CN-122023500-A - Method, equipment and storage medium for estimating body size of penaeus vannamei boone

CN122023500ACN 122023500 ACN122023500 ACN 122023500ACN-122023500-A

Abstract

The invention discloses a method, equipment and a storage medium for estimating the body size of penaeus vannamei boone, which relate to the technical field of image processing in the breeding industry and comprise the steps of collecting a penaeus vannamei boone data set image and labeling the penaeus vannamei boone data set image; and (3) constructing a shrimp body detection model, improving the shrimp body detection model, training the improved shrimp body detection model based on the labeled penaeus vannamei boone data set image, measuring a penaeus vannamei boone body ruler based on the trained shrimp body detection model, and estimating the shrimp body length. The method enhances the expression and fusion capability of multi-scale features, improves the positioning precision of prawn feature points, and develops a set of prawn body ruler non-contact estimation system. And accurately acquiring coordinates of characteristic points of the shrimp bodies, calculating body scale indexes, and accurately estimating the shrimps with different integrity. The method realizes rapid and accurate non-contact type prawn body ruler measurement, and provides powerful technical support for intelligent and accurate culture of the penaeus vannamei boone.

Inventors

DING QI
You Guanqi
ZHAI ZHAOYU
ZHANG ZIHAN
CHE JIANHUA
YU HONGFENG
XU HUANLIANG

Assignees

南京农业大学

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. The method for estimating the body size of the penaeus vannamei boone is characterized by comprising the following steps of: Collecting a penaeus vannamei boone data set image, and labeling the penaeus vannamei boone data set image; Constructing a shrimp detection model, improving the shrimp detection model, and training the improved shrimp detection model based on the marked penaeus vannamei data set image; and measuring the body size of the penaeus vannamei boone and estimating the length of the penaeus vannamei boone based on the trained penaeus vannamei boone detection model.
2. The method for estimating a body size of penaeus vannamei boone of claim 1, wherein the step of acquiring the data set image of penaeus vannamei boone comprises the steps of fixing a high-definition camera at a central position right above an observation platform and capturing morphological characteristics of penaeus vannamei boone by adopting a nodding view angle; the image of the data set of the penaeus vannamei comprises feed residue interference, shielding among shrimp bodies and a dense state of shrimp groups.
3. The method for estimating the body size of penaeus vannamei boone as claimed in claim 1 or 2, wherein the step of labeling the images of the penaeus vannamei boone data set comprises the steps of labeling the data set by adopting a Labelme labeling tool, labeling an external rectangular area containing the whole penaeus vannamei boone body by using a rectangular frame tool, positioning characteristic points of each penaeus vannamei boone by means of a point labeling tool, and distributing an ID for each penaeus vannamei boone.
4. The method for estimating a body size of penaeus vannamei boone as claimed in claim 3, wherein the step of improving the shrimp detection model comprises constructing a shrimp detection model based on the YOLO11-Pose model, and optimizing and improving the network structure by introducing the DBB module, the coordinate attention mechanism and the C3k2_ OREPA module.
5. The method for estimating a body size of penaeus vannamei boone as claimed in claim 4, wherein the DBB module comprises the steps of extracting feature information of different scales by adopting four parallel feature extraction paths, and the calculation flow of the DBB module is expressed as follows: , , , , , Wherein, the For the input feature map of the DBB, Is the first The output profile of the individual branches, Is the first The batch normalization layer corresponding to the branch, Is the first A 1x 1 convolution in each branch, Is the first A3 x 3 convolution in each branch, For the output characteristic map of the DBB, 、 Is the first Front and back batch normalization layers when multiple operations are connected in series in a branch, The pooling operation is averaged.
6. The method for estimating a body size of penaeus vannamei as set forth in claim 4 or 5, wherein the coordinate attention mechanism comprises performing direction encoding on the input feature map along a horizontal direction and a vertical direction respectively, preserving position information through one-dimensional global pooling, fusing the direction-encoded features, generating two independent attention weight maps through activation and convolution operations, and adaptively weighting the input features to be expressed as: , , , Wherein, the Is the first The first channel is The horizontal pooling corresponding to the rows outputs the result, Is the first The first channel is The horizontal pooling corresponding to the columns outputs the result, For the width of the feature map, For the height of the feature map, Inputting the first of the feature graphs for horizontal pooling The first channel is Line 1 The pixel values of the columns are used to determine, Inputting the first of the feature graphs for vertical pooling The first channel is Line 1 The pixel values of the columns are used to determine, To output the first of the characteristic diagrams The first channel is Line 1 The pixel values of the columns are used to determine, Is the first to pay attention to the horizontal The first channel is The attention weight corresponding to the row, Is the first of the vertical attention The first channel is The corresponding attention weights are listed.
7. The method for estimating a body size of penaeus vannamei boone as claimed in claim 6, wherein the C3k2_ OREPA module comprises the step of performing multipath supervision on OREPA in the gradient propagation process, wherein the multipath supervision is represented as follows: , , Wherein, the Is the first The output results of the individual branches are presented, Is the first The convolution operation in the individual branches, For the batch normalization operation, Is the first The channel-level linear scaling factors of the individual branches, In order to input the feature map, To output a feature map.
8. The method for estimating a body length of penaeus vannamei as set forth in any one of claims 1,2, 4, 5 and 7, wherein measuring the body length of penaeus vannamei and estimating the body length of penaeus vannamei comprises dividing the body of penaeus vannamei based on coordinates obtained by detecting a model of the body of penaeus vannamei when the body of penaeus vannamei is detected to be complete, estimating the body length of penaeus vannamei by arc length, obtaining a scale by observing a physical length of a diameter of a table and the length of the pixel, and obtaining the physical length of penaeus vannamei by the scale; When the incomplete shrimp body of the penaeus vannamei is detected, determining optimal parameters by adopting a least square method through the proportional relation between the divided part and the full length of the shrimp body, and establishing a regression equation fitting body ruler to estimate the length of the penaeus vannamei.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, performs the steps of the method for estimating the body size of penaeus vannamei according to any one of claims 1-8.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the steps of the method for estimating the body size of penaeus vannamei as set forth in any one of claims 1 to 8.

Description

Method, equipment and storage medium for estimating body size of penaeus vannamei boone Technical Field The invention relates to the technical field of image processing in the breeding industry, in particular to a method, equipment and storage medium for estimating a litopenaeus vannamei body ruler. Background As the aquaculture support industry, the prawn aquaculture industry faces serious challenges such as rising labor cost, frequent diseases (such as acute hepatopancreatic necrosis and white spot syndrome), environmental deterioration and the like, and precise management has become key. In penaeus vannamei boone cultivation, body length is a core index reflecting population health and growth rate, and accurate monitoring of body size parameters can provide important basis for growth assessment, harvest prediction, accurate feeding, density optimization and disease early warning. The traditional manual measurement method has low efficiency and poor consistency, and the frequent capturing and contact can cause stress, damage and even death of the shrimp bodies, so that the data reliability is seriously affected. Therefore, development of a non-contact, rapid and accurate prawn body size measurement technology has become an urgent need for improving industrial intelligence and economic benefits, and a computer vision technology has the advantages of strong self-adaption, low cost and easiness in deployment, can realize nondestructive real-time monitoring of the growth state of prawns, promotes digital upgrading of industrial circulating water and high-level pond culture, and lays a foundation for accurate feeding, health early warning and big data decision-making systems. The computer vision and deep learning method mainly comprises three types of example segmentation, target detection and feature point detection. The example segmentation can provide a pixel-level mask for body ruler estimation, but the shrimp body is semitransparent, the posture is changeable and the background is complex, so that the edge definition is difficult and the error is large, the object detection is used for estimating the body ruler through a boundary box, but the accuracy is insufficient in a high-density shielding scene, the feature point detection is good at locating semantic feature points, the method has been widely applied in the agricultural field, and a large space is still left for special research in the penaeus vannamei body ruler estimation. Under the natural state, the penaeus vannamei often presents obvious bending, stretching, twisting and other complex postures, and the posture changes can directly lead to large deviation of the positioning of head and tail characteristic points. The accuracy and the robustness of the currently mainstream feature point detection model still have obvious defects when facing such highly flexible biological targets. Therefore, development of a special feature point recognition model with strong gesture adaptation capability is needed to significantly improve the detection stability and accuracy under variable gestures. On a culture observation table, the prawns often show a high-density aggregation state, and partial shielding, overlapping or tight adhesion phenomena are very easy to occur among individuals, so that the difficulty of distinguishing examples is increased, and the problems of repeated detection or missed detection and the like are often caused. In order to solve the pain point, accurate distinguishing of overlapping individuals is needed to be realized, so that each shrimp can be independently and accurately identified and tracked, complex interference factors such as feed residues, impurity particles and the like often exist in an actual culture observation environment, and the background noise is easily and mistakenly identified as a shrimp target, so that the false detection rate is increased. In order to improve the practicability and reliability of the system, a highly-targeted anti-interference strategy, such as multi-feature fusion and other methods, must be introduced, so that the adverse effect of environmental factors on the detection performance is effectively reduced. Disclosure of Invention The present invention has been made in view of the above-described problems. The method solves the technical problems that the existing method for measuring the body ruler of the penaeus vannamei has the defects of insufficient positioning precision of characteristic points when the postures of the penaeus vannamei are various in a complex cultivation environment, large estimation error of the body ruler caused by individual adhesion and shielding under a group high-density aggregation condition, and easy misrecognition caused by environmental interference such as feed residues, and the like, and is difficult to realize stable and accurate estimation of the body ruler under the condition of incomplete prawn bodies or lack of the characteristic points. The method comp