Classification putting equipment suitable for fry breeding and control algorithm of classification putting equipment

Abstract

The invention relates to the technical field of fry screening, in particular to graded throwing equipment suitable for fry breeding and a control algorithm thereof.The graded throwing equipment comprises a box body, a fixing frame is fixedly connected to the inner wall of the box body, a center control device is fixedly connected to the outer wall of the fixing frame, and a display device is fixedly connected to the outer wall of the center control device; the outer wall of the fixing frame is fixedly connected with a recognition device; the pushing device is located in the box body, the pushing device is used for screening fish fries of different sizes, and through the design of the pushing device, the fish fries of different sizes can be accurately separated into different areas according to the set fish fry size grading standard, that is, small fish fries swim into a small fish area at the left end of the box body; large fries enter the adult fry area at the right end, so that complexity and inaccuracy of manual screening are avoided, the efficiency and accuracy of fry grading are improved, and a good foundation is provided for subsequent differentiated breeding of fries of different specifications.

Description

Technical Field

[0001] The present invention relates to the technical field of fry screening, and specifically to a grading and feeding device suitable for fry breeding and its control algorithm. Background Art

[0002] With the transformation and upgrading of China's aquaculture industry and the deployment of the smart fishery strategy, the aquaculture area has been increasing year by year, and the current total aquaculture output has ranked first in the world. The key pain points in the automated aquaculture process are as follows: problems such as disordered feeding, feed waste, and predation between large and small fish are particularly serious during the breeding process, which have a serious impact on the quality and efficiency of fishery breeding. The main reason for the above situation is that during the aquaculture process, it is necessary to efficiently and accurately sort and grade fry at different growth stages according to their sizes, so as to meet the requirement of pond breeding. However, at present, the sorting operation mainly relies on manual labor, and the special equipment for fry sorting is relatively backward, making it difficult to achieve precise grading according to the characteristics of the fry themselves, and thus unable to achieve high-quality fry sorting and intelligent feeding. Summary of the Invention

[0003] (I) Technical Problems to be Solved Aiming at the deficiencies of the prior art, the present invention provides a grading and feeding device suitable for fry breeding and its control algorithm, solving the problems that the current sorting operation mainly relies on manual labor, the special equipment for fry sorting is relatively backward, it is difficult to achieve precise grading according to the characteristics of the fry themselves, and thus unable to achieve high-quality fry sorting and intelligent feeding.

[0004] (II) Technical Solutions To achieve the above objectives, the present invention is implemented through the following technical solutions: A grading and feeding device suitable for fry farming, comprising: a box body, the inner wall of the box body is fixedly connected with a fixing frame, the outer wall of the fixing frame is fixedly connected with a central control device, the outer wall of the central control device is fixedly connected with a display device, and the outer wall of the fixing frame is fixedly connected with an identification device; a pushing device, the pushing device is located inside the box body, and the pushing device is used to screen fry of different sizes; the pushing device includes a limiting frame, the outer wall of the limiting frame is rotatably connected with a lead screw, the outer wall of the limiting frame is symmetrically and fixedly connected with limiting rods, one end of the lead screw away from the limiting frame is provided with a lead screw motor, the bottom end of the lead screw is provided with a screening plate, the bottom end of the screening plate is provided with an inclined plate, the outer wall of the screening plate is fixedly connected with a water scraping strip, the outer wall of the screening plate is symmetrically and fixedly connected with fixing parts through nuts, the outer wall of the screening plate is symmetrically and fixedly connected with driving parts through nuts, and the inner wall of the driving part is fixedly connected with a rotating ring. When the lead screw rotates, since the inner wall of the rotating ring is threadedly connected with the outer wall of the lead screw, and the inner wall of the fixing part slides on the outer wall of the limiting rod, the screening plate can move upward along the limiting rod under the pulling of the lead screw motor, so as to screen fry of different sizes. During the screening process, small fry will swim through the elliptical holes, flow holes, fish passing holes and water leakage holes of the inclined plate on the screening plate into the small fry area at the left end of the box body, while large fry will be pushed and pulled by the screening plate to the top of the inclined plate and then enter the adult fry area at the right end of the box body.

[0005] Preferably, the outer wall of the limiting frame is fixedly connected with the inner wall of the box body, the output end of the lead screw motor is fixedly connected with the outer wall of the lead screw, the outer wall of the lead screw motor is fixedly connected with the outer wall of the fixing frame, one end of the limiting rod away from the limiting frame is fixedly connected with the outer wall of the fixing frame, and the outer wall of the inclined plate is fixedly connected with the inner wall of the box body.

[0006] Preferably, the inner wall of the fixing part slides on the outer wall of the limiting rod, the inner wall of the rotating ring is threadedly connected with the outer wall of the lead screw, the outer wall of the water scraping strip slides on the inner wall of the box body, and the outer wall of the water scraping strip slides on the outer wall of the inclined plate. Since the screening plate needs to push and pull most of the water in the middle, in order to avoid breaking the screening plate, an acrylic plate with a thickness of 4 mm is used. There are 8-mm flow holes on both sides to screen fry of different sizes. Distributed holes are made in the middle, and 4 elliptical holes in the shape of rectangular blocks are designed on the four sides in the middle. The length of the elliptical hole diameter is 22 mm and the width is 8 mm, which is convenient for water flow and small fry to swim through the holes easily. Water scraping strips are hung on the left and right sides and the lower side of the screening plate to buffer the water body and gather the fry with the water scraping strips, preventing the fry from being injured due to too fast water flow.

[0007] Preferably, elliptical holes are formed in the wall of the screening plate, flow holes are symmetrically formed in the inner wall of the screening plate, fish passing holes are formed in the inner wall of the inclined plate, and water leakage holes are formed in the inner wall of the inclined plate. The inclined plate is made of acrylic material. The inclined plate cooperates with the screening plate to screen small fish and big fish. Through the investigation and screening of most fry on the market, the survival rate of most small fry is relatively high after the size reaches 7.5 mm. To prevent small fish from entering the big fry area when the survival rate is relatively low, the aperture size of the fish passing holes is 8 mm. Since the pressure difference in water is large and the volume of water is heavy, four water leakage holes are arranged in the middle to connect the water above the inclined plate and the water below the inclined plate, preventing the heavy volume of water from breaking the acrylic plate. Small fry can also return to the small fish area through the fish passing holes and water leakage holes in the inclined plate.

[0008] Preferably, a stretching motor is fixedly connected to the outer wall of the box body, a fish discharging port is formed in the inner wall of the box body, a plug plate is slidably connected to the outer wall of the box body, and the output end of the stretching motor is fixedly connected to the outer wall of the plug plate through a pull rope. When the central control device determines that the value of the fry reaches the preset value, the stretching motor will be started. The output end of the stretching motor is fixedly connected to the outer wall of the plug plate through a pull rope. At this time, the plug plate moves and the fish discharging port is opened, thus completing the classified feeding work of the fry.

[0009] A control method for classified feeding applicable to fry breeding includes the following steps: Step 1: Input screening information into the central control device, and then pour the fry between the inclined plate and the screening plate to complete the preparation for fry screening; Step 2: Start the lead screw motor through the central control device, and the lead screw motor pulls the screening plate to move for screening small and big fry; Step 3: After the recognition device finds that a living thing enters the adult fry area, it enters the working state for intelligent recognition, and displays the number of fry through the display device; Step 4: The recognition device transmits the recognized data to the central control device. When the value of the fry reaches the set value, the central control device starts the stretching motor to open the fish discharging port for fry feeding, thus completing the classified feeding of the fry.

[0010] Preferably, the central control device includes a voltage conversion module, a motor drive module, a voice recognition module, and a K210 recognition module; the voltage conversion module is used for voltage conversion; the motor drive module is used to control the start of the lead screw motor and the stretching motor; the voice recognition module enables users to perform voice control at a relatively long distance to prepare for the next batch of fry to be sorted, and there is no need to manually press the screening switch when the hands are wet; the K210 recognition module is used to recognize the fry through a visual recognition model. The developed machine vision tracking fry counting and recognition algorithm is practical and advanced. The fry are fed through voice control, reducing the risk of manual fish fishing injury and electric shock caused by wet hand operation.

[0011] To avoid burning out components due to unstable power input voltage, the voltage conversion module converts the voltage. The module input is 12V, which is respectively converted into 12V, 5V, and 3.3V. Among them, 12V powers the two-phase four-wire stepper motor, and 5V powers the STC32G12K128 chip and the drive board respectively; The STC32G12K128 single-chip microcomputer is selected as the main control chip of this system, and a chip in the ESP8266 Internet of Things field is selected. It can provide the following standard functions: built-in WiFi module, which can realize wireless network connection and support the 802.11 b / g / n protocol. It supports the TCP / IP protocol stack and can realize network data transmission and communication; The MaixBit development board of the K210 recognition module uses K210 as the core unit. In terms of Al machine vision performance, it has a variety of built-in hardware acceleration units, and the total computing power can reach up to 1TOPS. It can conveniently implement machine vision algorithms for various application scenarios, and its recognition module is used to identify and count large fry and then control the number of fry put in; Preferably, the establishment of the visual recognition model includes the following steps: Step S1: In the DarknetConv2D of the YOLOv3-Tiny model, the self-regularized non-monotonic neural activation function Mish is improved to the scaled exponential linear unit SELU; Step S2: Introduce the grouped convolution strategy in ResNeXt and the improvement on the CSPNet structure, control the number of groups through variable cardinality, and integrate the feature maps at the beginning and end of the network stage; Step S3: Introduce the improvement on channel pruning, add L1 regularization to constrain the BN layer coefficients to make the coefficients sparse, and at the same time crop the convolution kernels of the corresponding channels in the front of the BN layer to reduce the number of parameters; Step S4: Apply the OTSU algorithm to perform binary processing on the image to be recognized, and superimpose the image processed by the OTSU algorithm on the image to be processed processed by the CLAHE algorithm.

[0012] The model improves the activation function of DarknetConv2D in Yolov3-tiny, improves the self-regularized non-monotonic neural activation function Mish, and uses the modified scaled exponential linear unit SELU. This activation function can induce the self-normalization property, thus avoiding the explosion and disappearance of gradients. The SELU function multiplies the ELU function by the coefficient λ, that is, SELU(x)=λ×ELU(x). In this model, by modifying the SELU function:

[0013] For the positive value part in, increase the smoothness of its positive offset , so that it has faster convergence compared to SELU, and at the same time is closer to the natural gradient, avoiding the occurrence of gradient explosion. The values of the hyperparameters α and λ here are: α = 1.6732632423543772848170429916717, λ = 1.0507009873554804934193349852946 To improve the detection speed of the algorithm in the embedded classification model, a grouped convolution strategy between ordinary convolution kernels and depthwise separable convolutions and improvements on the cross-stage local network structure are introduced. The number of groups is controlled by the variable base to integrate the feature maps at the beginning and end of the network stage to respect the variability of the gradient. This model improves the Bottleneck in the backbone network of the Yolov3-tiny model in the channel dimension, asymmetrically groups the input and output, and improves the feature pointing of a single parameter through a smaller number of unified units; To improve the recognition speed of the model and reduce the number of parameters, here the improvement by Zhuang L et al. on channel pruning is introduced. By adding L1 regularization to constrain the coefficients of the BN layer, the coefficients are made sparse, and at the same time, the convolution kernels corresponding to the channels in front of the BN layer are cropped. The L1 regularization constraint used here is: Among them, the first term is the loss function, the second term introduces the constraint g(s)=|s|, x and y are the training samples and corresponding labels, W is the weight coefficient vector, λ is the regularization coefficient, and both γ and Γ are target weight penalty terms. The parameters can be made sparse according to the adjustment of the dataset; To improve the recognition accuracy of the model, special grayscale processing needs to be performed on the image to be recognized: the image to be processed is binarized based on the OTSU algorithm. This algorithm is simple to calculate and is not affected by the brightness and contrast of the image; the image processed by the OTSU algorithm is superimposed on the image to be processed processed by the CLAHE algorithm. Thereby enhancing the contrast of the image to be detected and suppressing the noise of the image to be detected.

[0014] (III) Beneficial effects The present invention provides a grading and feeding device suitable for fry farming and its control algorithm. It has the following beneficial effects: 1. Through the design of the pushing device, it is possible to accurately separate fry of different sizes into different areas according to the set fry size grading standard. That is, small fry swim into the small fry area at the left end of the box body, and large fry enter the adult fry area at the right end, avoiding the cumbersome and inaccurate manual screening, greatly improving the efficiency and accuracy of fry grading, and providing a good foundation for the differential cultivation of fry of different specifications in the follow-up; 2. Equipped with an identification device and a central control device, the identification device can intelligently identify the fry entering the adult fry area and transmit the data to the central control device. The central control device can automatically start the stretching motor according to the preset fry values to control the opening and closing of the fish discharge port, thus completing the accurate placement of fry. This intelligent operation not only reduces the labor input but also ensures that the number of fry placed each time meets the breeding requirements, effectively improving the scientificity and standardization of fry breeding; 3. Realize the automatic grading and screening of fry and the accurate placement of fry, enabling fry of different specifications to grow in a suitable environment, reducing problems such as growth differences, resource competition, and disease transmission caused by the mixing of fry specifications, helping to improve the survival rate and growth rate of fry, and then enhancing the economic and production benefits of fry breeding and promoting the development of the fry breeding industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic structural diagram of the whole of the present invention; Figure 2 It is a schematic internal structure diagram of the present invention; Figure 3 It is a schematic structural diagram of the driving member of the present invention; Figure 4 It is a schematic structural diagram of the screening plate of the present invention; Figure 5 It is a schematic structural diagram of the inclined plate of the present invention.

[0016] In the figure: 1. Box body; 2. Fish discharge port; 3. Fixed frame; 4. Pushing device; 41. Lead screw motor; 42. Lead screw; 43. Screening plate; 44. Inclined plate; 45. Limiting frame; 46. Limiting rod; 47. Driving member; 48. Swivel ring; 49. Wiper strip; 410. Oval hole; 411. Flow hole; 412. Fish passing hole; 413. Leakage hole; 414. Fixing member; 5. Central control device; 6. Identification device; 8. Stretching motor; 9. Plug plate; 10. Display device. DETAILED DESCRIPTION OF THE INVENTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] Please refer to Figures 1-5 , the present invention provides a technical solution: a grading and feeding device suitable for fry farming, including: a box body 1, a fixing frame 3 is fixedly connected to the inner wall of the box body 1, a central control device 5 is fixedly connected to the outer wall of the fixing frame 3, a display device 10 is fixedly connected to the outer wall of the central control device 5, and an identification device 6 is fixedly connected to the outer wall of the fixing frame 3; A pushing device 4, the pushing device 4 is located inside the box body 1, and the pushing device 4 is used for screening fry of different sizes; The pushing device 4 includes a limiting frame 45, a lead screw 42 is rotatably connected to the outer wall of the limiting frame 45, limiting rods 46 are symmetrically and fixedly connected to the outer wall of the limiting frame 45, one end of the lead screw 42 away from the limiting frame 45 is provided with a lead screw motor 41, the bottom end of the lead screw 42 is provided with a screening plate 43, the bottom end of the screening plate 43 is provided with an inclined plate 44, a water scraping strip 49 is fixedly connected to the outer wall of the screening plate 43, fixing members 414 are symmetrically and fixedly connected to the outer wall of the screening plate 43 through nuts, driving members 47 are symmetrically and fixedly connected to the outer wall of the screening plate 43 through nuts, and a rotating ring 48 is fixedly connected to the inner wall of the driving member 47.

[0019] The outer wall of the limiting frame 45 is fixedly connected to the inner wall of the box body 1, the output end of the lead screw motor 41 is fixedly connected to the outer wall of the lead screw 42, the outer wall of the lead screw motor 41 is fixedly connected to the outer wall of the fixing frame 3, one end of the limiting rod 46 away from the limiting frame 45 is fixedly connected to the outer wall of the fixing frame 3, and the outer wall of the inclined plate 44 is fixedly connected to the inner wall of the box body 1.

[0020] The inner wall of the fixing member 414 is slidably connected to the outer wall of the limiting rod 46, the inner wall of the rotating ring 48 is threadedly connected to the outer wall of the lead screw 42, the outer wall of the water scraping strip 49 is slidably connected to the inner wall of the box body 1, and the outer wall of the water scraping strip 49 is slidably connected to the outer wall of the inclined plate 44.

[0021] An elliptical hole 410 is opened in the wall of the screening plate 43, flow holes 411 are symmetrically opened in the inner wall of the screening plate 43, a fish passing hole 412 is opened in the inner wall of the inclined plate 44, and a water leakage hole 413 is opened in the inner wall of the inclined plate 44.

[0022] A stretching motor 8 is fixedly connected to the outer wall of the box body 1, a fish discharge port 2 is opened in the inner wall of the box body 1, a plug plate 9 is slidably connected to the outer wall of the box body 1, and the output end of the stretching motor 8 is fixedly connected to the outer wall of the plug plate 9 through a pulling rope.

[0023] A control method for hierarchical feeding applicable to fry farming, comprising the following steps: Step 1: Input screening information into the central control device 5, and then pour the fry between the inclined plate 44 and the screening plate 44 to complete the preparation for fry screening; Step 2: Start the lead screw motor 41 through the central control device 5, and the lead screw motor 41 pulls the screening plate 43 to move for screening of large and small fry; Step 3: After the recognition device 6 detects that a living object enters the fry area, it enters the working state for intelligent recognition, and displays the number of fry through the display device 10; Step 4: The recognition device 6 transmits the recognized data to the central control device 5. When the value of the fry reaches the set value, the central control device 5 starts the stretching motor 8 to open the fish discharge port 2 for fry feeding, completing the hierarchical feeding of the fry.

[0024] The central control device 5 includes a voltage conversion module, a motor drive module, a voice recognition module, and a K210 recognition module; the voltage conversion module is used to convert multiple voltage output forms for the motor drive module; the motor drive module is used to control the start of the lead screw motor 41 and the stretching motor 8; the voice recognition module is used for users to perform voice control at a relatively long distance to prepare for the next batch of fry to be sorted, and there is no need to manually press the screening switch when the hands are wet; the K210 recognition module is used to recognize the fry through a visual recognition model.

[0025] The establishment of the visual recognition model includes the following steps: Step S1: In the DarknetConv2D of the YOLOv3-Tiny model, improve the self-regularized non-monotonic neural activation function Mish to the scaled exponential linear unit SELU; Step S2: Introduce the grouped convolution strategy in ResNeXt and the improvement on the CSPNet structure, control the number of groups through variable cardinality, and integrate the feature maps at the beginning and end of the network stage; Step S3: Introduce the improvement on channel pruning, add L1 regularization to constrain the BN layer coefficients to make the coefficients sparse, and at the same time crop the convolution kernels of the corresponding channels in the front of the BN layer to reduce the number of parameters; Step S4: Apply the OTSU algorithm to perform binary processing on the image to be recognized, and superimpose the image processed by the OTSU algorithm on the image to be processed processed by the CLAHE algorithm.

[0026] In use, first, pour the fry into the area between the inclined plate 44 and the screening plate 43. Before that, screening information such as the grading standard of fry size needs to be input into the central control device 5 first. Then, the central control device 5 starts the lead screw motor 41. Since the output end of the lead screw motor 41 is fixedly connected to the outer wall of the lead screw 42, when the lead screw 42 rotates, because the inner wall of the rotating ring 48 is threadedly connected to the outer wall of the lead screw 42, and the inner wall of the fixing member 414 is slidably connected to the outer wall of the limiting rod 46, the screening plate 43 can move upward along the limiting rod 46 under the pulling of the lead screw motor 41, so as to screen the fry of different sizes. During the screening process, the small fry will swim into the small fry area at the left end of the box body 1 through the elliptical holes 410, flow holes 411 of the screening plate 43, the fish passing holes 412 and water leakage holes 413 of the inclined plate 44, while the large fry will be pushed and pulled by the screening plate 43 to the top of the inclined plate 44 and then enter the adult fry area at the right end of the box body 1; When the recognition device 6 detects that a living object enters the adult fry area, it will end the sleep state and enter the working state, intelligently recognize the fry, and display the number of fry through the display device 10. Then, the recognition device 6 transmits the recognized data to the central control device 5. When the central control device 5 determines that the value of the fry reaches the preset value, it will start the stretching motor 8. The output end of the stretching motor 8 is fixedly connected to the outer wall of the plug plate 9 through a pull rope. At this time, the plug plate 9 moves, and the fish discharge port 2 is opened, thus completing the grading and placing work of the fry, realizing the functions of automatically grading and screening the fry according to size and accurately placing them, which helps to improve the efficiency and quality of fry breeding and meet the need for separate breeding of fry of different specifications during the fry breeding process.

[0027] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0028] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A grading delivery device suitable for fish fry farming, characterized in that: include: A box body (1), wherein the inner wall of the box body (1) is fixedly connected to a fixing frame (3), the outer wall of the fixing frame (3) is fixedly connected to a central control device (5), the outer wall of the central control device (5) is fixedly connected to a display device (10), and the outer wall of the fixing frame (3) is fixedly connected to an identification device (6); A pushing device (4), the pushing device (4) being located inside the box (1), and the pushing device (4) being used to screen fry of different sizes; The pushing device (4) comprises a limit frame (45), the outer wall of the limit frame (45) is rotatably connected to a screw rod (42), the outer wall of the limit frame (45) is symmetrically fixedly connected to a limit rod (46), one end of the screw rod (42) away from the limit frame (45) is provided with a screw motor (41), the bottom end of the screw rod (42) is provided with a screening plate (43), the bottom end of the screening plate (43) is provided with an inclined plate (44), the outer wall of the screening plate (43) is fixedly connected to a wiper strip (49), the outer wall of the screening plate (43) is symmetrically fixedly connected to a fixing member (414) via a nut, the outer wall of the screening plate (43) is symmetrically fixedly connected to a driving member (47) via a nut, and the inner wall of the driving member (47) is fixedly connected to a swiping ring (48).

2. The grading delivery device suitable for fry farming according to claim 1, characterized in that: The outer wall of the limit frame (45) is fixedly connected to the inner wall of the box body (1), the output end of the screw motor (41) is fixedly connected to the outer wall of the screw (42), the outer wall of the screw motor (41) is fixedly connected to the outer wall of the fixing frame (3), one end of the limit rod (46) away from the limit frame (45) is fixedly connected to the outer wall of the fixing frame (3), and the outer wall of the inclined plate (44) is fixedly connected to the inner wall of the box body (1).

3. The grading delivery device suitable for fry farming according to claim 1, characterized in that: The inner wall of the fixing member (414) is slidably connected to the outer wall of the limiting rod (46), the inner wall of the rotating ring (48) is threadedly connected to the outer wall of the screw rod (42), the outer wall of the wiper strip (49) is slidably connected to the inner wall of the box body (1), and the outer wall of the wiper strip (49) is slidably connected to the outer wall of the inclined plate (44).

4. The grading delivery device suitable for fry farming according to claim 1, characterized in that: An elliptical hole (410) is provided in the wall of the screening plate (43), flow holes (411) are symmetrically provided on the inner wall of the screening plate (43), a fish hole (412) is provided on the inner wall of the inclined plate (44), and a water leakage hole (413) is provided on the inner wall of the inclined plate (44).

5. The grading delivery device suitable for fry farming according to claim 1, characterized in that: The outer wall of the box body (1) is fixedly connected to a stretching motor (8), the inner wall of the box body (1) is provided with a fish discharge port (2), the outer wall of the box body (1) is slidably connected to a blocking plate (9), and the output end of the stretching motor (8) is fixedly connected to the outer wall of the blocking plate (9) via a pull rope.

6. A control method for graded release of fry culture, used for the use of the graded release device for fry culture in claims 1-5, characterized in that: The following steps are involved: Step 1: input the screening information into the central control device (5), and then pour the fry between the inclined plate (44) and the screening plate (44), thereby completing the preparation for fry screening; Step 2: The central control device (5) starts the screw motor (41), and the screw motor (41) pulls the screening plate (43) to move and screen the large and small fry; Step 3: After the identification device (6) finds that a living thing has entered the frying area, it enters a working state to perform intelligent identification and displays the number of fry through the display device (10); Step 4: The identification device (6) transmits the identified data to the central control device (5). When the value of the fry reaches the set value, the central control device (5) starts the stretching motor (8), opens the fish outlet (2) to release the fry, and completes the graded release of the fry.

7. A control method for graded release of fry according to claim 6, characterized in that: The central control device (5) comprises a voltage conversion module, a motor drive module, a voice recognition module, and a K210 recognition module; the voltage conversion module is used for performing voltage conversion; the motor drive module is used for controlling the start of the lead screw motor (41) and the stretching motor (8); the voice recognition module is used for enabling the user to perform voice control at a long distance; and the K210 recognition module is used for identifying fry using a visual recognition model.

8. A control method for graded release of fry according to claim 7, characterized in that: The establishment of the visual recognition model comprises the following steps: Step S1, in DarknetConv2D of the YOLOv3-Tiny model, the self-regularized non-monotonic neural activation function Mish is improved to a scaled exponential linear unit SELU; Step S2: Introduce the grouped convolution strategy in ResNeXt and the improvement in the CSPNet structure, control the number of groups through the variable cardinality, and integrate the feature maps at the beginning and end of the network stage; Step S3: Introduce improvements in channel pruning, add L1 regularization to constrain the BN layer coefficients to make the coefficients sparse, and at the same time trim the convolution kernels of the corresponding channels at the front of the BN layer to reduce the number of parameters; Step S4: Apply the OTSU algorithm to perform binarization processing on the image to be recognized, and superimpose the image processed by the OTSU algorithm on the image to be processed by the CLAHE algorithm.

Images