A computer vision-based non-destructive shrimp body length measuring device and method

Abstract

This invention discloses a non-destructive measurement device and method for shrimp body length based on computer vision. The method acquires images of shrimp sliding on a track, preprocesses them, and then uses the YOLOv11 instance segmentation model to segment the shrimp bodies in the preprocessed images. After smoothing, the first and last points are initially determined by combining the curvature peak points. It is then determined whether the candidate peak points meet predetermined conditions; if so, the final tail point is determined, obtaining the key points for the first and last ends. A midline fitting method is used to obtain the growth curve, thereby determining the true body length of the shrimp. This invention combines instance segmentation model and geometric feature analysis technology to achieve fully automated, high-precision, non-destructive measurement of shrimp body length, providing reliable data support for monitoring and evaluating growth dynamics during the aquaculture process, and contributing to the digitalization and precision development of aquaculture.

Description

Technical Field

[0001] This invention belongs to the fields of aquaculture and computer vision technology, specifically relating to a non-destructive measurement device and method for shrimp body length based on computer vision. This device can measure the body length of Litopenaeus vannamei in real time and without damage. Background Technology

[0002] As one of the world's largest-scale and most economically beneficial aquatic species, the body length of the Pacific white shrimp is a key biometric indicator for assessing its growth status, health level, and aquaculture efficiency. Body length data not only reflects shrimp growth and development but also serves as an important basis for regulating feed input, rationally allocating stocking densities, and implementing precision management. The Food and Agriculture Organization of the United Nations (FAO) recommends conducting multiple systematic body length measurements of Pacific white shrimp within a standard culture cycle to ensure scientific decision-making and maximize profits. However, current traditional shrimp body length measurements mainly rely on manual harvesting and contact measurements using measuring tapes or calipers. This method is not only labor-intensive but also prone to human error and shrimp stress, leading to shrimp injury, disease transmission risks, and aquatic pollution. Especially in high-density aquaculture environments, manual measurement is inefficient and does not meet the demands of modern intensive and intelligent aquaculture. Therefore, developing a real-time, non-destructive, automated, and high-precision Pacific white shrimp body length measurement technology has become an urgent technological innovation direction for the aquaculture industry.

[0003] Automated body length measurement technology based on computer vision, due to its non-contact, efficient, and real-time characteristics, can conveniently achieve rapid measurement and dynamic monitoring of shrimp body length. However, as a non-rigid aquatic organism, the Pacific white shrimp naturally exhibits a highly flexible bending posture. Individuals often present complex morphologies with multiple angles and curvatures during imaging, especially when swimming naturally or affected by water flow. This non-linear three-dimensional bending projected onto a two-dimensional image poses a significant challenge to body length estimation using traditional linear or keypoint-based methods. Simple straight-line measurements or keypoint connections often fail to accurately reflect the true length of the shrimp, resulting in measurement bias. Furthermore, the bifurcated structure of the Pacific white shrimp's tail creates complex geometric features. The tail fan is composed of multiple flexible appendages, and the outline region exhibits obvious sharp bifurcations and overlaps, making it difficult for existing algorithms to maintain continuity and stability. These two major characteristics present serious recognition and fitting challenges for Pacific white shrimp body length measurement, limiting the widespread application of visual measurement technology in precision aquaculture. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a non-destructive measurement device and method for shrimp body length based on computer vision. This invention combines instance segmentation models with geometric feature analysis technology to achieve fully automated, high-precision, and non-destructive measurement of shrimp body length, providing reliable data support for monitoring and evaluating growth dynamics during the aquaculture process, and contributing to the digitalization and precision development of aquaculture.

[0005] The technical solution adopted in this invention is as follows:

[0006] A computer vision-based non-destructive measurement device for shrimp body length includes a slide, a camera, a variable frequency water pump and flow rate adjustment module, a data processing unit, and an auxiliary lighting device. The slide is inclined, with shrimp entering from the top and flowing continuously with the water flow in a stress-free state, exiting from the bottom. The camera faces the slide to capture clear images of the flowing shrimp. The auxiliary lighting device evenly illuminates the measurement area to prevent ambient light changes from affecting image quality. The variable frequency water pump and flow rate adjustment module are installed at the upper entrance of the slide to regulate the water flow speed, ensuring that the shrimp pass through the measurement area in a single, unobstructed path, thus ensuring the stability and continuity of image acquisition. The data processing unit is connected to the camera for real-time processing of received image data and body length measurement. Simultaneously, the data processing unit is connected to the variable frequency water pump and flow rate adjustment module for coordinated control of the water pump and system alarms, achieving automated and efficient operation of the measurement process.

[0007] Based on the above-mentioned measuring device, the present invention provides a non-destructive measurement method for shrimp body length based on computer vision, the method comprising the following steps:

[0008] Step 1: The data processing unit controls the variable frequency water pump and flow rate adjustment module to open, allowing the shrimp to be put into the shrimp inlet at the top of the slide. The camera captures the video of the whiteleg shrimp flowing through the slide in real time without stress, and transmits the clear high-definition image to the data processing unit through the high-speed data interface.

[0009] Step 2: The data processing unit preprocesses the received raw video, segments the video frame by frame to extract the image information of each frame, and uses Gaussian filtering to reduce environmental noise and enhance the contrast of the shrimp body edges.

[0010] Step 3: Use the YOLOv11 instance segmentation model to perform instance segmentation on the shrimp bodies in the preprocessed image and output a binary shrimp body mask.

[0011] Step 4: Extract the shrimp body contour point set {P0, P1, P2...P} from the mask. n}, preferably using cubic B-spline curves to smooth the contour point set. Based on the recursively defined basis functions, a smoothing curve is constructed:

[0012]

[0013] This eliminates jagged edges and local irregularities in the YOLOv11 instance segmentation results;

[0014] Step 5: Calculate the curvature value on the smooth contour curve, use the positive and negative changes in curvature to identify convex and concave features, thereby classifying the contour points according to their geometric characteristics and finding the peak curvature points on the contour.

[0015] Step 6: Select the maximum value point from the peak curvature as one of the key points (first feature point) of the beginning and end, and then find the corresponding key point at the other end according to the contour arc length (that is, select the point that is farthest from the first feature point on the contour among the remaining positive curvature peak points as the second feature point) to complete the preliminary positioning of the shrimp's beginning and end.

[0016] Step 7: For the complex bifurcation structure at the tail, a neighboring peak merging strategy is adopted: for feature point P i and its neighboring peak point P j First, calculate the arc distance between the two points. If the distance between the two peak curvature points is less than the preset threshold δ·D arc Then, it checks whether there is a negative curvature representing a concavity in the contour region between these two points. If so, it finally determines whether the curvatures of the two points are close. Where δ is the distance threshold factor, k(P) i ) is P i Point curvature, τ is the scaling factor, D arc The arc length of the overall contour. If all three conditions are met, the two points are considered to be a pair of neighboring peak points. Subsequently, the median arc length between the two points is used to replace the original two bifurcation apexes, achieving a unified representation of the shrimp tail bifurcation structure and reducing the impact of the bifurcation structure on measurement accuracy. Once the final tail point is determined, the other point in step 6 that is farthest from the tail point is the final determined first point;

[0017] Step 8: Using the first and last key points as boundaries, a continuous and smooth body length curve is generated using the midline fitting method. Specifically: from the first point to the last point, the complete contour is divided into two sub-contours along clockwise and counterclockwise paths. The arc length of the two contours is calculated and equal arc length sampling is performed. The corresponding sampling point pairs are obtained, and each pair of sampling points is connected and the midpoint of their geometric connection is calculated. Finally, all midpoints are connected in sequence with a smooth curve to obtain a complete and continuous contour midline. The pixel length of the body length curve is calculated and multiplied by a scaling factor to obtain the true body length of the shrimp, where the scaling factor is a built-in parameter of the camera.

[0018] Step 9: Based on the measurement results, the data processing unit adjusts the variable frequency pump flow rate via the control interface to ensure measurement stability. Measurement data is uploaded to the display terminal and the aquaculture management system in real time, supporting historical data storage and analysis.

[0019] The device of the present invention consists of a slide, a camera, a variable frequency water pump and flow rate adjustment module, a data processing unit and an auxiliary lighting device, which constitute a complete automated measurement device that can measure the body length of shrimp of various postures and sizes in real time and without damage.

[0020] The beneficial effects of this invention are:

[0021] This invention relates to a non-destructive body length measurement device for Litopenaeus vannamei (whiteleg shrimp). The device is simple in structure, easy to operate, and suitable for any aquaculture environment. Utilizing advanced visual recognition and curvature analysis technology, it achieves rapid and accurate measurement of shrimp body length. The system adapts to the natural bending posture and forked tail structure of shrimp, significantly improving measurement accuracy and stability. The non-contact measurement method effectively avoids damage and stress to the shrimp caused by traditional manual measurement, ensuring a safe and hygienic aquaculture environment. Simultaneously, the device supports real-time high-throughput detection, greatly improving aquaculture management efficiency and making it suitable for large-scale, intelligent Litopenaeus vannamei aquaculture production. This technology not only enhances the automation and intelligence of measurement but also provides reliable data support for farmers, demonstrating excellent prospects for promotion and application value. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of a non-destructive measurement device for shrimp body length based on computer vision in one embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of shrimp body segmentation in an embodiment of the present invention.

[0024] Figure 3 Yes Figure 2 The result of smoothing the obtained mask is shown in the image.

[0025] Figure 4a This is a graph showing the curvature calculation results in an embodiment of the present invention.

[0026] Figure 4b These are the peak curvature points on the contour obtained by screening in the embodiments of the present invention.

[0027] Figure 5a This refers to the selection of the first feature point and the second feature point in the embodiments of the present invention.

[0028] Figure 5b These are the initial and final points determined in the embodiments of this invention.

[0029] Figure 6aThis is a schematic diagram of the forked structure treatment of the shrimp tail in an embodiment of the present invention.

[0030] Figure 6b In this embodiment of the invention, the median point of the arc length is used as the final determined tail point.

[0031] Figure 7a This is a schematic diagram of two contour lines formed by using the first and last key points as boundaries in an embodiment of the present invention.

[0032] Figure 7b This is a schematic diagram of equal arc length sampling of two contour parts in an embodiment of the present invention.

[0033] Figure 7c This is a schematic diagram of taking the midpoint after connecting corresponding sampling points on two contours in an embodiment of the present invention.

[0034] Figure 7d This is the body length curve obtained in the embodiments of the present invention. Detailed Implementation

[0035] The following description, in conjunction with the accompanying drawings, provides a more detailed account of the computer vision-based non-destructive measurement device and method for the body length of Litopenaeus vannamei.

[0036] According to a specific embodiment of the present invention, a computer vision-based non-destructive measurement device for shrimp body length is as follows: Figure 1 As shown, the overall structure of this automatic shrimp measuring device is a closed trapezoidal box. The top of the box features a large and deep open concave pool (shrimp inlet pool). The bottom of the concave pool transitions downwards to a transparent and smooth slide 1, forming a sloping measurement area that guides the shrimp smoothly to the measurement area for body length measurement. A high-definition camera 2 is fixed above the slope, capable of capturing images of the slope plane. Multiple LED strip lights 3 are evenly distributed along the slope surface, conforming to the lower side of the slope to ensure uniform illumination. A variable frequency water pump and flow rate adjustment module 4 are installed at the connection between the shrimp inlet of the concave pool and the upstream of the slope, allowing adjustment of the shrimp's entry speed and interval via a mechanical gate or vibration device. The end of the slope gradually transitions to a horizontal outlet slide, with the horizontal section extending outside the box to ensure the shrimp smoothly enter the external tank or water body. A data processing and control system 5 is installed directly below the sloping measurement area inside the device, and is sealed to the camera, lighting, and flow rate adjustment module to complete the automatic measurement operation.

[0037] The above-mentioned device was used to perform non-destructive measurement of the body length of Litopenaeus vannamei. The measurement method included the following steps:

[0038] Step 1: The industrial camera captures real-time dynamic images of shrimp flowing through the measuring area of ​​the slide. The image format is RGB color, with a resolution of 1920×1080, and is transmitted to the data processing unit.

[0039] Step 2: The data processing unit preprocesses the received raw video, extracting the image information of each frame and using Gaussian filtering to reduce environmental noise and enhance the contrast of the shrimp's edges. The standard Gaussian kernel function is:

[0040]

[0041] Where σ = 1.0 is the standard deviation, i and j are two input vectors, and the filter window size is 5×5.

[0042] Step 3: The trained YOLOv11 deep learning model is used to segment shrimp instances from the preprocessed image. The model's inference output includes a binary mask for each shrimp. The YOLOv11 instance segmentation model includes a C2PSA component, whose spatial attention mechanism enables accurate identification of shrimp contours in complex environments, performing well even under partial occlusion. Secondly, the model significantly reduces computational overhead through optimized C3k2 block design, achieving a segmentation speed of 82.0 FPS, enabling near real-time processing on medium-configuration hardware and meeting the real-time requirements of high-throughput detection. Finally, YOLOv11 provides models of various sizes with a parameter size of only 22.3M, offering a significant lightweight advantage compared to other models, making it more suitable for deployment in resource-constrained aquaculture monitoring environments. The experimental test results of each model are as follows, with YOLOv11 showing the best performance in the application scenario of this invention.

[0043] Model mAP = 0.5:0.95 Params / M FPS Mask R-CNN 0.751 44.0 17.2 YOLACT 0.649 34.7 24.4 SOLOv2 0.753 46.2 15.4 Mask2Former 0.776 44.0 4.14 QueryInst 0.762 172.45 4.55 CondInst 0.765 33.98 11.8 YOLOv8 0.742 22.7 80.9 YOLOv11 0.741 22.3 82.0

[0044] like Figure 2 As shown, the output mask is in the form of:

[0045]

[0046] Step 4: Extract the shrimp body contour point set {P0, P1, P2...P} from the mask. n The contour is a closed curve. The contour exhibits jagged edges and local noise. A cubic B-spline curve is used to smooth the contour point set. Its inherent local control point characteristics allow for precise fitting of various characteristic curved structures such as the shrimp's head, abdomen, and tail through fine-tuning of the control points, without affecting the overall contour. Secondly, the C-axis of the cubic B-spline curve... 2 By combining continuous data with the morphological characteristics of shrimp, naturally smooth curves can be generated, effectively eliminating jagged edges in the segmentation results while maintaining biologically reasonable curvature features. Third, in real-time processing scenarios involving large batches of images, the recursive calculation method of B-spline curves can significantly reduce computational overhead, meeting the demands of rapid processing—an effect difficult to achieve with other smoothing methods. The parameterized form is as follows:

[0047]

[0048] Where u∈[0,1] are normalization parameters, N i,p(u) Let N be a p-order B-spline basis function. For higher-order basis functions, N... i,p(u) It can be defined recursively:

[0049]

[0050] N i,p (u)=[(ut i ) / (t i+p -t i )]·N i,p-1 (u)+[(t i+p+1 -u) / (t i+p+1 -t i+1 )]

[0051] ·N i+1,p-1 (u)

[0052] Where t i These are the node values ​​in the knot vector. Weights and knot vectors are selected uniformly to ensure the C-axis of the curve. 2 Continuity, eliminating jagged edges effectively preserves the shrimp's morphological characteristics, and the smoothed effect is as follows: Figure 3 As shown.

[0053] Step 5: Calculate the curvature value on the smooth profile curve:

[0054] x′(u)≈[x(u+Δu)-x(u-Δu)] / (2Δu)

[0055] x″(u)≈[x(u+Δu)-2x(u)+x(u-Δu)] / (Δu) 2

[0056]

[0057] Where k is the curvature, x′ and y′ are the first derivatives of x and y, respectively, and x″ and y″ are the second derivatives. The positive and negative changes in curvature are used to identify convex and concave features, finding the peak curvature points on the contour. Positive curvature corresponds to convex features, often the head or tail tip of a shrimp, while negative curvature corresponds to concave features, often the abdomen of a shrimp in a bent posture, facilitating subsequent head and tail point localization. Figure 4a and Figure 4b .

[0058] Step 6: Select the point with the maximum curvature peak as one of the key points of the head and tail, corresponding to the most prominent convex areas such as the head or tail tip. Then, among the remaining positive curvature peak points, select the point that is furthest from the first feature point on the contour as the second feature point. This operation is based on the biological characteristic of the Litopenaeus vannamei that the arc distance between the head and tail points accounts for about half of the total contour length. The two feature points obtained in this way are the head and tail points of the shrimp. Figure 5a , Figure 5b As shown.

[0059] Step 7: For the complex bifurcation structure at the tail, a neighboring peak merging strategy is adopted: for feature point P i and its neighboring peak point P j First, calculate the arc distance between the two points. If the distance between the two peak curvature points is less than the preset threshold δ·D arc Then, it checks whether there is a negative curvature representing a concavity in the contour region between these two points. If so, it finally determines whether the curvatures of the two points are close. Where δ is the distance threshold factor, k(P) i ) is P i Point curvature, τ is the scaling factor, D arc The arc length of the overall profile. If all three conditions are met, the two points are considered to be a pair of neighboring peak points. Subsequently, the median arc length between the two points is used to replace the original two bifurcation apexes, achieving a unified representation of the shrimp tail bifurcation structure and reducing the impact of the bifurcation structure on measurement accuracy. Thus, the median arc length is determined as the final tail point, and the other point farthest from this point in step 6 is the final head point. Figure 6a and Figure 6b As shown.

[0060] Step 8: Using the first and last key points as boundaries, divide the complete contour into two sub-contours along clockwise and counterclockwise paths from the first to the last point, corresponding to the "upper" (or left) and "lower" (or right) curves of the shrimp body, respectively. Calculate the arc length of each contour and perform equal arc length sampling to obtain corresponding sampling point pairs. Since the corresponding sampling points on the left and right (or upper and lower) sides are biologically roughly located in the same segment, when connecting each pair of sampling points and calculating the midpoint of their geometric connection, these midpoints will naturally fall at the center of the shrimp body. Finally, connect all midpoints sequentially with a smooth curve to obtain a complete and continuous contour centerline. Calculate the pixel length of the body length curve and multiply by a scaling factor to obtain the true body length of the shrimp, as shown in the following formula:

[0061]

[0062] L real =L pixel ×S scale

[0063] Where the proportionality coefficient S scale These are the camera's built-in parameters. For example... Figures 7a-7d As shown.

[0064] Step 9: Based on the measurement results, the data processing unit adjusts the variable frequency pump flow rate via the control interface to ensure measurement stability. Measurement data is uploaded to the display terminal and the aquaculture management system in real time, supporting historical data storage and analysis.

[0065] The above-disclosed embodiments are merely specific examples of the present invention, but the present invention is not limited thereto. For those skilled in the art, any modifications made without departing from the present invention should be considered to fall within the scope of protection of the present invention.

Claims

1. A non-destructive method for measuring shrimp body length based on computer vision, characterized in that, This method involves acquiring images of shrimp sliding on a slide, preprocessing them, using the YOLOv11 instance segmentation model to segment the shrimp bodies in the preprocessed images, smoothing the images, and using curvature peak points to initially determine the head and tail points. It then determines whether the candidate peak points meet predetermined conditions; if so, the final tail point is determined, resulting in the final head and tail key points. Finally, a midline fitting method is used to obtain the growth curve, thereby determining the true body length of the shrimp. The specific steps include: Step 1: The data processing unit controls the variable frequency water pump and flow rate adjustment module to open, so that the shrimp are put into the shrimp inlet at the top of the slide. The camera captures the video of the shrimp flowing through the slide in real time without stress and transmits it to the data processing unit. Step 2: The data processing unit preprocesses the received raw video, segments the video frame by frame to extract the image information of each frame, and uses Gaussian filtering to reduce environmental noise and enhance the contrast of the shrimp body edges. Step 3: Use the YOLOv11 instance segmentation model to perform instance segmentation on the shrimp body in the preprocessed image obtained in Step 2, and output a binary shrimp body mask; Step 4: Extract the shrimp body contour point set from the mask The YOLOv11 instance segmentation results are smoothed using b-spline curves to eliminate jagged edges and local irregularities. Step 5: Calculate the curvature value on the smoothed contour curve, use the positive and negative changes in curvature to identify convex and concave features, thereby classifying the contour points according to their geometric characteristics, and then filtering to obtain the positive curvature peak points on the contour. Step 6: Identify the peak point with the largest curvature from the peak points obtained in Step 5 as the first feature point. Then, among the remaining peak points with positive curvature, select the point that is farthest from the first feature point on the contour as the second feature point; thus obtaining the preliminarily determined start and end points. Step 7: For the complex bifurcation structure of the tail, a neighboring peak merging strategy is adopted. Specifically, the complex bifurcation structure of the shrimp tail will cause the contour curvature curve to have a bimodal feature. When two candidate peak points meet the predetermined conditions, they are considered to be a pair of neighboring peak points. Then, the median arc length between the two points is used to replace the two bifurcation tips of the original tail to achieve a unified representation of the shrimp tail bifurcation structure. If this point is determined as the final tail point, then the other point far away from this tail point in step 6 is the final determined first point. Step 8: Based on the key points identified in Steps 6 and 7, a continuous and smooth body length curve is generated using the midline fitting method. The pixel length of the body length curve is calculated and multiplied by a scaling factor to obtain the true body length of the shrimp. The predetermined conditions are as follows: For feature points and its neighboring peak points First, calculate the arc distance between the two points. If the distance between the two peak curvature points is less than a preset threshold... Then, it checks whether there is a negative curvature representing a concavity in the contour region between these two points. If so, it finally determines whether the curvatures of the two points are close. in Distance threshold factor, for Point curvature, As a scaling factor, This is the arc length of the overall outline.

2. The non-destructive measurement method for shrimp body length based on computer vision according to claim 1, characterized in that, In step 4, a cubic B-spline curve is used to smooth the contour point set. Based on the recursively defined basis functions, a smooth curve is constructed: ， Where n is the number of contour points. For the i-th contour point, It is a p-th degree B-spline basis function.

3. The non-destructive measurement method for shrimp body length based on computer vision according to claim 1, characterized in that, Step 5 involves calculating the curvature value on the smoothed contour curve and determining the concave / convex features based on the positive or negative curvature value. Positive curvature corresponds to convex features, while negative curvature corresponds to concave features, which are used for subsequent positioning of the start and end points.

4. The non-destructive measurement method for shrimp body length based on computer vision according to claim 1, characterized in that, The method for locating the head and tail points in step 6 is as follows: Identify the peak point with the largest curvature from the peak points obtained in step 5 as the first feature point, and then select the point that is farthest from the first feature point on the contour from the remaining positive curvature peak points as the second feature point. The two feature points obtained in this way are the preliminarily determined head and tail points of the shrimp.

5. The non-destructive measurement method for shrimp body length based on computer vision according to claim 1, characterized in that, The midline fitting method described in step 8 is as follows: From the first point to the last point, the complete contour is divided into two sub-contours along two paths, clockwise and counterclockwise. The arc length of the two contours is calculated and equal arc length sampling is performed. The corresponding sampling point pairs are obtained, each pair of sampling points is connected, and the midpoint of their geometric connection is calculated. Finally, all midpoints are connected in sequence with a smooth curve to obtain a complete and continuous contour midline. The true body length of the shrimp is obtained by multiplying the pixel length of the body length curve by a scaling factor, where the scaling factor is a built-in parameter of the camera.

6. A non-destructive measurement device for shrimp body length based on computer vision, characterized in that, The system includes a slide, a camera, a variable frequency water pump and flow rate adjustment module, a data processing unit, and an auxiliary lighting device. The slide is inclined, with an inlet at the top and an outlet at the bottom. The shrimp to be tested enters the water from the top and flows continuously with the water flow in a stress-free state, exiting from the bottom. The camera is positioned directly in front of the slide to capture clear images of the flowing shrimp. The auxiliary lighting device is used to uniformly illuminate the area to be measured to prevent changes in ambient light from affecting image quality. The variable frequency water pump and flow rate adjustment module are installed at the inlet at the top of the slide to ensure that the shrimp pass through the measurement area without obstruction by adjusting the water flow speed. The data processing unit is connected to the camera and is used to process the received image data in real time and measure the body length, implementing the method described in any one of claims 1-5. Simultaneously, the data processing unit is connected to the variable frequency water pump and flow rate adjustment module to coordinate the control of the water pump and system alarm.

Images