorting healthy fish and diseased fish into different channels, a diseased fish sorting component (22) located in the fish school channel component and used for sorting diseased fish of different types into different channels, and a controller (12) for controlling the fish sorting component (21) and the diseased fish sorting component (22) according to the identification result of the visual processing unit.

2. The farmed fish sorting device according to claim 1, Features: The fish school channel assembly comprises a fish inlet channel (13), a fish identification channel (14) and a fish sorting channel (16) with two outlets, which are sequentially connected to a fish pond (1) to be sorted, wherein one outlet of the fish sorting channel (16) is connected to a healthy fish pond (2) via a healthy fish channel (15), and the other outlet of the fish sorting channel (16) is connected to a plurality of sick fish ponds via a sick fish channel (17) with multiple outlets; and fish outlet channels (18) are installed at the outlet end of the healthy fish channel (15) and the multiple outlet ends of the sick fish channel (17).

3. The farmed fish sorting device according to claim 2, Features: The fish outlet passage (18) is provided with a pneumatic acceleration component (10) for preventing fish from being trapped and blocked at the end of the fish outlet passage (18). The pneumatic acceleration component (10) comprises an air pump (101) and a delivery hose (102) connected between the air pump (101) and the fish outlet passage (18). A counter (23) connected to a controller (12) is provided at the end of the fish outlet passage (18). The controller (12) controls the action of the air pump (101) according to the count of the counter (23).

4. The farmed fish sorting device according to claim 2, Features: The diameter of the fish inlet passage (13) gradually decreases along the direction of water flow. A first fish limiting component (19) and a second fish limiting component (20) for limiting the number of fish entering are sequentially arranged inside the fish inlet passage (13). The first fish limiting component (19) comprises a first upper limit fish baffle (191) and a first lower limit fish baffle (192) symmetrically rotatably mounted on the inner wall of the fish inlet passage (13). A first upper limit spring (193) is mounted between the first upper limit fish baffle (191) and the inner wall of the fish inlet passage (13). A first lower limit spring (194) is mounted between the first lower limit fish baffle (192) and the inner wall of the fish inlet passage (13). When the fish separator (191) and the first lower limit fish separator (192) are opened to the limit position, two fish can be allowed to pass through; the second fish limiting component (20) comprises a second upper limit fish separator (201) and a second lower limit fish separator (202) symmetrically rotatably mounted on the inner wall of the fish inlet passage (13); a second upper limit spring (203) is mounted between the second upper limit fish separator (201) and the inner wall of the fish inlet passage (13); a second lower limit spring (204) is mounted between the second lower limit fish separator (202) and the inner wall of the fish inlet passage (13); when the second upper limit fish separator (201) and the second lower limit fish separator (202) are opened to the limit position, one fish can be allowed to pass through.

5. The farmed fish sorting device according to claim 2, Features: The fish sorting assembly (21) comprises a fish sorting sealing box (215) arranged outside the fish sorting channel (16), a first motor (214) installed in the fish sorting sealing box (215) and having an output shaft extending into the fish sorting channel (16), a first sealing ring (212) used for sealing the connection between the output shaft of the first motor (214) and the fish sorting channel (16), a fish sorting partition (211) connected to the output shaft of the first motor (214) and located in the fish sorting channel (16), and a first magnet (213) fixedly installed on the outside of the fish sorting partition (211) and used for fixing the fish sorting partition (211) after the first motor (214) is powered off, wherein the fish sorting partition (211) is located at a three-port intersection node of the fish sorting pipeline (16), and the first motor (214) drives the fish sorting partition (211) to rotate to realize the opening and closing of the two outlets of the fish sorting channel (16).

6. The farmed fish sorting device according to claim 2, Features: The sick fish classification component (22) comprises a sick fish sealing box (225) arranged outside the sick fish passage (17), a second motor (224) installed in the sick fish sealing box (225) and having an output shaft extending into the sick fish passage (17), a second sealing ring (222) used to seal the connection between the output shaft of the second motor (224) and the sick fish passage (17), a sick fish classification partition (221) connected to the output shaft of the second motor (224) and located in the sick fish passage (17), and a second magnet (223) fixedly installed on the outside of the sick fish classification partition (221) and used to fix the sick fish classification partition (221) after the second motor (224) is powered off, wherein the sick fish classification partition (221) is located at a multi-port intersection node of the sick fish passage (17), and the second motor (224) drives the sick fish classification partition (221) to rotate to realize the opening and closing of the multiple outlets of the sick fish passage (17).

7. The farmed fish sorting device according to claim 2, Features: Auxiliary water inlet pipes (9) are installed at the ends of the two outlets of the fish sorting channel (16), and the other ends of the auxiliary water inlet pipes (9) extend into the fish pond (1) to be sorted. Filter screens (24) are installed at both ends of the auxiliary water inlet pipes (9).

8. The farmed fish sorting device according to claim 1, Features: The water levels of the healthy fish pond (2) and the plurality of diseased fish ponds are lower than the water level of the fish pond to be classified (1); the inner wall of the fish pond to be classified (1) is installed with a fish repellent (6); and the inner walls of the healthy fish pond (2) and the plurality of diseased fish ponds are installed with fish attracting lamps (7).

9. A method for classifying farmed fish, Features: The following steps are involved: The fish enter the fish entry channel (13), and the first fish limiting component (19) and the second fish limiting component (20) allow the fish to enter the fish identification channel (14) one by one; The camera (11) acquires images of the fish passing through the fish identification channel (14) one by one, and transmits the images to the visual processing unit of the controller (12); The detection model carried by the visual control unit recognizes and judges the input fish image, and transmits the recognition and judgment result to the central control unit of the controller (12), and the central control unit controls the rotation of the first motor (214) and the second motor (224) to realize automatic classification; The counter (23) of each fish outlet channel (18) records the number of the fish outlet channel (18) and transmits the counting information to the counting unit of the controller (12). The counting unit classifies and summarizes the information and transmits the counting data to the central control unit, which controls the action of the pneumatic acceleration component (10).

10. The method for classifying farmed fish according to claim 9, Features: The training steps of the detection model carried by the visual control unit are as follows: Collect image data related to various types of diseased fish in farms, use Labelimg annotation tools to annotate them, perform data expansion, establish a diseased fish dataset, and divide the dataset into a training set and a validation set; Improve the YOLOv8 algorithm model, build a CCE module, and replace all bottlenecks in C2f with the CCE module to form a C2f-CCE module, and replace all C2f in Backbone and Neck with C2f-CCE modules; the CCE module consists of a Conv convolution module and an ELA efficient local attention module; the Conv convolution module includes two, the first one is used to reduce the dimension of the input feature map and extract preliminary features, and the second one is used to perform group convolution operations to achieve dimensionality increase and deep-level feature extraction; the ELA efficient local attention module performs weighted processing on the output feature map of the second convolution to enhance important features and suppress irrelevant or redundant features; The training set is imported into the improved YOLOv8 algorithm model. First, the features of the image are extracted using the Backbone feature extraction part. Then, in the Neck part, the extracted image features are upsampled, downsampled, and feature concatenated to obtain a fused feature map. Finally, a preliminary model is obtained in the Head part. Import the validation set, use the preliminary model, calculate the loss value, and then perform backpropagation and parameter update until the loss function value is minimized and tends to be stable. The iteration stops and the final detection model is obtained.

Citation Information

Cited By

  • Automatic fish sorting method, device and equipment based on visual large model

    CN121121456A