A fully automatic sea cucumber high-throughput growth trait measurement and prediction device and method

Abstract

The application provides a kind of full-automatic sea cucumber high-throughput growth character measurement and prediction device and method, using the image acquisition module installed in the walking unit of support obtains the image containing sea cucumber target, image recognition module uses the visual recognition model of pre-construction and identifies sea cucumber target, with the longest place of sea cucumber target edge and the widest place as critical point, draw identification frame, data processing module calculates the actual size of identification frame, prediction model is based on the relationship between the actual size of identification frame and sea cucumber growth character and measures and predicts the growth character data of sea cucumber target, selects multiple fitting models with the highest confidence and confidence greater than threshold value, calculates the average value of the fitting model to obtain the measurement data of sea cucumber target length and width and the prediction data of wet weight and dry weight, can be automatically measured and predicted once in the growth character data of sea cucumber target in identification area.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sea cucumber cultivation, and provides a full-automatic sea cucumber high-throughput growth trait measurement and prediction device and method. BACKGROUND

[0002] Due to the unique nutritional value and market demand, sea cucumber has become a valuable fishery resource. In the future, with the economic development and acceptance, the demand for sea cucumber products of Chinese residents will further increase. In the process of sea cucumber cultivation, it is necessary to regularly evaluate the growth status and quantity of sea cucumber to determine whether the growth status meets the expectation. If abnormal conditions such as growth stagnation are found, measures such as adjusting the feed formula and improving the water quality can be taken in time to promote the normal growth of sea cucumber. If sea cucumber is counted and measured by artificial, it is time-consuming and laborious, and has a great influence on sea cucumber, which is easy to cause stress reaction of sea cucumber, and the sea cucumber will expel the internal organs to affect the growth of sea cucumber. Therefore, it is of great significance to provide a device for automatically measuring the quantity and growth status of sea cucumber for the sea cucumber cultivation industry. SUMMARY

[0003] In order to solve the problems in the prior art, the present application provides a full-automatic sea cucumber high-throughput growth trait measurement and prediction device, and provides a full-automatic sea cucumber high-throughput growth trait measurement and prediction method.

[0004] The present application first provides a full-automatic sea cucumber high-throughput growth trait measurement and prediction device, which comprises:

[0005] A support is arranged on a to-be-measured area, a linear guide rail is arranged in the support, and a walking unit movable along the linear guide rail is arranged in the linear guide rail;

[0006] An image acquisition module is arranged on the walking unit, a polarizing lens is arranged on a lens of the image acquisition module, and the image acquisition module is used for acquiring an image containing a sea cucumber target;

[0007] A distance measurement module is arranged on the walking unit, and is used for acquiring a distance between an optical center of the image acquisition module and a plane where the sea cucumber target is located;

[0008] An image recognition module acquires the image containing the sea cucumber target, uses a pre-constructed visual recognition model to recognize the sea cucumber target, draws a recognition box taking the longest part and the widest part of the edge of the sea cucumber target as critical points, records the size of the recognition box, sequentially sorts each recognition box in the recognition area, and counts the total number of sea cucumber target individuals;

[0009] A data processing module acquires the size of the recognition box and the distance from the image acquisition module to the plane where the sea cucumber target is located, and calculates the actual size of the recognition box based on the parameters of the image acquisition module.

[0010] a prediction module, which obtains the actual size of the identification frame, uses the actual size of the sea cucumber target identification frame as a prediction parameter in a pre-constructed prediction model, selects a plurality of fitting models with the highest confidence and a confidence greater than a threshold value, and calculates the average value of the fitting models to obtain the measurement data of the length and width of the sea cucumber target and the prediction data of the wet weight and dry weight.

[0011] Specifically, the linear guide is provided with a first motor, the first motor drives the movement of the walking unit, a steering device is further provided in the walking unit, a second motor for driving the rotation of the steering device is provided in the steering device, and the image acquisition module and the distance measuring module are provided in the steering device.

[0012] The application also provides a full-automatic sea cucumber high-throughput growth trait measurement and prediction method, comprising:

[0013] The calibrated image acquisition module photographs an image containing a sea cucumber target;

[0014] A pre-constructed visual recognition model identifies the sea cucumber target image, and draws an identification frame with the longest and widest parts of the sea cucumber target edge as critical points, and records the size of the identification frame;

[0015] The distance from the image acquisition module to the plane where the sea cucumber target is located is obtained, the actual size of the identification frame is calculated based on the image acquisition module parameters, and each identification frame is sequentially sorted, and the total number of sea cucumber target individuals is counted;

[0016] The actual size of the identification frame is loaded into the pre-constructed prediction model as a prediction parameter, the prediction model measures and predicts the growth trait data of the sea cucumber target based on the relationship between the actual size of the identification frame and the growth trait of the sea cucumber, selects a plurality of fitting models with the highest confidence and a confidence greater than a threshold value, and calculates the average value of the fitting models to obtain the measurement data of the length and width of the sea cucumber target and the prediction data of the wet weight and dry weight.

[0017] Specifically, the calculation method of the actual size is as follows: based on the calibrated image acquisition module parameters, the identification frame size and the distance from the image acquisition module to the plane where the sea cucumber target is located, the identification frame size is converted into a first actual size of the identification frame, the second actual size of the identification frame is calculated by using the plane perspective, the error between the first actual size and the second actual size is verified, the identification frame with an error value within an error threshold is obtained, and the average value of the first actual size and the second actual size is obtained to obtain the actual size of the identification frame.

[0018] Specifically, the calculation formula of the first actual size and the second actual size is as follows:

[0019] The first actual size includes a first actual length and a first actual width, the first actual length is calculated by,

[0020]

[0021] wherein, l a is the first actual length, l is a camera sensor length, H is a shooting height, P la is a bounding box length pixel number, P l is an image length pixel number, f x is a calibrated x-axis focal length, and μ1 is a first correction coefficient;

[0022] The first actual width is calculated by,

[0023]

[0024] wherein, w a is the first actual width, w is a camera sensor width, H is a shooting height, P wa is a bounding box width pixel number, P w is an image width pixel number, f y is a calibrated y-axis focal length;

[0025] The second actual size includes a second actual length and a second actual width, the second actual length is calculated by,

[0026]

[0027] wherein, l b is the second actual length, θ l is a calibrated horizontal viewing angle, μ2 is a second correction coefficient,

[0028] The second actual width is calculated by,

[0029]

[0030] wherein, w b is the second actual width, θ w is a calibrated vertical viewing angle.

[0031] Specifically, the image recognition module adopts an improved YOLOv8-seacucumber visual recognition model to recognize sea cucumbers, and identifies a target with a confidence greater than a threshold value as a sea cucumber target and draws an identification box, the YOLOv8-seacucumber visual recognition model comprises an input end, a backbone network and a detection head, on the basis of an original YOLOv8 visual recognition model, a C2f-F-SA module is used to replace a C2f module in the backbone network and the detection head, the C2f-F-SA module comprises a convolution layer, a segmentation layer, a plurality of FasterBlock structures, a connection layer and a convolution layer arranged in sequence, a bottleneck layer in the original C2f module is replaced by a FasterBlock structure, and a Shuffle Attention attention mechanism is added in the FasterNet Block module; a Conv module of the detection head part is replaced by an Adown module.

[0032] Specifically, the Shuffle Attention attention mechanism comprises an input end, a segmentation layer, an upper branch, a lower branch, a connection layer and an output end.

[0033] The input end inputs a sea cucumber feature Figure X ∈R (C，H，W) , the sea cucumber feature map is divided into G groups in the channel dimension, and is represented as:

[0034] X=[X1,…,X G ],X k ∈R (C / G×H×W) ,

[0035] Wherein C is the channel number of the sea cucumber feature map, H is the spatial height of the sea cucumber feature map, W is the spatial width of the sea cucumber feature map, R represents the set of all images in the data set, X k is the input feature;

[0036] The segmentation layer divides the input feature X k into an upper branch X K1 and a lower branch X K2 in the channel dimension, X K1 , X K2 ∈R (C / 2G×H×W) The upper branch adopts channel attention, generates channel information S through transformation parameters w1, b1 scaling and displacement, and then performs global average pooling (GAP) embedding global information to generate channel information S∈R (C / 2G×1×1) , the channel information is calculated by shrinking X k1 in the spatial dimension HxW,

[0037]

[0038] The channel information amount S is transformed by using a full connection layer and a sigma activation function (sigmoid) to generate a weight matrix and multiplied by X k1 , to obtain weighted features Figure X ′ k1 :

[0039] X′ k1 = sigma (F C (S)) cdot X k1 = sigma (w1S + b1) cdot X k1 ,

[0040] wherein w1 element-of R (C / 2G×1×1) , b1 element-of R (C / 2G×1×1) ;

[0041] The lower branch uses spatial attention to capture the spatial dependency between features, and the spatial information amount is scaled and shifted by transforming parameters w2, b2, and the spatial attention is generated by using group norm (GN):

[0042] X′ k2 = sigma (w2 cdot GN(X k2 ) + b2) cdot X k2 ,

[0043] wherein w2 element-of R (C / 2G×1×1) , b2 element-of R (C / 2G×1×1) ;

[0044] The connection layer combines X′ K1 and X′ K2 obtained by processing the upper and lower branches, so that the output feature map and the input size are the same,

[0045] X′ k = [X′ k1， X′ k2 ]R (C / G×H×W) .

[0046] Specifically, the ADown module includes an input end, an average pooling layer, a segmentation layer, a first path, a second path, a connection layer and an output end,

[0047] The input end inputs a feature Figure X element-of R (C，H，W) , the average pooling layer performs an average pooling operation on the input feature Figure X by using an average pooling kernel k with a size of 2 and a step s of 1 without padding (p = 0) to obtain an average pooling feature Figure X avg ,

[0048] X avg = Avgpool (X, k = 2, s = 1, p = 0),

[0049] Splitting layer splits the average-pooled feature Figure X avg into two sub-features along the channel dimension Figure X 1 and X2, the sub-features Figure X 1 and X2 are down-sampled by the first path and the second path respectively, to get feature maps Y1 and Y2,

[0050] The first path is that, the sub-feature Figure X 1 is directly down-sampled by a traditional convolutional layer to get the feature map Y1,

[0051] Y1 = Conv1(X1, k = 3, s = 2, p = 1);

[0052] The second path is that, the sub-feature Figure X 2 is firstly down-sampled by a max-pooling layer with a kernel size of 3, a stride of 2 and a padding of 1 to extract the maximum value X2 in the local window, highlighting the significant features, and then the feature map is further down-sampled by a convolutional layer with a kernel size of 1, a stride of 1 and a padding of 0 to extract and refine the features, to get the feature map Y2,

[0053] X 2，max = MaxPool(X2, k = 3, s = 2, p = 1),

[0054] Y2 = Conv2(X 2，max , k = 1, s = 1, p = 0),

[0055] Two learnable weight factors a and b are introduced in the connection layer, which are optimized through the training process to weight the down-sampling results, and the weight factors a and b control the contribution of Y1 and Y2 in the fusion process respectively, and the dimensions of the weight factors a and b are both 1x1x1, and the weighted fused feature map Y fusion is:

[0056] Y fusion = a · Y1 + b · Y2,

[0057] where · represents element-wise multiplication,

[0058] The weight factors a and b are normalized to ensure that their sum is 1,

[0059]

[0060]

[0061] The weighted fused feature map is recalculated using the normalized weight factors as,

[0062] Y fusion = anorm • Y1+β norm • Y2,

[0063] The weighted fused feature map Y fusion As output, the dimension is In the formula, C' is the number of output channels.

[0064] Specifically, the YOLOv8-seacucumber visual recognition model construction method is to group sea cucumbers according to their regions and varieties, and to sample and train them, to obtain photos of each sea cucumber under different shooting angles, different illuminations and different backgrounds in natural conditions, and to divide the pictures into training set and test set according to the proportion, to manually label the sea cucumbers in the training set using the boundary box, to train the model through the convolutional neural network, to test the model using the test set, to manually review the training results, and to obtain the visual recognition model.

[0065] Specifically, the prediction model construction method is to fit the actual size of the recognition box of the sea cucumber in the same region and the same variety with the growth trait data based on the labeled sea cucumber image and the sea cucumber growth trait data, to use multiple regression models and machine learning to train the model and obtain the prediction model.

[0066] Compared with the prior art, the present application has the following advantages:

[0067] The full-automatic sea cucumber high-throughput growth trait measurement and prediction device and method provided by the present application use the improved YOLOv8-seacucumber visual recognition model to recognize the sea cucumber target, draw a recognition box with the longest and widest parts of the sea cucumber target edge as the critical points, calculate the real size of the recognition box based on the pixel size of the recognition box, use the real size of the recognition box as the prediction parameter, select multiple fitting models using the pre-constructed prediction model, obtain the measurement data of the length and width of the sea cucumber target and the prediction data of the wet weight and dry weight, and thus achieve the purpose of measuring and predicting the growth traits of the sea cucumber. Moreover, the use of the actual size of the recognition box to predict the growth trait data can significantly reduce the calculation intensity and improve the detection efficiency.

[0068] The improved YOLOv8-seacucumber visual recognition model is used for identifying sea cucumbers, and the C2f-F-SA module is used to replace the C2f module in the backbone network and the detection head of the original YOLOv8 visual recognition model, the C2f-F-SA module comprises a convolution layer, a FasterBlock structure of a segmentation layer, a connection layer and a convolution layer arranged in sequence, the bottleneck layer in the original C2f module is replaced by the FasterBlock structure, so that the feature extraction efficiency of the network is improved, the calculation complexity in the backbone network is reduced, the device calculation resources are used more effectively, and the ability of the backbone network in spatial feature extraction is further improved, the key features of sea cucumbers in the image can be effectively extracted, the perception ability of the model to the spatial features of the input sea cucumbers is enhanced, the calculation amount is reduced, the detection speed is improved, and therefore the overall performance of the model in the sea cucumber image feature extraction task is improved. The Shuffle Attention attention mechanism is added in the FasterNet Block module, the convolution layer in the detection head part is replaced by the Adown module, potential information loss in the down-sampling process is prevented, the sensitivity of the model to the detail features is improved, and the weight factor learning mechanism is introduced to realize more fine down-sampling result fusion. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure X It is a schematic diagram of the overall structure of the application;

[0070] Figure 1 It is a schematic diagram of the use state of the application;

[0071] Figure 2 It is a flowchart of the application;

[0072] Figure 3 It is a calibration chessboard pattern;

[0073] Figure 4 It is a structure block diagram of the YOLOv8-seacucumber visual recognition model of the application;

[0074] Figure 5 It is a structure block diagram of the C2f-F-SA module of the application;

[0075] Figure 6 It is a structure block diagram of the FasterNet Block structure of the application;

[0076] Figure 7 It is a structure block diagram of the Shuffle Attention attention mechanism of the application;

[0077] Figure 8 It is a structure block diagram of the ADOWN module of the application;

[0078] Figure 9 YOLOv8-seacucumber visual recognition model recognition effect diagram of the present application;

[0079] Figure 10 Fitting model of sea cucumber quality and recognition box length and width. DETAILED DESCRIPTION

[0080] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0081] As shown in Figure 11 and Figure 1 , the present application first provides a full-automatic sea cucumber high-throughput growth trait measurement and prediction device, which comprises a support 1, an image acquisition module, a distance measurement module and a host 7, is suitable for a sea cucumber breeding pond 10, and measures and predicts the growth trait data of sea cucumber based on a prediction model by shooting sea cucumber images.

[0082] The support 1 is a detachable telescopic support, which is erected in the sea cucumber breeding pond 10 and comprises a support leg and a linear guide rail 2 detachably mounted on the support leg. A walking unit 3 is arranged in the linear guide rail 2. The support leg is of a telescopic structure, and the height of the support 1 can be adjusted according to the water depth in the sea cucumber breeding pond 10, and the height of the support 1 is fixed by using a bolt. The support 1 is designed to be detachable, which is convenient for storage when not in use and can be used after being assembled when in use.

[0083] A first motor 4 is arranged on one side of the linear guide rail 2, which drives the walking unit 3 to move along the linear guide rail 2. The image acquisition module and the distance measurement module are arranged in the walking unit 3. Generally, the environment of the sea cucumber breeding pond 10 is relatively dark, so a fill light is arranged in the walking unit 3 to improve the image clarity.

[0084] The image acquisition module is a camera 5, which shoots images containing sea cucumber targets. A polarizer is arranged on the lens of the camera 5, which is used to eliminate reflected light on the water surface and improve the clarity of shooting. The distance measurement module is a laser range finder 6, which is used to obtain the distance between the optical center of the image acquisition module and the plane where the sea cucumber target is located. When the laser beam emitted by the laser range finder 6 reaches the water surface, part of the echo will be reflected, and when the laser beam reaches the water bottom, part of the echo will be reflected again. The distance from the laser range finder 6 to the water surface and the water depth are calculated based on the time of measuring the two laser echoes, so as to obtain the distance from the laser range finder 6 to the plane where the sea cucumber target is located.

[0085] The camera is calibrated before shooting using Zhang's Calibration Method to obtain the actual external parameters and the actual internal parameters of the camera. The actual internal parameters of the camera include the camera sensor length, the camera sensor width, the image length pixel number, the image width pixel number, the focal length and the distortion coefficient. The actual external parameters of the camera include the horizontal viewing angle and the vertical viewing angle. The specific steps of camera calibration are as follows:

[0086] 1. Prepare the calibration board:

[0087] A high-precision checkerboard pattern with a known size is used. The checkerboard pattern is shown in Figure 2 The checkerboard pattern is black and white, and each square has a square block with a real size.

[0088] 2. Collect calibration images:

[0089] Fix the camera in a position to ensure that the camera does not move during the entire calibration process. Change the position and angle of the calibration board and take multiple images from different angles. Take at least 10 images to obtain more accurate results.

[0090] During shooting, ensure that different parts of the checkerboard are in the camera's field of view; ensure that the calibration board is shot at different angles and distances to increase diversity and obtain rich projection information; ensure that the image is clear and the corners are clearly visible.

[0091] 3. Extract the checkerboard corner points:

[0092] In the captured images, use the checkerboard corner point extraction algorithm in OpenCV to extract the inner corner points of each image (i.e. the four corner points of each small square). The corner point position of each checkerboard serves as the input data for calibration. These corner points serve as the geometric relationship data between the camera and the calibration board. The extraction steps are as follows:

[0093] (1) Corner detection: use the corner detection algorithm, i.e. cv: : findChessboardCorners in OpenCV to automatically identify the corner points of the checkerboard from the image;

[0094] (2) Sub-pixel level corner positioning: to improve accuracy, use the cv: : cornerSubPix function in OpenCV for sub-pixel level accurate positioning of the corner points;

[0095] (3) Store the corner point coordinates: the corner points of each image need to be saved and converted to two-dimensional coordinates in the image coordinate system.

[0096] 4. Construct the calibration matrix:

[0097] The corner point coordinates detected in each image are converted into coordinates in a world coordinate system, and an intrinsic matrix (containing focal length, principal point, etc.) and an extrinsic matrix (containing a rotation and a translation vector) of the camera are constructed;

[0098] The mathematical model of Zhang calibration is:

[0099] Suppose a point P in the world coordinate system w =(X, Y, Z) T , a point P in the camera coordinate system c =(X', Y', Z') T , and a coordinate p=(x, y) of the point P projected onto the image plane T The transformation is performed by a camera projection matrix P:

[0100]

[0101] In the formula, K is the intrinsic matrix of the camera, [R|t] is the extrinsic matrix of the camera, R is a rotation matrix, t is a translation vector, and (x, y) is an image coordinate;

[0102] The relationship between the image coordinate (x, y) and the normalized coordinate (x', y') is:

[0103]

[0104] In the formula, (c x , c y ) is the principal point coordinate, f x and f y are the focal lengths in the x and y directions;

[0105] The distortion model is:

[0106] x distorrted =(1+k1r 2 +k2r 4 +k3r 6 )+2p1x'y'+p2(r 2 +2x' 2 ),

[0107] y distorted =(1+k1r 2 +k2r 4 +k3r 6 )+p1(r 2 +2y' 2 )+2p2x'y',

[0108] where r 2 =x' 2 +y' 2, k1, k2, k3 are radial distortion coefficients, p1, p2 are tangential distortion coefficients;

[0109] The reprojection error is:

[0110]

[0111] where p i is the observed point in the image, P i is the point in the world coordinate system, K is the intrinsic matrix, R i and t i are the extrinsic matrix of the i-th image;

[0112] In the calibration process, the Levenberg-Marquardt algorithm is used to minimize the error between the actual corner points and the reprojection points on each image to estimate the intrinsic and extrinsic parameters of the camera. Iterative optimization is performed to continuously adjust the intrinsic and extrinsic parameters to minimize the overall reprojection error.

[0113] 5. Calculate camera parameters:

[0114] After calibration, the following important parameters are obtained:

[0115] Camera intrinsic matrix (Ihtrinsic Parameters):

[0116]

[0117] c x and c y are the principal point coordinates, usually close to the image center;

[0118] Distortion coefficients (Distortion Coefficients): including radial distortion coefficients (k1, k2, k3) and tangential distortion coefficients (P1, P2);

[0119] Camera extrinsic matrix (Extrinsic Parameters): including the rotation matrix R and the translation vector t of each image, used to convert the point in the world coordinate system to the camera coordinate system;

[0120] Reprojection error (Reprojection Error): represents the error between the theoretical projection and the actual corner point position in the image;

[0121] Correction coefficient: including the first correction coefficient μ1 and the second correction coefficient μ2, correcting the error between the calculated size and the actual size;

[0122] Use the focal length in the intrinsic matrix and the image resolution to calculate the horizontal and vertical viewing angles of the camera.

[0123] 6. Reprojection error analysis and result verification:

[0124] Using the obtained camera parameters, the corner coordinates in the world coordinate system are projected back to the image plane, and the error between the reprojected points and the actually detected corners is calculated to evaluate the accuracy of the calibration. Finally, the calibration results are verified by placing the calibration board at a known position and using the parameters obtained from the calibration to reconstruct its position in three-dimensional space.

[0125] A steering device can also be provided in the walking unit 3 to enable the camera 5 and laser range finder 6 installed in the walking unit 3 to be steered 90° along the length direction of the linear guide rail 2 to take pictures of sea cucumbers attached to the walls of the sea cucumber culture pond 10. The steering device includes a fixed frame and a second motor 8 installed on the outside of the fixed frame, a steering plate 9 is provided on the output shaft of the second motor 8, the camera 5, laser range finder 6 and fill light are arranged on the lower side of the steering plate 9, and the main machine 7 is arranged on the upper side of the steering plate 9. The second motor 8 can rotate clockwise or counterclockwise to make the steering plate 9 face the water surface or the side walls on both sides of the sea cucumber culture pond 10, respectively, to measure the distance and take pictures of sea cucumbers attached to the bottom of the pond and the walls on both sides of the pond.

[0126] The main machine 7 includes an image recognition module, a data processing module and a prediction module, and is in communication connection with the image acquisition unit and the distance measuring unit.

[0127] The image recognition module acquires images containing sea cucumber targets, uses a pre-constructed visual recognition model to recognize the sea cucumber targets, and only records sea cucumbers with the same feature points in consecutive images once.

[0128] As shown in Figure 4 The image recognition module uses an improved YOLOv8-seacucumber visual recognition model to recognize sea cucumbers and draw recognition boxes. The improved YOLOv8-seacucumber visual recognition model includes an input end (Input), a backbone network (Backbone) and a detection head (Head), the present application replaces the Bottleneck module in C2f with FasterNet Block module to obtain C2f-F (C2f-Faster) structure, and adds ShuffleAttention (SA) attention mechanism in FasterNet Block module, designs a new C2f-F-SA (C2f-Faster-ShuffleAttention) structure, and replaces the C2f structure in the traditional YOLOv8 visual recognition model; also replace all Conv modules in the Head part of YOLOv8 visual recognition model with Adown module. It provides strong technical support for the rapid and accurate identification of sea cucumbers.

[0129] As Figure 5 shown in FIG. 1, the C2f-F-SA module includes a convolutional layer (Conv), a split layer (Split), a plurality of FasterBlock structures, a concatenation layer (Concat), and a convolutional layer (Conv) arranged in sequence. The convolutional layer extracts features of the input data, the split layer divides the features into multiple branches, each branch further extracts features and fuses information through a FasterBlock structure composed of a plurality of sub-FasterBlock structures, the branches are spliced through the concatenation layer, and finally the convolutional layer performs final feature fusion and output.

[0130] The traditional C2f structure is composed of a series of bottleneck layers (Bottleneck), which are responsible for extracting and enhancing local features. In this technical solution, the bottleneck layer structure in the YOLOv8 model is replaced by the FasterBlock structure to improve the feature extraction efficiency of the network. This replacement strategy aims to reduce the computational complexity in the backbone network, thereby more effectively utilizing device computing resources and further improving the ability of the backbone network in spatial feature extraction. Through this replacement, the FasterBlock structure replaces the bottleneck layer structure in C2f, providing strong technical support for the fast and accurate identification of sea cucumbers.

[0131] The specific steps of replacing the bottleneck layer structure are as follows:

[0132] Step 1: Identification and positioning

[0133] First, analyze the C2f module of the YOLOv8 model in depth, accurately locate the existing bottleneck layer, and prepare for the subsequent replacement process;

[0134] Step 2: Replacement and integration

[0135] (1) Removal operation, gradually remove the existing bottleneck layer in the C2f module to release space for the integration of new FasterBlock;

[0136] (2) Insertion operation, insert the FasterBlock structure into the C2f module according to the original position order of the bottleneck layer to ensure the coherence of the network structure;

[0137] (3) Dimension matching, to ensure the compatibility of FasterBlock with the original network architecture, adjust the input and output dimensions of FasterBlock, which includes adding or removing a 1x1 convolutional layer (linear layer) when necessary to achieve dimensional consistency;

[0138] S3: Model reconstruction

[0139] After the integration of the FasterBlock structure, the entire model is recompiled to ensure that all network layers are connected correctly and that information flows smoothly within the model.

[0140] Step 4: Training and Optimization

[0141] Using the updated model structure for training, the hyperparameters are carefully adjusted to optimize the model's performance indicators. Through this series of training and optimization steps, the improved YOLOv8 model provided by the invention demonstrates excellent performance in sea cucumber identification tasks.

[0142] When identifying sea cucumbers, some sea cucumbers have a small proportion, leading to a high proportion of image background. In order to more accurately and quickly identify small and medium-sized sea cucumbers, such as Figure 6 , the C2f-F structure adds a SA attention mechanism (ShuffleAttention). When identifying individual sea cucumbers, the SA attention mechanism can effectively extract key features of sea cucumbers in images, enhancing the model's ability to perceive spatial features of input sea cucumbers while reducing computational complexity and improving detection speed, thereby improving the overall performance of the model in sea cucumber image feature extraction tasks.

[0143] As shown in Figure 7 , the SA attention mechanism includes an input end (Input), a split layer (Split), an upper branch, a lower branch, a connection layer (Concat), and an output end (Output).

[0144] The input end inputs the sea cucumber features Figure 8 ∈R (C，H，W) , the sea cucumber feature map is divided into G groups along the channel dimension, denoted as X = [X1,..., X G ], X k ∈R (C / G×H×W) , where C is the number of channels of the sea cucumber feature map, H is the spatial height of the sea cucumber feature map, W is the spatial width of the sea cucumber feature map, R represents the set of all images in the dataset, and X k is the input feature.

[0145] The split layer divides the input feature X k into an upper branch X K1 and a lower branch X K2 along the channel dimension, X K1 , X K2 ∈R (C / 2G×H×W) The upper branch uses channel attention, generating channel information S through transformation parameters w1, b1 scaling and displacement. First, global average pooling (GAP) is used to embed global information, generating channel information S ∈ R (C / 2G×1×1) . This channel information is used to adjust the channel dimension of the input feature Xk1 The contraction is performed on the spatial dimension HxW to calculate

[0146]

[0147] The channel information amount S is transformed by using the full connection layer and a sigma activation function (sigmoid) to generate a weight matrix and multiply X k1 to obtain the weighted features Figure X k1 :

[0148] X′ k1 = σ (F C (S)) · X k1 = σ (w1S + b1) · X k1 ,

[0149] where w1∈R (C / 2G×1×1) , b1∈R (C / 2G×1×1) ,

[0150] The mathematical expression of the sigma activation function is:

[0151] where z is the input of the function, which can be a scalar, a vector, a matrix or a higher-dimensional tensor;

[0152] The lower branch uses spatial attention to capture the spatial dependence between features, and the spatial information amount is scaled and shifted by transforming parameters w2, b2, and the group norm (GN) is used to generate the statistics of spatial attention:

[0153] X′ k2 = σ (w2·GN(X k2 ) + b2) · X k2 ,

[0154] where w2∈R (C / 2G×1×1) , b2∈R (C / 2G×1×1) .

[0155] The connection layer combines X′ K1 and X′ K2 processed by the upper and lower branches, so that the output feature map and the input size are the same,

[0156] X′ k = [X′ k1 , X′ k2 ]R (C / G×H×W) .

[0157] ​To address the challenges faced by backbone networks in performing sea cucumber feature extraction tasks, an innovative ADown module is used to optimize the downsampling process. Traditional standard convolution downsampling reduces the spatial size of feature maps by increasing the stride, however, this method often leads to a significant reduction in feature map size, causing the sea cucumber region, which occupies a small proportion in the original image, to be incorrectly merged into the larger background region, thereby causing the loss of key information. This information loss phenomenon makes the features of small sea cucumbers become blurred or diluted during the downsampling process, severely affecting the model's ability to recognize these targets in subsequent layers.

[0158] As shown in Figure X , the ADown downsampling method combines the advantages of average pooling (AvgPool2d) and maximum pooling (MaxPool2d), where average pooling is responsible for retaining global information, while maximum pooling emphasizes significant features. By performing convolution operations on the outputs of these two pooling methods respectively, the ADown module can effectively retain multi-level information, thereby preventing potential information loss during downsampling and improving the model's sensitivity to detailed features. Meanwhile, a weight factor learning mechanism is introduced to achieve more refined downsampling result fusion.

[0159] By replacing the traditional ordinary downsampling convolution layer with the downsampling Adown module, the model's perception of the target is significantly improved, thereby improving the accuracy and reliability of sea cucumber detection. This innovative design provides a new solution for sea cucumber feature extraction and detection, with wide application prospects and practical value.

[0160] The ADown module includes an input end (Input), an average pooling layer (AvgPool2d), a split layer (Split), a first path (cv1), a second path (cv2), a connection layer (Concat), and an output end.

[0161] The input end inputs features Figure 9 ∈R (C，H，W) The average pooling layer performs an average pooling operation on the input features Figure X by an average pooling layer with a pooling kernel k size of 2, a stride s of 1, and no padding (p = 0), to obtain average pooling features Figure X avg ,

[0162] X avg = AvgpoolX, k = 2, s = 1, p = 0).

[0163] The split layer splits the average pooling features Figure X avg into two sub-features Figure X 1 and X2 along the channel dimension. Figure X1 and X2 are down-sampled through the first path and the second path respectively, to obtain feature maps Y1 and Y2.

[0164] The first path is: sub-feature Figure X 1 is directly down-sampled through a traditional convolutional layer to obtain a feature map Y1,

[0165] Y1 = Conv1 (X1, k = 3, s = 2, p = 1);

[0166] The second path is: sub-feature Figure X Figure X 2 is first extracted through a max-pooling layer with a pooling kernel k of 3, a step length s of 2, and a padding p of 1 to highlight the significant features, and then the feature map is further extracted and refined through a convolutional layer with a convolution kernel k of 1, a step length s of 1, and a padding p of 0 to obtain a feature map Y2,

[0167] X 2，max = MaxPool (X2, k = 3, s = 2, p = 1),

[0168] Y2 = Conv2 (X 2，max , k = 1, s = 1, p = 0),

[0169] Two learnable weight factors α and β are introduced in the connection layer, which are optimized through the training process to weight the down-sampling results. The weight factors α and β control the contribution of Y1 and Y2 in the fusion process respectively. The dimensions of the weight factors α and β are both 1x1x1. The feature map Y fusion after weighted fusion is:

[0170] Y fusion = α·Y1+β·Y2,

[0171] In the formula, · represents element-wise multiplication.

[0172] The weight factors α and β are normalized to ensure that their sum is 1:

[0173]

[0174] The feature map Y fusion after weighted fusion is recalculated using the normalized weight factors as: norm norm

[0175] Y fusion = α fusion ·Y1+β norm ·Y2.

[0176] The feature map Y fusion after weighted fusion is output from the output end, and its dimension is In the formula, C' is the number of output channels.

[0177] The method for constructing the visual recognition model is: grouping sea cucumbers according to the regions where the sea cucumbers are located and the varieties of the sea cucumbers, and making the sizes of the sea cucumbers in each group in a similar section, and the number of the sea cucumbers in each group is not less than 100; obtaining photos of each sea cucumber under different shooting angles, different illuminations and different backgrounds in a natural condition, the number of the photos is not less than 1000, obtaining growth trait data of each sea cucumber, the growth trait data includes body length, body width, wet weight and dry weight, matching the sea cucumber images with the growth trait data, and constructing a data set; dividing the shooting pictures into a training set and a test set according to a proportion, manually labeling the sea cucumbers in the training set using a bounding box, training a model for all labeled sea cucumber image information through a YOLOv8-seacucumber visual recognition model, testing the model using the test set, until the detection accuracy and speed of the model meet the requirements, and manually reviewing the training results; randomly testing the trained model, if the average precision (AP) of the test result is greater than a threshold value, the visual recognition model is successfully trained, and the average precision AP (IoU = 0.75) threshold value of the confidence is preferably 0.8. The visual recognition model can be used to identify sea cucumbers of the same variety in the region, if the region where the sea cucumbers are located or the variety of the sea cucumbers is changed, the model needs to be re-established by repeating the above steps. The YOLOv8-seacucumber visual recognition model identifies possible sea cucumber targets in the image and calculates the confidence thereof, identifies the targets with a confidence greater than a threshold value as sea cucumber targets, and marks them in the image, the threshold value of the confidence is preferably 0.9, draws a recognition box taking the longest and widest parts of the edge of the sea cucumber target as critical points, and records the size of the recognition box, the size includes the number of pixels of the length and width of the recognition box, and each recognition box is sequentially sorted, and the final sorting value of the recognition box is recorded as the total number of sea cucumber target individuals identified in this round.

[0178] Further, the image recognition module is further provided with a sea cucumber disease recognition model, the sea cucumber disease recognition model monitors the health status of the sea cucumbers according to the sea cucumber images, and can timely find the disease risk of the sea cucumbers. The sea cucumber disease library model is trained using photos of diseased sea cucumbers, the pictures are divided into a training set and a test set according to a proportion, the diseased conditions of the sea cucumbers in the training set are labeled manually using a bounding box, a model is trained for all labeled sea cucumber image information through a YOLOv8-seacucumber visual recognition model, and the model is tested using the test set, until the detection accuracy and speed of the model meet the requirements, and the training results are manually reviewed. The health status of the sea cucumbers is identified using the sea cucumber disease recognition model at the same time of the sea cucumber recognition, the sea cucumber individuals with diseases or abnormal states are additionally marked and labeled with a warning label, so as to be distinguished separately. Workers can take corresponding processing measures according to the identification information of the health status of the sea cucumbers, so as to ensure the healthy growth of the sea cucumbers.

[0179] The data processing module obtains the distance between the optical center of the camera and the shooting target measured by the laser range finder, calculates the actual size of the recognition frame based on the calibrated camera parameters and the size of the recognition frame, wherein the camera parameters include the camera sensor length, the camera sensor width, the image length pixel number, the image width pixel number, the focal length, the horizontal view angle and the vertical view angle. The actual size of the recognition frame includes converting the pixel size of the recognition frame into the first actual size of the recognition frame, and calculating the second actual size of the recognition frame using the plane view angle. The calculation results of the first actual size and the second actual size are verified with each other, the error between the first actual size and the second actual size is verified, and it is calculated whether the first actual size is within the error threshold of the second actual size. The recognition frame with the error value within the error threshold is obtained, and the average of the first actual size and the second actual size of the recognition frame is obtained to obtain the actual size of the recognition frame, and the error threshold is preferably 5%. The calculation formula of the first actual size and the second actual size is:

[0180] The first actual size includes the first actual length and the first actual width, and the calculation formula of the first actual length is:

[0181]

[0182] In the formula, l a is the first actual length, l is the camera sensor length, H is the shooting height, P la is the length pixel number of the recognition frame, P l is the image length pixel number, f x is the calibrated x-axis focal length, μ1 is the first correction coefficient, and the error between the average calculation value and the true measurement value of the first actual size is corrected,

[0183]

[0184] The calculation formula of the first actual width is:

[0185]

[0186] In the formula, w a is the first actual width, w is the camera sensor width, P wa is the width pixel number of the recognition frame, P w is the image width pixel number, f y is the calibrated y-axis focal length;

[0187] The second actual size includes the second actual length and the second actual width, and the calculation formula of the second actual length is:

[0188]

[0189] In the formula, lb is the second actual length, θ l is the calibrated horizontal viewing angle, μ2 is the second correction coefficient, correcting the error between the average calculated value and the true measured value of the second actual size,

[0190]

[0191] The calculation formula of the second actual width is,

[0192]

[0193] wherein w b is the second actual width, θ w is the calibrated vertical viewing angle.

[0194] If the first actual size and the second actual size are greater than the error threshold, the data information of the abnormal sea cucumber is recorded and a prompt is issued, and the abnormal sea cucumber data information is temporarily not recorded in the statistical result. The abnormal data is reviewed by manual, and the data is loaded into the statistical result after manual review. If the first actual size and the second actual size generally have large errors, usually because the camera shooting accuracy does not meet the use requirement, the camera needs to be recalibrated.

[0195] The prediction model is used to measure and predict the growth state of the sea cucumber target based on the actual size of the recognition box. The prediction model is based on the labeled sea cucumber image and the sea cucumber growth trait data, and the actual size of the recognition box of the same species of sea cucumber in the same region is fitted with its growth trait data, and multiple regression models and machine learning are used for model training to obtain the prediction model. The construction method of the prediction model is to use the image of the sea cucumber and its growth trait data as a data set, and use multiple regression models to train the data set with the true size of the recognition box in the sea cucumber image as a feature. The training model includes multiple linear regression (Multiple Linear Regression), polynomial regression (Polynomial Regression), generalized linear model (Generalized Linear Models, GLM), decision tree regression (Decision Tree Regression), ridge regression (Ridge Regression), random forest (Random Forests), and support vector regression (Support Vector Regression, SVR) and the like. The trained model is tested, and the coefficient of determination R 2 The model performance is evaluated, and the coefficient of determination R 2 represents the degree of explanation of the variables explained by the regression model to the target variable, the larger the coefficient of determination, the smaller the error, which means that the corresponding model is better, and if the evaluation result R 2greater than 0.9, the model is considered qualified.

[0196] The actual size of each sea cucumber target recognition box is loaded into the prediction model as a prediction parameter, the growth state of the sea cucumber target is fitted and predicted, a plurality of fitted models with the highest confidence and a confidence greater than a threshold value are selected as the prediction model, the growth trait data of the plurality of fitted models meeting the requirements are averaged to obtain the measurement data of the length and width of a single sea cucumber target and the prediction data of the current wet weight and dry weight, and the prediction value is stored in the database to record the growth trait data of each sea cucumber. Specifically, the actual size of a target sea cucumber recognition box is obtained and loaded into the prediction model, the prediction model selects three fitted models with the highest confidence and a confidence greater than 0.9 according to the actual size of the sea cucumber target recognition box, that is, the fitted models closest to the sea cucumber target, and the average of the three fitted models is calculated, that is, the growth trait data of the target sea cucumber is obtained.

[0197] The sea cucumber often appears twisted in the real growth state, it is difficult to measure the real length, width and other information of the sea cucumber in the live state of the sea cucumber without affecting the normal growth of the sea cucumber, and it is also difficult to measure the dry weight of the sea cucumber, and using the calculation of the length and width of the sea cucumber based on the twisted radian of the sea cucumber is also prone to errors in the calculation results and the real length of the sea cucumber due to the shrinkage of the sea cucumber itself. Therefore, the present application considers the twisted condition of the sea cucumber, that is, even if the sea cucumber is in a twisted state, the length-width ratio of the recognition box changes, a high-confidence regression model established according to the fitting of the growth trait data of the sea cucumber according to the actual size of the recognition box can be used to measure and predict the growth trait data of the sea cucumber, and the measurement data of the length and width of the sea cucumber and the prediction data of the wet weight and dry weight are obtained. In addition, compared with the calculation of the real length of the sea cucumber, the use of the actual size of the recognition box to measure and predict the growth trait data of the sea cucumber can significantly reduce the calculation intensity and improve the detection efficiency.

[0198] In this paper, specific examples are used to illustrate the principles and embodiments of the present application, and the above examples are only used to help understand the method and core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific embodiments and application scope will be changed. In view of the above, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A fully automated method for high-throughput measurement and prediction of growth traits in sea cucumbers, characterized in that, include: The calibrated image acquisition module captures images containing sea cucumber targets; A pre-built visual recognition model identifies sea cucumber target images and draws a recognition box using the longest and widest points of the sea cucumber target's edge as critical points, and records the size of the recognition box; The distance from the image acquisition module to the plane where the sea cucumber target is located is obtained, and the actual size of the recognition box is calculated based on the parameters of the image acquisition module. Each recognition box is sorted in turn, and the total number of individual sea cucumber targets is counted. The actual size of the identification box is used as a prediction parameter and loaded into a pre-built prediction model. The prediction model measures and predicts the growth trait data of the sea cucumber target based on the relationship between the actual size of the identification box and the growth traits of the sea cucumber. Multiple fitting models with the highest confidence and confidence greater than the threshold are selected, and the average value of the fitting models is calculated to obtain the measurement data of the length and width of the sea cucumber target and the prediction data of wet weight and dry weight. The image recognition module uses an improved YOLOv8-seacucumber visual recognition model to identify sea cucumbers. Targets with a confidence level greater than a threshold are identified as sea cucumber targets and bounding boxes are drawn. The YOLOv8-seacucumber visual recognition model includes an input terminal, a backbone network, and a detection head. Based on the original YOLOv8 visual recognition model, the C2f module in the backbone network and detection head is replaced with the C2f-F-SA module. The C2f-F-SA module includes a convolutional layer, a segmentation layer, multiple FasterBlock structures, a connection layer, and another convolutional layer arranged sequentially. The bottleneck layer in the original C2f module is replaced with a FasterBlock structure, and a Shuffle Attention mechanism is added to the FasterNet Block module. The Conv module in the detection head is replaced with an Adown module. The Shuffle Attention mechanism includes an input layer, a segmentation layer, an upper branch, a lower branch, a connection layer, and an output layer. Input sea cucumber feature image The sea cucumber feature maps are divided into G groups according to the channel dimension, as follows: ， Where C is the number of channels in the sea cucumber feature map, H is the spatial height of the sea cucumber feature map, W is the spatial width of the sea cucumber feature map, R represents the set of all images in the dataset, and X k Input features; The segmentation layer takes the input feature X k Divided into upper branch X along the channel dimension K1 With the lower branch X K2 , The upper branch employs channel attention, generating channel information S by scaling and shifting parameters w1 and b1. Global average pooling is then used to embed global information, generating channel information S. The information content of this channel is obtained by X k1 The calculation is performed by shrinking the spatial dimension H×W. ， The channel information S is transformed using a fully connected layer, and a weight matrix is ​​generated by applying the σ activation function and multiplying by X. k1 The weighted feature map X' is obtained. k1 : ， In the formula, , ; The lower branch employs spatial attention to capture the spatial dependencies between features. By scaling and shifting the spatial information content through transformation parameters w2 and b2, the statistics of spatial attention are generated using the group norm. ， In the formula, , ; The connection layer processes the upper and lower branches to obtain X' K1 and X' K2 This combination ensures that the output feature map and the input feature map have the same size. 。 2. The fully automated high-throughput growth trait measurement and prediction method for sea cucumbers according to claim 1, characterized in that, The method for calculating the actual size is as follows: based on the calibrated image acquisition module parameters, the recognition box size, and the distance from the image acquisition module to the plane where the sea cucumber target is located, the recognition box size is converted into the first actual size of the recognition box, and the second actual size of the recognition box is calculated using a planar perspective. The error between the first actual size and the second actual size is verified, the recognition box with the error value within the error threshold is obtained, and the average value of the first actual size and the second actual size is taken to obtain the actual size of the recognition box.

3. The fully automated high-throughput growth trait measurement and prediction method for sea cucumbers according to claim 2, characterized in that, The formulas for calculating the first actual dimension and the second actual dimension are as follows: The first actual dimension includes a first actual length and a first actual width, and the formula for calculating the first actual length is as follows: ， In the formula, l a Let l be the first actual length, l be the camera sensor length, H be the shooting height, and P be the actual length. la P is the number of pixels for the recognition bounding box length. l f is the number of pixels in the image length. x The x-axis focal length is calibrated, and μ1 is the first correction factor; The formula for calculating the first actual width is as follows: ， In the formula, w a The first actual width is given by w, where w is the width of the camera sensor, H is the shooting height, and P is the actual width. wa To identify the number of pixels in the width of the bounding box, P w f is the number of pixels in the image width. y The calibrated focal length along the y-axis; The second actual dimension includes a second actual length and a second actual width. The formula for calculating the second actual length is as follows: ， In the formula, l b For the second actual length, θ l The horizontal viewing angle is calibrated, and μ2 is the second correction factor. The formula for calculating the second actual width is as follows: ， In the formula, w b For the second actual width, θ w The vertical viewing angle is calibrated.

4. The fully automated high-throughput growth trait measurement and prediction method for sea cucumbers according to claim 1, characterized in that: The ADown module includes an input layer, an average pooling layer, a segmentation layer, a first path layer, a second path layer, a connection layer, and an output layer. Input feature map The average pooling layer passes the input feature map X through an average pooling layer with a kernel size k of 2, a stride s of 1, and no padding, to obtain the average pooled feature map X. avg , ， The segmentation layer will average pool the feature map X avg The feature map is divided into two sub-feature maps X1 and X2 along the channel dimension. Sub-feature maps X1 and X2 are then downsampled simultaneously through the first path and the second path, respectively, to obtain feature maps Y1 and Y2. The first approach involves downsampling the sub-feature map X1 directly through a traditional convolutional layer to obtain the feature map Y1. ； The second path involves first passing the sub-feature map X2 through a max-pooling layer with a kernel size k of 3, a stride s of 2, and p = 1 to extract the maximum value X2 within the local window, highlighting salient features. Subsequently, this feature map is passed through a convolutional layer with a kernel size k of 1, a stride s of 1, and p = 0 to further extract and refine features, resulting in the feature map Y2. ， ， Two learnable weight factors, α and β, are introduced into the connection layer. These are optimized during training, and the downsampling results are weighted and fused. Weight factors α and β control the contributions of Y1 and Y2 in the fusion process, respectively. Both weight factors α and β have a dimension of 1×1×1. The weighted fused feature map Y... fusion for: ， In the formula, ∙ denotes element-wise multiplication. The weighting factors α and β are normalized to ensure that their sum is 1. ， ； The weighted fused feature map is recalculated using the normalized weight factors. ， The weighted and fused feature map Y is obtained from the output. fusion As output, its dimension is In the formula, C' represents the number of output channels.

5. The fully automated high-throughput growth trait measurement and prediction method for sea cucumbers according to claim 1, characterized in that: The YOLOv8-seacucumber visual recognition model construction method involves grouping sea cucumbers according to their region and species, and then sampling and training them. Photos of each sea cucumber are obtained under natural conditions, at different shooting angles, under different lighting, and against different backgrounds. The images are divided into training and testing sets according to proportions. The sea cucumbers in the training set are manually labeled using bounding boxes. The model is trained using a convolutional neural network on all labeled sea cucumber image information. The model is tested using the testing set. The training results are manually reviewed to obtain the visual recognition model.

6. The fully automated high-throughput growth trait measurement and prediction method for sea cucumbers according to claim 1, characterized in that, The prediction model construction method is as follows: based on the labeled sea cucumber images and sea cucumber growth trait data, the actual size of the identification box of the same variety of sea cucumber in the same region is fitted with its growth trait data, and multiple regression models and machine learning are used to train the model to obtain the prediction model.

7. A fully automated high-throughput sea cucumber growth trait measurement and prediction device, using the fully automated high-throughput sea cucumber growth trait measurement and prediction method as described in any one of claims 1-6, characterized in that, include: A support is provided, which is set on the area to be measured. A linear guide rail is provided in the support, and a walking unit that can move along the linear guide rail is provided in the linear guide rail. An image acquisition module is mounted on the walking unit, and a polarizing lens is mounted on the lens of the image acquisition module to acquire images containing sea cucumber targets. A ranging module, which is mounted on the walking unit, is used to obtain the distance between the optical center of the image acquisition module and the plane where the sea cucumber target is located; The image recognition module acquires an image containing sea cucumber targets, identifies the sea cucumber targets using a pre-built visual recognition model, draws a recognition box with the longest and widest points of the sea cucumber target's edge as the critical points, records the size of the recognition box, sorts each recognition box in the recognition area in turn, and counts the total number of individual sea cucumber targets. The data processing module obtains the size of the recognition frame and the distance from the image acquisition module to the plane where the sea cucumber target is located, and calculates the actual size of the recognition frame based on the parameters of the image acquisition module; The prediction module obtains the actual size of the recognition box, uses a pre-built prediction model with the actual size of the sea cucumber target recognition box as the prediction parameter, selects multiple fitting models with the highest confidence and a confidence greater than a threshold, and calculates the average value of the fitting models to obtain the measurement data of the length and width of the sea cucumber target and the prediction data of wet weight and dry weight.

8. The fully automated high-throughput sea cucumber growth trait measurement and prediction device according to claim 7, characterized in that: A first motor is installed on the linear guide rail, which drives the walking unit to move. A steering device is also installed in the walking unit, and a second motor is installed in the steering device to drive the steering device to rotate. The image acquisition module and the ranging module are installed in the steering device.

Images