Factory farming biomass real-time monitoring device fusing machine vision

Abstract

The present application relates to the technical field of breeding monitoring, and discloses a factory breeding biomass real-time monitoring device fusing machine vision, which is used to solve the problem that non-living organisms such as bubbles are mistaken for living organisms when monitoring breeding organisms, and comprises the following steps: underwater image acquisition to obtain an underwater image; obtaining a fusion sensor; preliminary living organism identification on the underwater image to obtain a preliminary living organism image; obtaining living organism information of the preliminary living organism image through the fusion sensor; obtaining a living organism determination index according to the living organism information evaluation; living organism determination according to the living organism determination index to obtain a determined living organism; obtaining the number, size and density of the determined living organism; analyzing the biomass according to the number, size and density of the determined living organism; visualizing the biomass; and updating the real-time data, which effectively improves the accuracy of biomass data acquisition and the efficiency of subsequent breeding management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cultivation monitoring, more particularly to a factory farming biomass real-time monitoring device fusing machine vision. BACKGROUND

[0002] The rapid development of Internet of Things technology has promoted the deep integration of machine vision and intelligent sensing technology in the field of factory farming. By deploying distributed visual monitoring nodes, combining edge computing and cloud analysis, real-time monitoring and precise control of key indicators such as the growth state of the cultivated organisms and environmental parameters can be achieved. Efficient and accurate biomass monitoring not only helps to optimize feeding strategies and reduce farming costs, but also provides data support for disease early warning and yield prediction.

[0003] In the prior art, machine vision-based cultivation biomass monitoring usually adopts static feature recognition and fixed threshold determination methods. For example, target detection algorithms are used to identify individual organisms, and then morphological analysis is used to estimate biomass. Some optimization schemes introduce multi-sensor fusion to improve detection accuracy in complex environments.

[0004] However, in the process of implementing the technical scheme of the present application, the present application has found that the above-mentioned technology at least has the following technical problems:

[0005] In the factory farming scenario, a large number of bubbles will be generated in the water due to the movement of organisms, etc. The device may mistakenly identify bubbles as organisms, resulting in an overestimation of the number of organisms detected, which affects subsequent farming management and strategy selection. SUMMARY

[0006] In order to overcome the above-mentioned defects of the prior art, the present application provides a factory farming biomass real-time monitoring device fusing machine vision to solve the problems existing in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0008] The factory farming biomass real-time monitoring device fusing machine vision comprises:

[0009] An image acquisition module for underwater image acquisition by a high-resolution camera device, obtaining underwater images, and transmitting the underwater images to a primary organism recognition module;

[0010] A sensor fusion module for fusing environmental sensors, depth sensors, and ultrasonic sensors to obtain a fusion sensor;

[0011] The preliminary biometrics module is used to perform preliminary biometrics based on underwater images using an edge detection algorithm, obtain a preliminary organism image, and transmit the preliminary organism image to the organism determination module;

[0012] The organism identification module is used to acquire organism information from preliminary organism images through fusion sensors. The organism information includes movement trajectory, morphological information, optical characteristics, and size changes. An organism identification index is obtained based on the organism information. The organism is then identified based on the organism identification index to obtain the identified organism. The identified organism is then transmitted to the biomass analysis module.

[0013] The biomass analysis module is used to obtain the number, size, and density of organisms, analyze the biomass based on the determined number, size, and density of organisms, and then transmit the biomass to the visualization module.

[0014] The visualization module is used to visualize biomass and update data in real time.

[0015] Preferably, the step of performing preliminary biometric identification based on the underwater image using an edge detection algorithm to obtain a preliminary biological image is as follows:

[0016] The underwater images are preprocessed, including noise reduction and illumination correction.

[0017] The Sobel operator is used to calculate the gradient magnitude in the horizontal and vertical directions. Pixels with gradient magnitudes higher than the threshold are set to 255, and the rest are set to 0.

[0018] Connectivity analysis was used to extract all closed edges, resulting in a preliminary image of the organism.

[0019] Preferably, the step of obtaining the organism determination index is as follows:

[0020] Using the initial biological image as the target image, the position coordinate sequence of the target image in consecutive frames within the detection time period is obtained, and the motion purposefulness coefficient is evaluated based on the position coordinate sequence.

[0021] Obtain the contour data of the target image, and evaluate the irregularity coefficient based on the contour data of the target image.

[0022] A gray-level co-occurrence matrix is ​​constructed from the target image, and texture coefficients are calculated based on the gray-level co-occurrence matrix.

[0023] Obtain the size data of the target image within the detection time period, and calculate the size stability coefficient based on the size data within the detection time period;

[0024] The motion purpose coefficient, the irregular coefficient, the texture coefficient and the size stability coefficient are normalized, and the organism determination index is calculated according to the normalized motion purpose coefficient, the normalized irregular coefficient, the normalized texture coefficient and the normalized size stability coefficient, and the specific acquisition steps are as follows:

[0025] LO=a1×PE+a2×IL+a3×TU+a4×DS;

[0026] In the formula, LO represents the organism determination index, PE represents the normalized motion purpose coefficient, IL represents the normalized irregular coefficient, TU represents the normalized texture coefficient, DS represents the normalized size stability coefficient, a1, a2, a3 and a4 represent the weight coefficients of the motion purpose coefficient, the irregular coefficient, the texture coefficient and the size stability coefficient.

[0027] Preferably, the motion purpose coefficient acquisition step is as follows:

[0028] The position coordinate sequence of the target image in the continuous frames in the detection time period is acquired through a target tracking algorithm;

[0029] The horizontal distance and the vertical distance between adjacent frames are calculated, and the displacement vector between adjacent frames is calculated through the Euclidean distance;

[0030] The degree of change of the motion direction is quantified through the included angle of adjacent displacement vectors, and the specific acquisition steps are as follows:

[0031]

[0032] In the formula, θ t represents the motion direction change angle between the tth frame and the (t-1)th frame, Δx t represents the horizontal distance from the tth frame to the (t-1)th frame, Δx t-1 represents the horizontal distance from the (t-1)th frame to the (t-2)th frame, Δy t represents the vertical distance from the tth frame to the (t-1)th frame, Δy t-1 represents the vertical distance from the (t-1)th frame to the (t-2)th frame, d t represents the displacement vector of the tth frame, d t-1 represents the displacement vector of the (t-1)th frame;

[0033] The motion direction change angle standard deviation and the displacement vector variance of each continuous frame in the detection time period are calculated, the maximum displacement distance of the continuous frames in the detection time period is acquired, and the motion purpose coefficient is calculated according to the motion direction change angle standard deviation, the displacement vector variance and the maximum displacement distance, and the specific acquisition steps are as follows:

[0034]

[0035] PE is the motion purpose coefficient, σ θ is the standard deviation of the motion direction change angle, Var(d) is the variance of the displacement vector, and max(d) is the maximum displacement distance.

[0036] Preferably, the irregularity coefficient obtaining step is:

[0037] An edge detection algorithm is used on the target image to extract a set of outer contour pixel points of the target, a minimum circumscribed circle of the contour is fitted, and a center and a radius of the minimum circumscribed circle are obtained;

[0038] The distance of each contour point to the center is calculated, and a mean value is calculated to obtain an average center distance;

[0039] According to the distance of each contour point to the center and the average center distance, a standard deviation of the contour distance is calculated;

[0040] The standard deviation of the contour distance is calculated by ratio with the average center distance to obtain an irregularity coefficient.

[0041] Preferably, the texture coefficient obtaining step is:

[0042] The target image is subjected to grayscale processing to convert it into a single-channel grayscale image, the frequency of occurrence of each pixel pair is counted based on a set direction and distance, and a gray level co-occurrence matrix is constructed according to the frequency of occurrence of each pixel pair;

[0043] The square of the gray level difference of each pair of adjacent pixels is calculated, and is denoted as the difference intensity;

[0044] According to the gray level co-occurrence matrix, the frequency of occurrence of all adjacent pixel pairs is counted, and the frequency of occurrence of all adjacent pixel pairs and the difference intensity are weighted and summed to obtain a texture coefficient.

[0045] Preferably, the size stability coefficient obtaining step is:

[0046] The target image is subjected to binary segmentation to obtain a mask value, and the pixel area of the target image is calculated according to the mask value;

[0047] The area sequence in the continuous frames of the target image in the detection time period is obtained, and the relative area change rate is calculated frame by frame;

[0048] The average relative area change rate in the detection time period is calculated, and the size stability coefficient is obtained according to the average relative area change rate.

[0049] Preferably, the biological body determination according to the biological body determination index is obtained by the following steps:

[0050] The organism determination index is compared with a preset threshold value, if the organism determination index is greater than or equal to the preset threshold value, the organism is determined; if the organism determination index is less than the preset threshold value, the organism is determined.

[0051] Preferably, the biomass acquisition step is:

[0052] According to the area of each determined organism, the average organism area is calculated, and the number of determined organisms is calculated by the ratio of the underwater image area to obtain the organism density.

[0053] According to the number of determined organisms, the average organism area and the organism density, the biomass is calculated, and the specific acquisition steps are:

[0054] BS=NUM S x AVG C x BD.

[0055] In the formula, BS represents the biomass, NUM S represents the number of determined organisms, AVG C represents the average organism area, and BD represents the organism density.

[0056] The technical effects and advantages of the present application are:

[0057] The underwater image is collected to obtain the underwater image, the fusion sensor is acquired, the preliminary organism image is obtained by preliminary organism identification of the underwater image, the organism information of the preliminary organism image is acquired through the fusion sensor, the organism determination index is evaluated according to the organism information, the organism is determined according to the organism determination index, the determined organism is obtained, the number, size and density of the determined organism are acquired, the biomass is analyzed according to the number, size and density of the determined organism, the biomass is visually displayed, and real-time data updating is performed, which effectively improves the accuracy of biomass data acquisition and improves the subsequent breeding management efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 The fusion machine vision factory breeding biomass real-time monitoring device structure diagram provided by the embodiment of the present application. DETAILED DESCRIPTION

[0059] The technical solutions in the present application will be described clearly and completely in combination with the drawings in the present application, and the forms of each structure described in the following embodiments are only examples, and the fusion machine vision factory breeding biomass real-time monitoring device involved in the present application is not limited to each structure described in the following embodiments. All other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0060] The present application provides a factory farming biomass real-time monitoring device fused with machine vision, such as Figure 1 As shown, the device comprises:

[0061] An image acquisition module for underwater image acquisition by a high-resolution camera, obtaining an underwater image, and transmitting the underwater image to a preliminary biological recognition module. The high-resolution camera can be an industrial underwater camera, an underwater monitoring camera, and a multispectral / special camera, etc.

[0062] This module acquires high-quality underwater images in real time through high-resolution underwater cameras, providing a raw data basis for subsequent analysis. Its advantage is that it can clearly capture the details of tiny organisms, adapt to turbid water or low-light environments, ensure the integrity and accuracy of image information, and support flexible equipment replacement through modular design to meet the needs of different monitoring scenarios.

[0063] A sensor fusion module for fusing environmental sensors, depth sensors, and ultrasonic sensors to obtain a fusion sensor.

[0064] The module integrates environmental sensors, depth sensors, and ultrasonic data to generate multi-dimensional, high-precision environmental parameters. Its core advantage is to utilize multi-source data complementarity to overcome the limitations of single sensors (such as the failure of optical cameras in turbid water), improve the environmental adaptability and data reliability of the system, such as verifying the authenticity of visual targets through sonar data to reduce false positives.

[0065] A preliminary biological recognition module for preliminary biological recognition based on the underwater image through an edge detection algorithm to obtain a preliminary biological image, and transmitting the preliminary biological image to a biological body determination module.

[0066] Edge detection algorithm is a technology that extracts the target boundary by identifying the pixel gray value mutation area in the image. Its core is to calculate the image gradient or second derivative to locate the edge. In underwater biological recognition, the Canny algorithm is commonly used because it can effectively suppress false edges caused by turbid water or uneven lighting, output continuous, single-pixel-wide biological contours, and provide a basis for subsequent target segmentation and tracking.

[0067] The module can quickly filter out candidate biological targets. Its advantage is that it efficiently preprocesses images, significantly reducing the subsequent computational burden, and the adaptive threshold mechanism can dynamically adjust to adapt to different water quality and lighting conditions, ensuring the accuracy and stability of preliminary recognition.

[0068] In this embodiment, it is specifically pointed out that the preliminary biological recognition based on the underwater image through the edge detection algorithm to obtain the preliminary biological image step is:

[0069] The underwater image is preprocessed, and the preprocessing includes denoising and light correction. The denoising is to smooth the image by using Gaussian filtering to suppress particle noise and fine suspended matter. The light correction is to use a contrast-limited adaptive histogram to equalize and compensate for non-uniform illumination, avoiding local overexposure or darkness.

[0070] The Sobel operator is used to calculate the gradient amplitude in the horizontal and vertical directions. The pixels with a gradient amplitude higher than a threshold value are set to 255 (white edge), and the rest are 0 (black background).

[0071] All closed edges are extracted using connected component analysis to obtain a preliminary organism image.

[0072] The organism determination module is used to fuse the sensor to obtain the organism information of the preliminary organism image. The organism information includes motion trajectory, morphological information, optical features, and size change. The organism determination index is obtained according to the organism information. The organism determination is performed according to the organism determination index, and the determined organism is obtained. The determined organism is transmitted to the biomass analysis module.

[0073] The module combines sensor data and visual features to calculate the organism determination index, accurately determines whether the target is a real organism. The core advantage is multi-feature fusion decision, which greatly reduces the false detection rate and supports dynamic weight adjustment to adapt to different biological species and environmental changes.

[0074] In this embodiment, it needs to be specifically pointed out that the organism determination index acquisition step is:

[0075] The preliminary organism image is taken as a target image, the position coordinate sequence of the target image in the continuous frames in the detection time period is obtained, and the motion purpose coefficient is evaluated according to the position coordinate sequence;

[0076] The target image contour data is obtained, and the irregularity coefficient is evaluated according to the target image contour data;

[0077] The gray level co-occurrence matrix is constructed by the target image, and the texture coefficient is calculated according to the gray level co-occurrence matrix;

[0078] The size data of the target image in the detection time period is obtained, and the size stability coefficient is calculated according to the size data in the detection time period;

[0079] The motion purpose coefficient, the irregularity coefficient, the texture coefficient, and the size stability coefficient are normalized, and the organism determination index is calculated according to the normalized motion purpose coefficient, the irregularity coefficient, the texture coefficient, and the size stability coefficient. The specific acquisition steps are:

[0080] LO = a1 x PE + a2 x IL + a3 x TU + a4 x DS;

[0081] In the formula, LO represents the biological determination index, PE represents the normalized motion purpose coefficient, when the target motion trajectory presents directionality and predictability, the motion purpose coefficient is higher, and the confidence of the system determining the real biological body is higher. This design is based on the fact that the motion of natural organisms is usually purposeful, while the motion of interference objects is usually passive and is characterized by randomness or irregularity. Therefore, this coefficient effectively distinguishes biological and non-biological targets by quantifying the intelligence of the motion pattern, and can significantly reduce the misjudgment rate, especially in complex scenes such as heat source areas or turbid water areas. IL represents the normalized irregularity coefficient, when the complexity of the target contour is higher (such as the presence of fins, gills, tentacles, etc.), the irregularity coefficient is larger, and the confidence of the system determining the real biological body is higher. This design is based on the fact that natural organisms usually have non-geometric shapes, while interference objects usually have near-circular or smooth edges. By quantifying the degree of deviation from the spherical shape of the contour, this coefficient effectively captures the unique morphological characteristics of organisms, which is particularly important in transparent organisms (such as jellyfish) or highly reflective environments, and can reduce misjudgments caused by similar shapes. TU represents the normalized texture coefficient, when the texture features of the target surface are more complex, the texture coefficient is higher, and the confidence of the system determining the real biological body is also improved. This correlation is based on the essential difference between biological and non-biological objects. The surface of a natural organism usually has rich microstructure, while the surface of a bubble or suspended matter is usually smooth or homogeneous. By strengthening the weight of texture features, this coefficient can effectively identify biological targets in low-contrast environments, which is particularly important for accurate detection of transparent or small organisms. DS represents the normalized size stability coefficient, when the physical size of the target remains relatively stable in a short period of time, the size stability coefficient is higher, and the confidence of the system determining the real biological body is significantly improved. This correlation is based on the difference in dynamic behavior. The growth or motion of a biological body is usually continuous, while the size of a bubble or other interference object may fluctuate dramatically due to merging, breaking or water flow impact. By tracking the area change rate of the target in the time sequence frame, false targets caused by environmental disturbances can be effectively filtered, which is particularly suitable for foam dense areas or turbulent scenes, and enhances the anti-interference ability of the system. a1, a2, a3, a4 represent the weight coefficients of the motion purpose coefficient, the weight coefficients of the irregularity coefficient, the weight coefficients of the texture coefficient, and the weight coefficients of the size stability coefficient, and a1+a2+a3+a4=1. a1, a2, a3, a4 are obtained by the analytic hierarchy process, which is a systematic decision-making method. By constructing a hierarchical structure model, a complex problem is divided into multiple interrelated elements, and a two-by-two comparison matrix is used to quantify the relative importance of each element. Finally, the weight coefficients are calculated by the characteristic vector. Its core advantage is to combine subjective experience and mathematical consistency test to ensure the rationality of weight distribution.In the embodiment, the analytic hierarchy process is used to determine the contribution weight of the motion purposiveness coefficient, the irregularity coefficient, the texture coefficient and the size stability coefficient to the organism determination index, so as to balance the discrimination priority of different characteristics in a complex environment.

[0082] In the embodiment, it is particularly pointed out that the motion purposiveness coefficient acquisition step is:

[0083] The position coordinate sequence of the target image in the continuous frames in the detection time period is acquired through the target tracking algorithm, the target tracking algorithm is a technology of generating a motion trajectory by associating the positions of the detected target in the continuous frames in a video sequence, and provides a data basis for motion purposiveness analysis;

[0084] The horizontal distance and the vertical distance between adjacent frames are calculated, the displacement vector between adjacent frames is calculated through the Euclidean distance, the Euclidean distance is a classical method for measuring the straight-line distance between two points, and in a two-dimensional space, it is expressed as the square sum of the horizontal and vertical coordinate differences of two points and then square root, and directly reflects the actual displacement length of the target. In target tracking, it is used to quantify the moving distance (in pixels) of the target center between adjacent frames, provides a basic physical quantity for motion analysis, is simple in calculation and conforms to the human cognition of "shortest path", is a standard calculation method of the displacement vector module length, and the specific acquisition steps are as follows:

[0085]

[0086] In the formula, d t represents the displacement vector of the tth frame, Δx t represents the horizontal distance from the tth frame to the (t-1) th frame, Δy t represents the vertical distance from the tth frame to the (t-1) th frame;

[0087] The degree of change of the motion direction is quantified through the included angle of adjacent displacement vectors, and the specific acquisition steps are as follows:

[0088]

[0089] In the formula, θ t represents the change angle of the motion direction between the tth frame and the (t-1) th frame, Δx t represents the horizontal distance from the tth frame to the (t-1) th frame, Δx t-1 represents the horizontal distance from the (t-1) th frame to the (t-2) th frame, Δy t represents the vertical distance from the tth frame to the (t-1) th frame, Δy t-1 represents the vertical distance from the (t-1) th frame to the (t-2) th frame, d t represents the displacement vector of the tth frame, d t-1 represents the displacement vector of the (t-1) th frame, and if θ tIf θ ≈ 0, it means the moving direction is consistent; if θ t If θ ≈ π, it means the direction changes suddenly.

[0090] The standard deviation of the moving direction change angle of each continuous frame in the detection time period is calculated, which reflects the stability of the direction. The variance of the displacement vector of each continuous frame in the detection time period is calculated, which reflects the fluctuation of the speed. The maximum displacement distance of the continuous frames in the detection time period is obtained. The purpose coefficient of the motion is calculated according to the standard deviation of the moving direction change angle, the variance of the displacement vector and the maximum displacement distance. The specific obtaining steps are as follows:

[0091]

[0092] In the formula, PE represents the purpose coefficient of the motion, σ θ represents the standard deviation of the moving direction change angle, Var(d) represents the variance of the displacement vector, max(d) represents the maximum displacement distance, PE ∈ [0, 1], and the value is closer to 1, the stronger the purpose of the motion.

[0093] In this embodiment, it needs to be specifically explained that the irregular coefficient obtaining step is as follows:

[0094] An edge detection algorithm is used on the target image to extract a set of outer contour pixel points of the target. A function is used to fit the minimum circumscribed circle of the contour to obtain the center and radius of the minimum circumscribed circle.

[0095] The distance of each contour point to the center is calculated, and the mean value is calculated to obtain the average center distance.

[0096] According to the distance of each contour point to the center and the average center distance, the standard deviation of the contour distance is calculated, which reflects the fluctuation degree of the contour.

[0097] The standard deviation of the contour distance and the average center distance are calculated by ratio, and the irregular coefficient is obtained. The closer the irregular coefficient is to 0, the closer the target image is to a circle. The closer the irregular coefficient is to 1, the greater the fluctuation of the target image contour is.

[0098] In this embodiment, it needs to be specifically explained that the texture coefficient obtaining step is as follows:

[0099] The target image is subjected to grayscale processing and converted into a single-channel grayscale image. The frequency of occurrence of pixel pairs in a specific direction and distance is counted. A gray level co-occurrence matrix is constructed according to the frequency of occurrence of each pixel pair. The gray level co-occurrence matrix is a statistical tool for quantifying the texture features of an image. A two-dimensional matrix is generated by calculating the gray level co-occurrence frequency of paired pixels in a specific direction and distance in the image. The core idea is that the adjacent pixel gray level difference of a texture complex region is large and changes frequently, which will be manifested as a scattered high value distribution in the matrix; and the smooth region has uniform gray level, and the matrix value is concentrated near the diagonal line. By analyzing the matrix, the roughness and regularity of the target surface can be objectively described, thereby distinguishing biological and non-biological targets.

[0100] The square of the gray level difference of each pair of adjacent pixels is calculated, denoted as the difference intensity.

[0101] According to the gray level co-occurrence matrix, the frequency of occurrence of all adjacent pixel pairs is counted, and the weighted sum of the frequency of occurrence of all adjacent pixel pairs and the difference intensity is calculated to obtain the texture coefficient.

[0102] In this embodiment, it needs to be specifically pointed out that the size stability coefficient acquisition step is:

[0103] The target image in the current frame is subjected to binary segmentation to obtain a mask value. The mask value is the pixel value of the target region after binary segmentation, and 0 (black) is usually used to represent the background and 1 (white) is used to represent the target organism. In image processing, the mask is a binary matrix with the same size as the original image, which distinguishes the target and the background through the 0 / 1 pixel value, and is convenient for subsequent calculation of the target area, contour or logical operation with other images. For example, after segmentation of fish, the pixel value in the contour is 1 and the water background is 0, which can be directly used to count the number of pixels occupied by the organism or measure the morphological parameters. The specific acquisition steps are:

[0104]

[0105] In the formula, A t represents the pixel area of the target image in the t frame, M t (x, y) represents the mask value of the target image in the t frame, 1 is the target pixel and 0 is the background, and W and H represent the width and height of the target image.

[0106] The area sequence in the continuous frames of the target image in the detection time period is obtained, and the area relative change rate is calculated frame by frame to reflect the stability.

[0107] The average area relative change rate in the detection time period is calculated, and the size stability coefficient is obtained according to the average area relative change rate. The specific acquisition steps are:

[0108] DS = 1 - min (AVGA ,1);

[0109] In the formula, DS represents the size stability coefficient, AVG A represents the average area relative change rate, min(AVG A ,1) is a minimum function, which is to truncate the result to a maximum value of 1 to prevent extreme cases. If the size stability coefficient tends to 1, it indicates a low change rate, such as slow growth of fish bodies. If the size stability coefficient tends to 0, it indicates a high change rate, such as bubble merging or breaking.

[0110] In this embodiment, it needs to be specifically explained that the organism determination is determined according to the organism determination index, and the determination organism step is obtained as follows:

[0111] The organism determination index is compared with the preset threshold value. If the organism determination index is greater than or equal to the preset threshold value, it is judged as a determined organism. If the organism determination index is less than the preset threshold value, it is judged as a non-organism. The preset threshold value is obtained by an adaptive threshold method. The adaptive threshold method is a technology for dynamically adjusting the discrimination threshold. The core is to automatically optimize the threshold setting by real-time analysis of the biological feature data distribution under the current environment, rather than relying on fixed empirical values.

[0112] The biomass analysis module is used to obtain the number, size and density of the determined organisms, and analyze the biomass according to the number, size and density of the determined organisms. The biomass is transmitted to the visualization module;

[0113] The module accurately calculates the total biomass based on the determined organism data. Its advantage is to provide high-precision quantitative indicators to directly guide breeding management decisions such as feeding amount adjustment or sorting strategies, while supporting group statistics to optimize resource allocation and growth monitoring.

[0114] In this embodiment, it needs to be specifically explained that the biomass acquisition step is as follows:

[0115] According to the area of each determined organism, the average biological area is calculated, and the biological density is calculated by ratio calculation of the number of determined organisms and the area of the underwater image;

[0116] The biomass is calculated according to the number of determined organisms, the average biological area and the biological density. The specific acquisition steps are as follows:

[0117] BS = NUM S × AVG C × BD;

[0118] In the formula, BS represents the biomass, NUM S represents the number of determined organisms, AVG C represents the average biological area, and BD represents the biological density.

[0119] A visualization module is used for visualizing biomass and updating real-time data.

[0120] The visualization module is the interactive core of the factory farming biomass monitoring system. Its role is to visually display the real-time calculated biomass data and environmental parameters through dynamic charts, digital panels, and 3D spatial distribution maps, support multi-view switching and historical trend backtracking. This module connects edge computing or cloud data streams to achieve second-level updates, helping farmers quickly grasp the growth state of the organism, density abnormalities, and feeding effects, assisting in decision-making optimization, and improving management efficiency and farming benefits.

[0121] Finally, the above is only the preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0122] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which shall be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A real-time monitoring device for biomass in factory farming integrating machine vision, characterized in that, The device includes; The image acquisition module is used to acquire underwater images through a high-resolution camera device, obtain underwater images, and transmit the underwater images to the initial biometric identification module. The sensor fusion module is used to fuse environmental sensors, depth sensors, and ultrasonic sensors to obtain a fused sensor. The preliminary biometrics module is used to perform preliminary biometrics based on underwater images using an edge detection algorithm, obtain a preliminary organism image, and transmit the preliminary organism image to the organism determination module; The organism identification module is used to acquire organism information from preliminary organism images through fusion sensors. The organism information includes movement trajectory, morphological information, optical characteristics, and size changes. An organism identification index is obtained based on the organism information. The organism is then identified based on the organism identification index to obtain the identified organism. The identified organism is then transmitted to the biomass analysis module. The biomass analysis module is used to obtain the number, size, and density of organisms, analyze the biomass based on the determined number, size, and density of organisms, and then transmit the biomass to the visualization module. The visualization module is used to visualize biomass and update data in real time. The steps for obtaining the organism determination index are as follows: Using the initial biological image as the target image, the position coordinate sequence of the target image in consecutive frames within the detection time period is obtained, and the motion purposefulness coefficient is evaluated based on the position coordinate sequence. Obtain the contour data of the target image, and evaluate the irregularity coefficient based on the contour data of the target image. A gray-level co-occurrence matrix is ​​constructed from the target image, and texture coefficients are calculated based on the gray-level co-occurrence matrix. Obtain the size data of the target image within the detection time period, and calculate the size stability coefficient based on the size data within the detection time period; The motion purposefulness coefficient, irregularity coefficient, texture coefficient, and dimensional stability coefficient are normalized. Based on these normalized coefficients, the organism determination index is calculated. The specific steps for obtaining this index are as follows: ; In the formula, Represented as an organismal determinant index. This is expressed as the normalized motion purposefulness coefficient. Represented as the normalized irregularity coefficient. Represented as normalized texture coefficients, This is expressed as the normalized dimensional stability coefficient. , , , The weighting coefficients are represented as the weighting coefficients for motion purposefulness, irregularity, texture, and dimensional stability.

2. The real-time monitoring device for biomass in factory farming integrating machine vision as described in claim 1, characterized in that: The step of performing preliminary biometric identification based on underwater images using an edge detection algorithm to obtain preliminary biological images is as follows: The underwater images are preprocessed, including noise reduction and illumination correction. The Sobel operator is used to calculate the gradient magnitude in the horizontal and vertical directions. Pixels with gradient magnitudes higher than the threshold are set to 255, and the rest are set to 0. Connectivity analysis was used to extract all closed edges, resulting in a preliminary image of the organism.

3. The real-time monitoring device for biomass in factory farming integrating machine vision as described in claim 1, characterized in that, The steps for obtaining the motion purposefulness coefficient are as follows: The target tracking algorithm is used to obtain the position coordinate sequence of the target image in consecutive frames within the detection time period; Calculate the horizontal and vertical distances between adjacent frames, and calculate the displacement vectors between adjacent frames using Euclidean distance; The degree of change in the direction of motion is quantified by the angle between adjacent displacement vectors. The specific steps are as follows: ; In the formula, It is represented as the angle of change in motion direction between frame t and frame (t-1). It represents the horizontal distance from frame t to frame (t-1). It is represented as the horizontal distance from frame t-1 to frame t-2. This represents the vertical distance from frame t to frame (t-1). This is represented as the vertical distance from frame t-1 to frame t-2. Let it be represented as the displacement vector of the t-th frame. It is represented as the displacement vector of the (t-1)th frame; The standard deviation of the motion direction change angle and the variance of the displacement vector for each consecutive frame within the detection time period are calculated. The maximum displacement distance of the consecutive frames within the detection time period is obtained. The motion purposefulness coefficient is calculated based on the standard deviation of the motion direction change angle, the variance of the displacement vector, and the maximum displacement distance. The specific steps for obtaining this coefficient are as follows: ; In the formula, Represented as the motion purposefulness coefficient, Expressed as the standard deviation of the angle of change of the direction of motion. Expressed as the variance of the displacement vector. This represents the maximum displacement distance.

4. The real-time monitoring device for biomass in factory farming integrating machine vision as described in claim 1, characterized in that: The steps for obtaining the irregularity coefficient are as follows: An edge detection algorithm is used on the target image to extract the set of pixels of the outer contour of the target, and the minimum circumcircle of the contour is fitted to obtain the center and radius of the minimum circumcircle. Calculate the distance from each contour point to the center of the circle, and then calculate the average distance to the center of the circle. The standard deviation of the contour distance is calculated based on the distance from each contour point to the center of the circle and the average distance to the center of the circle. The irregularity coefficient is obtained by calculating the ratio of the standard deviation of the contour distance to the average center distance.

5. The real-time monitoring device for biomass in factory farming integrating machine vision as described in claim 1, characterized in that: The steps for obtaining the texture coefficients are as follows: The target image is converted to grayscale and transformed into a single-channel grayscale image. Based on the set direction and distance, the frequency of pixel pairs is counted, and a grayscale co-occurrence matrix is ​​constructed according to the frequency of each pixel pair. For each pair of adjacent pixels, calculate the square of the grayscale difference, and denot it as the difference intensity; Based on the gray-level co-occurrence matrix, the occurrence frequency of all adjacent pixel pairs is counted, and the texture coefficient is calculated by weighted summation of the occurrence frequency and difference intensity of all adjacent pixel pairs.

6. The real-time monitoring device for biomass in factory farming integrating machine vision as described in claim 1, characterized in that: The steps for obtaining the size stability coefficient are as follows: Perform binary segmentation on the target image to obtain the mask value, and calculate the pixel area of ​​the target image based on the mask value; The area sequence of consecutive frames of the target image is obtained within the detection time period, and the relative rate of change of area is calculated frame by frame. Calculate the average relative change rate of area during the detection period, and obtain the size stability coefficient based on the average relative change rate of area.

7. The real-time monitoring device for biomass in factory farming integrating machine vision as described in claim 1, characterized in that: The steps for determining an organism based on an organism identification index are as follows: The organism identification index is compared with a preset threshold. If the organism identification index is greater than or equal to the preset threshold, it is identified as an organism; if the organism identification index is less than the preset threshold, it is identified as a non-organism.

8. The real-time monitoring device for biomass in factory farming integrating machine vision as described in claim 1, characterized in that: The biomass acquisition steps are as follows: The average biological area is calculated based on the area of ​​each identified organism, and the biological density is calculated by the ratio of the number of identified organisms to the area of ​​the underwater image. Biomass is calculated based on the determined number of organisms, average biological area, and biological density. The specific steps for obtaining this biomass are as follows: ; In the formula, Expressed as biomass, This indicates the number of organisms. This is expressed as average biological area. Represented as biological density.

Images