Fish size measurement system based on 3D visual reconstruction

Abstract

The invention discloses a fish size measurement system based on 3D visual reconstruction, and relates to the technical field of fish measurement, the system realizes high-precision measurement of underwater fish body size by fusing acoustic and visual data, the acoustic data obtains information such as reflection intensity, propagation distance, frequency change and the like of the surface of a fish body in real time through a distance measuring sensor, and the measurement accuracy is improved. And the visual data is used for extracting geometrical characteristics of the surface of the fish body through an underwater image acquisition system. After acoustic and visual data are subjected to feature extraction and three-dimensional reconstruction processing, the size of the fish body can be effectively calculated. By constructing the three-dimensional fusion model, carrying out weighted fusion on sound wave and visual data and outputting the fused fish body size Xfusion, the accuracy of underwater fish body size measurement is greatly improved, especially in a complex environment, the influence of factors such as underwater illumination and transparency on visual measurement is overcome, and high reliability of a measurement result is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of fish measurement, and in particular to a fish size measurement system based on 3D vision reconstruction. Background Art

[0002] With the continuous development of science and technology, measurement technology has been widely used in many fields, especially in underwater environments. Underwater detection technology involves multiple fields such as ocean exploration, deep-sea exploration, and fishery resource monitoring. In the field of fisheries, in particular, the size and number of fish are one of the key factors affecting the efficiency of fishery production. Traditional fish measurement methods mostly rely on manual measurement or simple estimation based on sonar. These methods are not only inefficient but also difficult to guarantee accuracy in practical applications. In recent years, with the development of 3D visual reconstruction technology, the technology of combining acoustic and visual data to measure fish size has gradually become a research hotspot. By combining visual and acoustic data, efficient and accurate measurement of underwater fish bodies can be achieved, especially in complex underwater environments with low visibility, which has important application prospects.

[0003] Currently, traditional underwater fish size measurement techniques rely primarily on acoustic or manual measurements, which have numerous shortcomings. First, while acoustic-based measurement techniques can achieve relatively simple distance estimation in underwater environments, they can only provide general information about the fish and lack a detailed description of its surface morphology, leading to large measurement errors. Furthermore, relying solely on visual sensors for 3D reconstruction of underwater fish is affected by factors such as underwater lighting and turbid water quality, making it impossible to stably obtain accurate data in complex underwater environments. While combining acoustic and visual data has great potential, effectively combining their strengths, eliminating noise interference, and improving fusion accuracy remain major challenges in their implementation. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a fish size measurement system based on 3D visual reconstruction, which solves the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0006] Acoustic and visual data acquisition module: Distance measurement sensors and visual sensors are installed in the test smart fish pond to collect acoustic and visual data of underwater fish in real time, and transmit the acoustic and visual data to the fish measurement system on the ground;

[0007] Data processing module: Extract features from the acoustic and visual data in the fish measurement system, obtain the reflection intensity R and the fish surface area A, perform preprocessing, obtain the acoustic fish 3D vector Xsonic and the visual fish 3D vector Xvis, then construct a fish database and store the acoustic fish 3D vector Xsonic and the visual fish 3D vector Xvis in the fish database;

[0008] 3D fusion module: Build a 3D fusion model, fuse the sonic fish body 3D vector Xsonic and the visual fish body 3D vector Xvis, and calculate the fused fish body size Xfusion;

[0009] Fish size analysis module: After fusion, the fish size is inferred using the reverse calculation formula based on the fused fish size Xfusion, and the measured fish size Sfinal is obtained by integrating the fish size.

[0010] Error analysis module: Calculates the difference between the total fish size Sfinal and the actual fish size in the test smart fish pond, outputs the deviation value Sreal, performs error evaluation, and triggers the error optimization mechanism based on the error evaluation results.

[0011] Preferably, the acoustic and visual data acquisition module includes a data acquisition unit and a data transmission unit;

[0012] The data acquisition unit sets up an array of monitoring points at equal intervals in the test smart fish pond, and installs a ranging sensor and a visual sensor in each monitoring point. The ranging sensor is set to a fixed acoustic signal to collect acoustic and visual data in real time.

[0013] Acoustic wave data includes propagation characteristics, frequency variations, and reflection characteristics;

[0014] Visual data includes underwater fish images;

[0015] The data transmission unit establishes a communication connection between the fish body measurement system and the communication modules of the ranging sensor and the visual sensor by setting up underwater electromagnetic communication, and transmits the collected acoustic wave data and visual data to the fish body measurement system.

[0016] Preferably, the data processing module includes an acoustic wave feature extraction unit, an image feature extraction unit, a data integration processing unit and a data storage unit;

[0017] The acoustic wave feature extraction unit extracts the propagation distance d, the acoustic wave incident angle Rs and the azimuth angle Fw of the acoustic wave signal by performing feature extraction based on the acoustic wave data;

[0018] The propagation time t of the output sound wave signal to the surface of the fish body is calculated based on the double value of the propagation distance d of the sound wave signal divided by the propagation speed of the sound wave in water;

[0019] Based on the specific surface characteristics combined with the sound wave incident angle Rs and azimuth Fw, the reflection intensity R is extracted; then, based on the relationship between the reflection intensity R and the distance, the geometric dimensions of the fish in all directions are calculated. The acoustic fish size of the fish is gradually derived through the reflection intensity R at different angles.

[0020] The acoustic fish body size includes the acoustic fish body length Ls, the acoustic fish body width Ws and the acoustic fish body height Hs.

[0021] Preferably, the image feature extraction unit performs depth processing on the visual data by image processing technology, extracts the depth value Z of the fish body surface on the pixel point (x, y) plane in the visual data, combines the pixel point (x, y) coordinates of the two-dimensional plane, and uses a three-dimensional reconstruction algorithm to convert the image in the visual data into three-dimensional space coordinates;

[0022] After 3D reconstruction, the visual size of the fish is preliminarily estimated based on the size and angle of the visual data and the geometric model of the fish body. The visual size includes the fish body length Lv, fish body width Wv and fish body height Hv.

[0023] Based on the depth value Z of the fish body surface on the pixel point (x, y) plane, the curvature of each pixel point (x, y) on the fish body surface is calculated, and the surface is discretized and integrated using numerical methods to obtain the fish body surface area A.

[0024] Preferably, the data integration processing unit obtains the acoustic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis by summarizing the obtained visual fish body size and acoustic fish body size, wherein Xsonic=(Ls, Ws, Hs), Xvis=(Lv, Wv, Hv);

[0025] The data storage unit constructs a fish body database to automatically encode each fish, and associates and stores the corresponding encoded sonic fish body three-dimensional vector Xsonic and visual fish body three-dimensional vector Xvis into the fish body database.

[0026] Preferably, the three-dimensional fusion module constructs a three-dimensional fusion model, extracts the sonic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis, inputs the extracted data into the three-dimensional fusion model, adjusts the fusion size of the sonic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis by introducing a fusion coefficient a, and outputs the fused fish body size Xfusion, where the fused fish body size Xfusion represents the fused fish body length Lf, the fused fish body width Wf, and the fused fish body height Hf;

[0027] The fused fish size Xfusion is output through the following 3D fusion model;

[0028] Xfusion=a·Xvis+(1-a)·Xsonic;

[0029] Where a represents the fusion coefficient, which controls the relative weight of visual data and acoustic data and is initially set to 0.5.

[0030] Preferably, the fish size analysis module includes a back-estimation unit and a size analysis unit;

[0031] The reverse calculation unit uses the reverse calculation formula to reversely calculate the fish body size (L, W, H) based on the fusion fish body size Xfusion obtained after fusion. The reverse calculation formula uses a linear regression algorithm for reverse calculation optimization.

[0032] The size analysis unit integrates the fish body dimensions (L, W, H) to perform comprehensive calculations and output the total fish body size Sfinal.

[0033] Preferably, the error analysis module includes an error calculation unit, an error evaluation unit and an error optimization unit;

[0034] The error calculation unit is used to calculate the difference between the actual measured fish body size and the obtained total fish body size Sfinal and output a deviation value Sreal.

[0035] Preferably, the error evaluation unit sets the allowable deviation range [-0.5, 0.5] according to the allowable error by the user, performs error evaluation based on the output result of the deviation value Sreal, analyzes the detection accuracy of the fish body measurement system, and triggers the error optimization mechanism according to the error evaluation result. The specific evaluation content is as follows;

[0036] When -0.5≤deviation value Sreal≤0.5, it indicates that the measurement accuracy of the fish body measurement system is normal. At this time, the test ends and the current fish body measurement system is executed;

[0037] When the deviation value Sreal>0.5, it indicates that the measurement accuracy of the fish body measurement system is positively abnormal, and the error optimization mechanism is triggered.

[0038] When the deviation value Sreal≤-0.5, it indicates that the measurement accuracy of the fish body measurement system is abnormal, and the error optimization mechanism is triggered.

[0039] Preferably, the error optimization unit triggers the error optimization mechanism by outputting the deviation value Sreal as the objective function, deriving the objective function with respect to the fusion coefficient a, obtaining the gradient of the error function, and continuously updating the fusion coefficient a using the gradient descent formula until the error function converges and stops updating. The updated fusion coefficient a is substituted into the three-dimensional fusion model for iterative analysis until the measurement accuracy reaches normal and the iteration and update are stopped.

[0040] The present invention provides a fish size measurement system based on 3D visual reconstruction. It has the following beneficial effects:

[0041] (1) The system achieves high-precision measurement of underwater fish size by fusing acoustic and visual data. Acoustic data uses ranging sensors to obtain real-time information such as the reflection intensity, propagation distance, and frequency changes of the fish surface, while visual data uses an underwater image acquisition system to extract the geometric features of the fish surface. After feature extraction and three-dimensional reconstruction, the acoustic and visual data can effectively calculate the size of the fish. By constructing a three-dimensional fusion model, the acoustic and visual data are weightedly fused to output the fused fish size Xfusion, which greatly improves the accuracy of underwater fish size measurement. Especially in complex environments, it overcomes the influence of underwater lighting, transparency and other factors on visual measurement, ensuring the high reliability of the measurement results.

[0042] (2) The system achieves dynamic optimization of fish body measurement accuracy by introducing an error analysis and optimization mechanism. When there is a deviation between the measurement result and the actual fish body size, the system automatically calculates the error and evaluates the detection accuracy. By setting the allowable deviation range [-0.5, 0.5], the system triggers the error optimization mechanism according to the error value Sreal, automatically adjusts the fusion coefficient a, and uses the gradient descent algorithm to update the fusion coefficient until the error function converges, effectively improving the measurement accuracy. This mechanism can automatically adapt to changes in the underwater environment, ensuring that each measurement can be performed with optimal accuracy, reducing the need for human intervention, and significantly improving the stability and robustness of the system. It is particularly suitable for long-term, dynamically changing underwater monitoring environments.

[0043] (3) The system achieves efficient data processing and storage by integrating acoustic feature extraction, image feature extraction and data integration processing units. The fish body measurement system can collect and process acoustic data and visual data in real time. Through an intelligent data storage mechanism, the acoustic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis are automatically associated and stored in the fish body database to ensure the integrity and traceability of the data. By automatically encoding and storing the measurement data of each fish, the system can quickly perform data comparison and analysis, while providing convenient support for subsequent data mining and big data analysis. The intelligent design of this module reduces manual intervention, optimizes data access and processing speed, and improves the overall efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a flow chart of a fish size measurement system based on 3D visual reconstruction according to the present invention;

[0045] Figure 2This is a data flow diagram of a fish size measurement system based on 3D visual reconstruction in the present invention. DETAILED DESCRIPTION

[0046] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0047] Example 1

[0048] See also Figure 1 and Figure 2 The present invention provides a fish size measurement system based on 3D visual reconstruction. To achieve the above purpose, the present invention is implemented through the following technical solutions:

[0049] Acoustic and visual data acquisition module: Distance measurement sensors and visual sensors are installed in the test smart fish pond to collect acoustic and visual data of underwater fish in real time, and transmit the acoustic and visual data to the fish measurement system on the ground;

[0050] Data processing module: Extract features from the acoustic and visual data in the fish measurement system, obtain the reflection intensity R and the fish surface area A, perform preprocessing, obtain the acoustic fish 3D vector Xsonic and the visual fish 3D vector Xvis, then construct a fish database and store the acoustic fish 3D vector Xsonic and the visual fish 3D vector Xvis in the fish database;

[0051] 3D fusion module: Build a 3D fusion model, fuse the sonic fish body 3D vector Xsonic and the visual fish body 3D vector Xvis, and calculate the fused fish body size Xfusion;

[0052] Fish size analysis module: After fusion, the fish size is inferred using the reverse calculation formula based on the fused fish size Xfusion, and the measured fish size Sfinal is obtained by integrating the fish size.

[0053] Error analysis module: Calculates the difference between the total fish size Sfinal and the actual fish size in the test smart fish pond, outputs the deviation value Sreal, performs error evaluation, and triggers the error optimization mechanism based on the error evaluation results.

[0054] In this embodiment, the system significantly improves the accuracy and stability of underwater fish size measurement through precise multi-module data acquisition, processing and fusion. The specific implementation method is as follows: a ranging sensor and a visual sensor are installed in the test smart fish pond to collect acoustic and visual data of underwater fish in real time and transmit it to the ground fish measurement system. The data processing module uses feature extraction technology to obtain reflection intensity and propagation characteristics from acoustic wave data, and extracts fish surface image information from visual data. Then, through three-dimensional reconstruction and preprocessing of acoustic and visual data, it generates acoustic and visual three-dimensional vector information and stores it in the database. The three-dimensional fusion module establishes a fusion model to perform weighted fusion of the three-dimensional vectors of acoustic and visual to generate the fused fish size Xfusion, and further deduce the final size of the fish Sfinal. If there is a deviation between the system measurement result and the actual fish size, the error analysis module will automatically calculate and output the deviation value Sreal, trigger error optimization through the error evaluation mechanism, automatically adjust the system parameters, and optimize the measurement accuracy. Through this implementation, the present invention achieves high-precision, real-time measurement of underwater fish dimensions, effectively addressing measurement errors caused by factors such as illumination, transparency, and distance in traditional measurement methods. Furthermore, by introducing an error optimization mechanism, the system can adapt to environmental changes, continuously improve measurement accuracy, and ensure measurement stability under diverse environmental conditions. Furthermore, intelligent data storage and management significantly enhances the system's data processing efficiency, providing reliable support for subsequent data analysis and multiple measurements.

[0055] Example 2

[0056] Specifically: the acoustic and visual data acquisition module includes a data acquisition unit and a data transmission unit;

[0057] The data acquisition unit sets up an array of monitoring points at equal intervals in the test smart fish pond, and installs a ranging sensor and a visual sensor in each monitoring point. The ranging sensor is set to a fixed acoustic signal to collect acoustic and visual data in real time.

[0058] Acoustic wave data includes propagation characteristics, frequency variations, and reflection characteristics;

[0059] Visual data includes underwater fish images;

[0060] The ranging sensor transmits sound wave signals to the fish in the water, which are reflected by the surface of the fish. The sensor receives the reflected echo signal and records the signal's propagation time, frequency change, and reflection intensity data.

[0061] The data transmission unit establishes a communication connection between the fish body measurement system and the communication modules of the ranging sensor and the visual sensor by setting up underwater electromagnetic communication, and transmits the collected acoustic wave data and visual data to the fish body measurement system.

[0062] In this embodiment, the system establishes an array of equidistant monitoring points within the test smart fish pond, with both a ranging sensor and a visual sensor installed at each point. The ranging sensor transmits a fixed-frequency acoustic signal to collect real-time acoustic data from the fish in the water, including signal propagation characteristics, frequency variation, and reflection intensity. Simultaneously, the visual sensor collects underwater image data of the fish. When the sound waves propagating through the water encounter the fish's surface, they reflect. The ranging sensor receives the reflected echo signal and records the propagation time, frequency variation, and reflection characteristics to obtain the acoustic characteristics of the fish. The visual sensor provides image information of the fish's surface for subsequent three-dimensional reconstruction and size estimation. All collected data is transmitted via an underwater electromagnetic communication transmission unit to a surface-based fish measurement system for further data processing and fusion analysis. Through this implementation, the acquisition module of the present invention enables real-time, synchronous collection of acoustic and visual data in underwater environments, effectively improving the multidimensionality and accuracy of the measured data. Compared to traditional measurement methods, this system not only overcomes issues such as unstable underwater lighting and limited visibility, but also complements visual data with acoustic data, providing more comprehensive and accurate fish information. In addition, through the application of underwater electromagnetic communication technology, the system achieves efficient data transmission, ensures the rapid delivery and processing of real-time data, and greatly improves the reliability of data acquisition and the system response speed.

[0063] Example 3

[0064] Specifically: the data processing module includes an acoustic feature extraction unit, an image feature extraction unit, a data integration processing unit and a data storage unit;

[0065] The acoustic feature extraction unit extracts the propagation distance d of the acoustic signal from the emission point to the fish body surface and then to the receiving point, the acoustic wave incident angle Rs and the azimuth angle Fw by performing feature extraction based on the acoustic wave data;

[0066] The propagation time t of the output sound wave signal to the surface of the fish body is calculated based on the double value of the propagation distance d of the sound wave signal divided by the propagation speed of the sound wave in water;

[0067] Based on the specific surface characteristics combined with the sound wave incident angle Rs and azimuth Fw, the reflection intensity R is extracted; then, based on the relationship between the reflection intensity R and the distance, the geometric dimensions of the fish in all directions are calculated. The acoustic fish size of the fish is gradually derived through the reflection intensity R at different angles.

[0068] The acoustic fish body size includes the acoustic fish body length Ls, the acoustic fish body width Ws and the acoustic fish body height Hs;

[0069] For example, the length of the fish body can be calculated based on the reflection intensity of the fish head and tail.

[0070] The specific extraction algorithm of reflection intensity R is: Where R(Rs, Fw) represents the reflection intensity at the incident angle Rs and azimuth Fw, C(Rs, fw) represents the area of a small region on the fish surface at the incident angle Rs and azimuth Fw, Cos represents the cosine function, and σ represents the sound wave reflection coefficient, which depends on the surface material of the fish body. The material of the fish scales and the surface roughness will affect the reflection intensity, as shown in Table 1.

[0071] Table 1:

[0072] Fish scale material Reflection coefficient σ surface properties scaleless fish 0.5 Smooth, partially reflective scaly fish 0.7 Rough, strong reflection Porous or soft surfaces 0.2 Soft tissue, highly absorbable

[0073] The image feature extraction unit performs deep processing on the visual data through image processing technology, extracts the depth value Z of the fish body surface on the pixel point (x, y) plane in the visual data, combines the pixel point (x, y) coordinates of the two-dimensional plane, and uses a three-dimensional reconstruction algorithm to convert the image in the visual data into three-dimensional space coordinates;

[0074] After 3D reconstruction, the visual size of the fish is preliminarily estimated based on the size and angle of the visual data and the geometric model of the fish body. The visual size includes the fish body length Lv, fish body width Wv and fish body height Hv.

[0075] Based on the depth value Z of the fish body surface on the pixel point (x, y) plane, the curvature of each pixel point (x, y) on the fish body surface is calculated, and the surface is discretized and integrated using numerical methods to obtain the fish body surface area A.

[0076] The data integration processing unit summarizes the obtained visual fish body size and acoustic fish body size respectively to obtain the acoustic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis, where Xsonic = (Ls, Ws, Hs) and Xvis = (Lv, Wv, Hv).

[0077] The data storage unit constructs a fish body database to automatically encode each fish, and associates and stores the corresponding encoded sonic fish body three-dimensional vector Xsonic and visual fish body three-dimensional vector Xvis into the fish body database.

[0078] In this embodiment, the system's acoustic feature extraction unit accurately extracts features such as distance and reflection intensity from acoustic data, and infers the geometric dimensions of the fish based on the propagation characteristics of underwater acoustic waves. The acoustic reflection intensity extraction algorithm further improves the accurate description of the fish's morphology based on different surface characteristics. Secondly, the image feature extraction unit deeply processes visual data and, in combination with a three-dimensional reconstruction algorithm, converts the two-dimensional image data into three-dimensional spatial coordinates, further inferring the visual dimensions of the fish. The surface area A of the fish is obtained by discretizing and integrating the surface. Furthermore, the data integration processing unit fuses acoustic and visual data, generating three-dimensional vectors to aggregate different types of dimensional information, providing high-quality input data for subsequent dimensional analysis and error optimization. The data storage unit establishes a fish database and associates and stores the corresponding dimensional data according to each fish's code, facilitating subsequent data query and optimization. Through this data processing process, the present invention effectively integrates acoustic and visual information, improving the overall accuracy of the underwater fish measurement system. Compared to traditional single-source data sources, this method, through multi-source data fusion, not only overcomes the limitations of single-sensor data but also enhances the ability to identify complex fish morphologies. This is particularly true in underwater environments with low light levels or limited visibility. Acoustic data complements visual data, ensuring comprehensive and efficient measurement results. Furthermore, the reflection intensity analysis algorithm and three-dimensional reconstruction technology employed improve the accuracy and real-time nature of fish size estimation, providing more precise and stable measurement results in practical applications, significantly enhancing the practicality and adaptability of underwater fish monitoring systems.

[0079] Example 4

[0080] Specifically: the three-dimensional fusion module constructs a three-dimensional fusion model, extracts the sonic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis, inputs the two into the three-dimensional fusion model, adjusts the fusion size of the sonic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis by introducing a fusion coefficient a, and outputs the fused fish body size Xfusion, where the fused fish body size Xfusion represents the fused fish body length Lf, the fused fish body width Wf, and the fused fish body height Hf;

[0081] The fused fish size Xfusion is output through the following 3D fusion model;

[0082] Xfusion=a·Xvis+(1-a)·Xsonic;

[0083] Where a represents the fusion coefficient, which controls the relative weight of visual data and acoustic data. It is initially set to 0.5, that is, the weights of the sonic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis are equal.

[0084] In this embodiment, the system effectively improves the accuracy and robustness of fish size measurement by introducing a deep fusion of acoustic and visual data. Specifically, the acoustic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis are first extracted, representing the fish size information obtained by the acoustic and visual data, respectively. These two three-dimensional vectors are then input into a three-dimensional fusion model, and the relative weights of the acoustic and visual data are adjusted by introducing a fusion coefficient a. By adjusting the fusion coefficient a, the module can flexibly adjust the contributions of acoustic and visual data according to different water environments, fish body characteristics, or measurement requirements, further optimizing measurement accuracy. The beneficial effect of this embodiment is that the three-dimensional fusion model can achieve higher-precision size estimation through the complementarity of data. Compared to traditional methods that rely solely on acoustic or visual data, the fusion strategy fully utilizes the respective advantages of acoustic and visual data, ensuring that in complex underwater environments, whether there is insufficient light, complex fish morphology, or turbid water, the system can provide more accurate fish size measurements. The fused size calculation not only improves the system's measurement accuracy, but also enhances the system's adaptability and reliability, especially in real-time monitoring and large-scale fish school analysis, and can provide more stable and accurate measurement results.

[0085] Example 5

[0086] Specifically: the fish body size analysis module includes a reverse estimation unit and a size analysis unit;

[0087] The reverse calculation unit uses the reverse calculation formula to reversely calculate the fish body size (L, W, H) based on the fusion fish body size Xfusion obtained after fusion. The reverse calculation formula uses a linear regression algorithm for reverse calculation optimization.

[0088] The fish body size (L, W, H) is calculated and output using the following algorithm formula;

[0089]

[0090] Where L represents the length of the fish body, W represents the width of the fish body, and H represents the height of the fish body. k1, k2, and k3 represent the regression coefficients of the fused fish body length Lf, fused fish body width Wf, and fused fish body height Hf in the fused fish body size Xfusion, respectively. b1, b2, and b3 represent the bias terms of the fused fish body length Lf, fused fish body width Wf, and fused fish body height Hf in the fused fish body size Xfusion, respectively, which are used as correction terms to adjust the output value.

[0091] The values of the regression coefficient k and the bias term b are automatically adjusted through optimization calculations, such as least squares regression, stochastic gradient descent, etc.

[0092] The size analysis unit integrates the fish body dimensions (L, W, H) to perform comprehensive calculations and output the total fish body size Sfinal.

[0093] In this embodiment, the method further improves the accuracy of fish body size estimation and the system's adaptability by introducing a back-calculation unit and a size analysis unit. The specific implementation method is to first use the back-calculation unit to back-calculate and optimize the true size of the fish body based on the fused fish body size Xfusion obtained after fusion using a linear regression algorithm. The back-calculation formula is adjusted by the regression coefficient k and the bias term b, and these parameters are automatically adjusted using optimization algorithms such as least squares regression or stochastic gradient descent to obtain a more accurate size estimate. In the size analysis unit, the fish body dimensions (L, W, H) obtained by back-calculation are integrated to calculate the final total fish body size Sfinal. This size comprehensively considers the measurement errors and optimization results of all dimensions, providing a more accurate basis for subsequent fish body monitoring and data analysis. The calculation formula for the total fish body size takes into account the mutual influence of each dimension, making the measurement results more reliable. By introducing the linear regression back-calculation algorithm and automatically adjusting the regression coefficient, the size deviation caused by measurement errors or changes in the external environment can be effectively corrected, further improving the accuracy and robustness of the system. Compared to traditional fixed-parameter measurement methods, this module automatically optimizes the calculation model based on actual measurement data, adjusting the regression coefficient and bias term in real time to accommodate different fish species and environmental changes, thereby ensuring high accuracy and consistency in measurement results. Furthermore, the sizing analysis module's comprehensive calculation of fish dimensions makes the final total fish size (Sfinal) more representative, providing accurate and reliable data support for subsequent scientific research and practical applications.

[0094] Example 6

[0095] Specifically: the error analysis module includes an error calculation unit, an error evaluation unit and an error optimization unit;

[0096] The error calculation unit is used to calculate the difference between the actual measured fish body size and the obtained total fish body size Sfinal and output a deviation value Sreal.

[0097] The error evaluation unit sets the allowable deviation range [-0.5, 0.5] according to the user's allowable error. Based on the output result of the deviation value Sreal, it performs error evaluation, analyzes the detection accuracy of the fish body measurement system, and triggers the error optimization mechanism according to the error evaluation result. The specific evaluation contents are as follows;

[0098] When -0.5≤deviation value Sreal≤0.5, it indicates that the measurement accuracy of the fish body measurement system is normal. At this time, the test ends and the current fish body measurement system is executed;

[0099] When the deviation value Sreal>0.5, it indicates that the measurement accuracy of the fish body measurement system is positively abnormal, and the error optimization mechanism is triggered.

[0100] When the deviation value Sreal≤-0.5, it indicates that the measurement accuracy of the fish body measurement system is abnormal, and the error optimization mechanism is triggered.

[0101] After the error evaluation triggers the error optimization mechanism, the error optimization unit outputs the deviation value Sreal as the objective function, derives the objective function with respect to the fusion coefficient a, obtains the gradient of the error function, and continuously updates the fusion coefficient a using the gradient descent formula until the error function converges and stops updating. The updated fusion coefficient a is substituted into the three-dimensional fusion model for iterative analysis until the measurement accuracy is normal and the iteration and update are stopped. Among them, gradient descent is an iterative method used to solve least squares problems (both linear and nonlinear). When solving the model parameters of machine learning algorithms, that is, unconstrained optimization problems, gradient descent is one of the most commonly used methods. Another commonly used method is the least squares method. When solving the minimum value of the loss function, the gradient descent method can be used to iterate step by step to obtain the minimized loss function and model parameter values. Conversely, if the maximum value of the loss function needs to be solved, the gradient ascent method needs to be used for iteration. In machine learning, two gradient descent methods have been developed based on the basic gradient descent method, namely the stochastic gradient descent method and the batch gradient descent method.

[0102] In this embodiment, the system further enhances the adaptive adjustment capability and accuracy assurance of the fish measurement system by introducing an error calculation unit, an error evaluation unit, and an error optimization unit. A specific implementation involves first calculating the difference between the actual measured fish size and the obtained total fish size Sfinal through the error calculation unit to obtain a deviation value Sreal, thereby quantifying the measurement error. The deviation value calculation ensures the accuracy of the error evaluation by considering the number of data points and the difference between the actual fish size and the measurement result. In the error evaluation unit, the system evaluates the deviation value Sreal based on a preset allowable deviation range. When the measurement accuracy is normal, the system terminates the test and executes the current measurement result. If the deviation value exceeds the normal range, the error optimization mechanism is triggered. This mechanism determines in real time whether further optimization is needed based on the error evaluation results, ensuring the reliability of the measurement accuracy. The error optimization unit optimizes the error using a gradient descent algorithm by introducing a fusion coefficient a as the objective function. The fusion coefficient a is continuously adjusted iteratively until the error function converges and meets the set accuracy requirements. This process effectively corrects errors caused by external factors or the system itself, resulting in more accurate measurement results and avoiding systematic biases caused by a single data source. Through real-time error calculation and optimization, it significantly improves system stability and measurement accuracy. Dynamic adjustment of measurement parameters allows the system to maintain high accuracy under varying environmental conditions. In particular, the introduction of an error optimization mechanism enables the system to automatically adjust itself in the event of measurement errors, effectively improving the reliability and consistency of fish measurements.

[0103] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. A fish size measurement system based on 3D visual reconstruction, characterized in that: include: Acoustic and visual data acquisition module, used to collect acoustic and visual data of underwater fish in real time; a data processing module for performing feature extraction on the acoustic wave data and the visual data to obtain a reflection intensity R and a fish body surface area A, preprocessing the reflection intensity R and the fish body surface area A to obtain an acoustic wave fish body three-dimensional vector Xsonic and a visual fish body three-dimensional vector Xvis, reconstructing a fish body database, and storing the acoustic wave fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis in the fish body database; A three-dimensional fusion module is used to construct a three-dimensional fusion model, fuse the sonic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis, and calculate the fused fish body size Xfusion; The fish size analysis module uses the reverse calculation formula to reversely calculate the fish size of the underwater fish based on the fused fish size Xfusion, and integrates the fish size to obtain the measured total fish size Sfinal; The error analysis module is used to calculate the difference between the total fish size Sfinal and the actual fish size in the test smart fish pond, output the deviation value Sreal, perform error evaluation on the deviation value Sreal, and trigger the error optimization mechanism based on the error evaluation result.

2. A fish size measurement system based on 3D visual reconstruction according to claim 1, characterized in that: The acoustic and visual data acquisition module includes a data acquisition unit and a data transmission unit; The data acquisition unit is configured by setting an array of monitoring points at equal intervals in the test smart fish pond, and installing a ranging sensor and a visual sensor in each monitoring point, wherein the ranging sensor is configured to generate a fixed acoustic wave signal to collect acoustic wave data and visual data in real time; The sound wave data includes propagation characteristics, frequency changes and reflection characteristics; The visual data includes underwater fish images; The data transmission unit establishes a communication connection between the fish body measurement system and the communication modules of the distance sensor and the visual sensor by setting up underwater electromagnetic communication, and transmits the collected acoustic wave data and visual data to the data processing module.

3. A fish size measurement system based on 3D visual reconstruction according to claim 2, characterized in that: The data processing module includes an acoustic wave feature extraction unit, which is used to extract features from the acoustic wave data to obtain the propagation distance d, the acoustic wave incident angle Rs and the azimuth Fw of the acoustic wave signal; and calculate the propagation time t of the acoustic wave signal to the surface of the fish body based on the propagation distance d; Based on the surface characteristics, combined with the sound wave incident angle Rs and azimuth angle Fw, the reflection intensity R is calculated and extracted; then, based on the relationship between the reflection intensity R and the distance, the geometric dimensions of the fish in all directions are calculated. The acoustic fish size of the fish is gradually derived based on the reflection intensity R at different angles; The acoustic fish body size includes the acoustic fish body length Ls, the acoustic fish body width Ws and the acoustic fish body height Hs.

4. A fish size measurement system based on 3D visual reconstruction according to claim 1, characterized in that: The data processing module also includes an image feature extraction unit, which performs depth processing on the visual data through image processing technology, extracts the depth value Z of the fish body surface on the pixel point (x, y) plane in the visual data, combines the pixel point (x, y) coordinates of the two-dimensional plane, and uses a three-dimensional reconstruction algorithm to convert the image in the visual data into three-dimensional space coordinates; After the three-dimensional reconstruction, the visual body size of the fish is preliminarily estimated based on the size and angle of the visual data and the geometric model of the fish body. The visual body size includes the fish body length Lv, the fish body width Wv and the fish body height Hv; Based on the depth value Z of the fish body surface on the pixel point (x, y) plane, the curvature of each pixel point (x, y) on the fish body surface is calculated, and the surface is discretized and integrated using a numerical method to obtain the fish body surface area A.

5. A fish size measurement system based on 3D visual reconstruction according to claim 4, characterized in that: The data processing module also includes a data integration processing unit and a data storage unit. The data integration processing unit is used to summarize the obtained visual fish body size and acoustic fish body size respectively, and output the sonic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis, wherein the sonic fish body three-dimensional vector Xsonic = (Ls, Ws, Hs), and the visual fish body three-dimensional vector Xvis = (Lv, Wv, Hv); The data storage unit constructs a fish body database to automatically encode each fish, and associates and stores the corresponding encoded sonic fish body three-dimensional vector Xsonic and visual fish body three-dimensional vector Xvis into the fish body database.

6. A fish size measurement system based on 3D visual reconstruction according to claim 5, characterized in that: The three-dimensional fusion module constructs a three-dimensional fusion model, extracts the sonic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis, inputs them into the three-dimensional fusion model, adjusts the fusion size of the sonic fish body three-dimensional vector Xsonic and the visual fish body three-dimensional vector Xvis by introducing a fusion coefficient a, and outputs the fused fish body size Xfusion; where the fused fish body size Xfusion represents the fused fish body length Lf, the fused fish body width Wf, and the fused fish body height Hf.

7. A fish size measurement system based on 3D visual reconstruction according to claim 6, characterized in that: The fish body size analysis module includes a reverse estimation unit and a size analysis unit; The reverse calculation unit uses the reverse calculation formula to reversely calculate the fish body size (L, W, H) based on the fusion fish body size Xfusion obtained after fusion. The reverse calculation formula uses a linear regression algorithm for reverse calculation optimization. The calculation formula of the fish body size (L, W, H) is: Where L represents the length of the fish body, W represents the width of the fish body, and H represents the height of the fish body. k1, k2, and k3 represent the regression coefficients of the fused fish body length Lf, fused fish body width Wf, and fused fish body height Hf in the fused fish body size Xfusion, respectively. b1, b2, and b3 represent the bias terms of the fused fish body length Lf, fused fish body width Wf, and fused fish body height Hf in the fused fish body size Xfusion, respectively, which are used as correction terms to adjust the output value. The size analysis unit performs comprehensive calculation by integrating the fish body dimensions (L, W, H) to output the total fish body size Sfinal. The calculation formula for the total fish body size Xfusion is: Xfusion=a·Xvis+(1-a)·Xsonic; Wherein, a represents the fusion coefficient, which controls the relative weight of visual data and acoustic data, Xsonic represents the three-dimensional vector of the sonic fish body, and Xvis represents the three-dimensional vector of the visual fish body Xvis.

8. A fish size measurement system based on 3D visual reconstruction according to claim 1, characterized in that: The error analysis module includes an error calculation unit and an error evaluation unit; The error calculation unit is used to calculate the difference between the actual measured fish body size and the obtained total fish body size Sfinal, and output a deviation value Sreal; The error evaluation unit is used to perform error evaluation based on the output result of the deviation value Sreal according to the allowable deviation range set by the user, analyze the detection accuracy of the fish body measurement system, and trigger the error optimization mechanism according to the error evaluation result.

9. A fish size measurement system based on 3D visual reconstruction according to claim 8, characterized in that: The error evaluation result triggers the error optimization mechanism, and the specific evaluation content is as follows; When -0.5≤deviation value Sreal≤0.5, it indicates that the measurement accuracy of the fish body measurement system is normal, and the test ends, and the current fish body measurement system is executed; When the deviation value Sreal>0.5, it indicates that the measurement accuracy of the fish body measurement system is positively abnormal, and the error optimization mechanism is triggered; When the deviation value Sreal≤-0.5, it indicates that the measurement accuracy of the fish body measurement system is reversely abnormal, and the error optimization mechanism is triggered.

10. A fish size measurement system based on 3D visual reconstruction according to claim 9, characterized in that: The error analysis module also includes an error optimization unit. When the error optimization mechanism is triggered, the error optimization unit sets the output result of the deviation value Sreal as the objective function, derives the objective function with respect to the fusion coefficient a, obtains the gradient of the error function, and continuously updates the fusion coefficient a using the gradient descent formula until the error function converges and stops updating. The updated fusion coefficient a is substituted into the three-dimensional fusion model for iterative analysis until the measurement accuracy is normal and the iteration and update are stopped.

Images