Living body culture lateolabrax japonicus muscle quality evaluation method and device based on hyperspectral image and medium
By using a portable land-based hyperspectral imaging system and machine learning models, the problem of non-destructive high-throughput detection of the muscle texture characteristics of sea bass has been solved, enabling rapid and accurate determination of the muscle texture characteristics of live sea bass, which is applicable to different aquaculture regions and methods.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- YELLOW SEA FISHERIES RES INST CHINESE ACAD OF FISHERIES SCI
- Filing Date
- 2026-01-21
- Publication Date
- 2026-05-05
AI Technical Summary
Current technologies lack non-destructive, high-throughput methods for real-time detection of the texture characteristics of sea bass muscle. Traditional methods are time-consuming, labor-intensive, and affect the samples, failing to meet the needs of the rapidly developing live sea bass farming industry.
A portable land-based hyperspectral imaging system was used to acquire hyperspectral images of the surface of live sea bass. A muscle quality evaluation model was constructed by combining machine learning methods to achieve non-destructive, rapid and accurate determination of muscle texture parameters.
It enables rapid, accurate, and non-destructive determination of the muscle texture characteristics of live sea bass, reducing working time and costs, and is applicable to different farming regions and methods, thus improving testing efficiency.
Smart Images
Figure CN121978028A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of marine aquaculture technology, and in particular relates to a method, equipment and medium for evaluating the muscle quality of live cultured sea bass based on hyperspectral images. Background Technology
[0002] Sea bass is an important marine aquaculture fish in my country, accounting for approximately 12% of the country's annual marine fish production, and thus possesses significant economic value. With the increase in sea bass farming output, consumers are paying increasing attention to its muscle quality. Muscle texture characteristics, encompassing the structural, mechanical, and surface properties of muscle and muscle products perceived during human chewing, are considered sensory factors directly influencing consumer satisfaction. Muscle texture characteristics are characterized by parameters such as hardness, elasticity, adhesion, cohesiveness, gelatinization, chewiness, and resilience. However, current methods for evaluating muscle texture characteristics primarily rely on texture analyzers for quantitative description. This method requires complex sample pretreatment, which is not only labor-intensive and time-consuming but also destructive to the samples, hindering real-time sample detection and impacting production. Furthermore, the heterogeneity of fish muscle introduces representativeness errors into traditional measurement methods. To date, the evaluation of sea bass muscle texture characteristics still lacks non-destructive, high-throughput digital methods, severely restricting the healthy and sustainable development of the sea bass aquaculture industry. Therefore, the non-destructive, high-throughput evaluation of the muscle quality of farmed sea bass using hyperspectral imaging technology is a key technology that urgently needs to be solved.
[0003] Hyperspectral imaging technology can simultaneously capture high-resolution spatial and spectral data, offering advantages such as speed, accuracy, non-destructive processing, and continuous monitoring. It has been widely applied in areas such as evaluating the authenticity, color, flavor, freshness, nutritional parameters, and microbial contamination of meat products. However, most applications focus on processed muscle samples, and its application in evaluating the quality of live marine-cultured fish muscle has not yet been reported.
[0004] In particular, spotted sea bass are farmed in all coastal provinces of my country, using methods including pond culture, nearshore cage culture, and recirculating aquaculture systems. Different farming regions and methods result in different muscle texture characteristics of spotted sea bass. Currently, there is a lack of inversion models for the muscle texture characteristics of live spotted sea bass applicable to the major spotted sea bass farming areas across the country. Summary of the Invention
[0005] The purpose of this invention is to provide a method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images. Therefore, for live cultured sea bass, a rapid, accurate, non-destructive, and high-throughput method for determining muscle texture parameters is constructed, which is the foundation for evaluating the muscle quality and screening germplasm resources of cultured sea bass, and is both challenging and innovative.
[0006] To achieve the above objectives, this invention provides a method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images, comprising: Acquire hyperspectral images of the body surface of the target live spotted bass; The hyperspectral image data of the target live sea bass body surface is input into a pre-constructed sea bass muscle quality evaluation model to obtain the target fish muscle texture parameters and the digital distribution map of the target fish muscle texture parameters.
[0007] Furthermore, the method for determining the evaluation model for the muscle quality of farmed sea bass is as follows: By collecting hyperspectral image data of live spotted bass from different spotted bass farming areas, and simultaneously measuring the muscle texture parameters of corresponding parts of farmed spotted bass; Based on the hyperspectral characteristic spectral information of the live cultured sea bass, the characteristic bands of the hyperspectral image data of the target live sea bass are extracted, and the characteristic bands include spectral information in the spectral range of 400-1000 nm. The characteristic bands are preprocessed to eliminate the influence of environmental noise. Using machine learning methods, an evaluation model for the quality of farmed sea bass muscle is constructed by combining the preprocessed hyperspectral images and measured muscle texture parameter data.
[0008] Optionally, hyperspectral image data of the lateral surface of live cultured sea bass can be acquired at a distance of 30-60 cm.
[0009] Optionally, the simultaneous determination of the textural properties of corresponding muscle parts in farmed sea bass using a texture analyzer specifically includes: Acquire hyperspectral data of one side of the muscle in farmed sea bass; The muscle obtained above is then removed, along with the surface skin and fascia. The muscles with the surface skin and fascia removed were cut into multiple small muscle cubes; The above-mentioned multiple muscle cubes were placed on the texture analyzer platform, and the texture characteristics parameters of the corresponding parts of the farmed sea bass were measured using the full texture analysis mode, including hardness, elasticity, adhesion, cohesiveness, adhesiveness, chewiness and resilience. The average value of the texture parameters collected from the above multiple muscle cubes is used to represent the muscle texture parameters of the corresponding target cultured sea bass.
[0010] Optionally, the step of extracting the feature bands of the hyperspectral image data of the target live spotted sea bass based on the hyperspectral feature spectral information of the live sea bass body surface specifically includes: Acquire the image data of the reflective reference plate; The hyperspectral image data of the live cultured sea bass was calibrated using the image data of the reflective reference plate, and the calibrated hyperspectral image data of the cultured sea bass was preprocessed to obtain preprocessed hyperspectral information. Based on the preprocessed hyperspectral information, the hyperspectral characteristic spectral information of the surface of farmed sea bass was obtained.
[0011] Optionally, the preprocessing method includes, but is not limited to, one or more combinations of Savitzky-Golay smoothing, standard normal transformation, detrending, multivariate scattering correction, first-order reciprocal processing, and second-order derivative processing.
[0012] Optionally, the machine learning method includes, but is not limited to, one or more of partial least squares regression, support vector machine regression, random forest regression, convolutional neural network, and backpropagation neural network.
[0013] Optionally, acquiring hyperspectral images of the target live farmed sea bass specifically includes: acquiring hyperspectral image data of the target sea bass's body surface using a portable land-based hyperspectral imaging system.
[0014] As a preferred embodiment, the spectral preprocessing for hardness testing uses the first derivative, and the machine learning uses a convolutional neural network. For adhesion detection, spectral preprocessing employs detrending, and random forest is used for machine learning. The spectral preprocessing for chewiness detection employed second-order derivative processing, and the machine learning used support vector machines. For adhesiveness detection, spectral preprocessing was performed using first-order derivatives, and random forest was used for machine learning. The spectral preprocessing for elasticity detection uses second-order derivative processing, and the machine learning uses a backpropagation neural network. The spectral preprocessing for cohesiveness detection employed Savitzky-Golay smoothing, and the machine learning used a backpropagation neural network. For reproducible detection, standard normal transformation was used for spectral preprocessing, and convolutional neural networks were used for machine learning.
[0015] The present invention also provides a computer device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images.
[0016] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images.
[0017] The beneficial effects of this invention compared to the prior art are as follows: Compared with traditional methods for determining muscle texture properties, this invention first achieves rapid and non-destructive determination of live farmed sea bass samples. In particular, by utilizing a portable land-based hyperspectral imaging system, it reduces working time and costs, allows for a wider range of applicable conditions, and makes it easier to construct estimation models, thus achieving efficient determination of the muscle texture parameters of target live farmed sea bass. Attached Figure Description
[0018] Figure 1 This is a schematic diagram of the process for evaluating the muscle quality of live cultured sea bass based on hyperspectral images, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of a portable land-based hyperspectral imaging system provided in an embodiment of the present invention; 1. Hyperspectral imager, 2. Halogen lamp, 3. Black Teflon plate, 4. Black light-absorbing flocked cloth, 5. Darkroom; Figure 3 This is a digital distribution map of the muscle texture parameters of farmed sea bass predicted using an estimation model, provided in an embodiment of the present invention. Figure 4 is a graph showing the relationship between the model-predicted muscle texture parameters of farmed sea bass and the measured muscle texture parameters provided in the embodiments of the present invention. Detailed Implementation
[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0021] This invention provides a method for estimating the textural properties of live farmed sea bass muscle using a portable land-based hyperspectral imaging system, particularly a method for evaluating the quality of live farmed sea bass muscle based on spectral images. Figure 1 As shown, it includes the following steps: Step 1: Obtain hyperspectral images of the body surface of the target live spotted bass; Step 2: Input the hyperspectral image data of the target live sea bass body surface into the pre-constructed sea bass muscle quality evaluation model to obtain the target fish muscle texture characteristics parameters and the digital distribution map of the target fish muscle texture characteristics parameters.
[0022] The method for determining the quality evaluation model for farmed spotted sea bass muscle is as follows: (1) Use a portable land-based hyperspectral imaging system to collect hyperspectral image data of live spotted bass from major aquaculture areas in my country (Guangdong, Fujian and Shandong, etc.), and use a texture analyzer to simultaneously measure the muscle texture parameters of corresponding parts of farmed spotted bass; Since spotted sea bass are farmed in all coastal provinces of my country, using methods such as pond farming, nearshore cage farming, and recirculating aquaculture systems, it is impossible to acquire hyperspectral image data of live spotted sea bass under different farming regions and methods using a fixed land-based hyperspectral imager. Therefore, this embodiment of the invention first constructs a portable land-based hyperspectral imaging system. Live farmed spotted sea bass from major sea bass farming areas in my country (Guangdong, Fujian, and Shandong, etc.) are anesthetized and placed in the portable land-based hyperspectral imaging system with their sides facing upwards. The hyperspectral imager is positioned directly above the live farmed spotted sea bass to acquire hyperspectral image data of their body surface.
[0023] An example is: Figure 2 As shown, the portable land-based hyperspectral imaging system includes a high-resolution land-based hyperspectral imager 1 (SpecimIQ sensor), two 150 W halogen lamps 2 and lamp stands, a 100 cm × 100 cm black light-absorbing flocked cloth 4 for imaging, and a 60 cm × 60 cm black Teflon plate 3. Anesthetized live farmed sea bass are placed sideways on the black Teflon plate, and the land-based hyperspectral imager is positioned 45 cm directly above the sideways of the sea bass. The two halogen lamps are positioned on either side of the fish's body axis, illuminating the sample area at a 45° angle. The hyperspectral image data of the live farmed sea bass's body surface is acquired in a darkroom 5 to ensure consistent illumination conditions for all samples.
[0024] (2) Using the ENVI image analysis software, based on the hyperspectral characteristic spectral information of the cultured sea bass body surface, the characteristic bands of the hyperspectral image data of the live cultured sea bass body surface are extracted. The characteristic bands include spectral information in the spectral range of 400-1000 nm. The specific operation is as follows: 1) Obtain the image data of the reflective reference plate.
[0025] During the acquisition of hyperspectral image data, a reflection reference plate is set up so that both the target area and the reflection reference plate are placed within the viewfinder of the hyperspectral imager during shooting, thereby acquiring reflection reference plate image data used for reflectivity calibration of the hyperspectral image data.
[0026] 2) The hyperspectral image data of the live cultured sea bass is calibrated using the image data of the reflective reference plate, and the calibrated hyperspectral image data of the live cultured sea bass is preprocessed to obtain preprocessed hyperspectral information.
[0027] One example is as follows: First, the hyperspectral image data is calibrated using reflectance reference plate image data. Then, the calibrated hyperspectral image data is used in ENVI 5.3 software to extract hyperspectral information from the surface of live farmed sea bass. Next, the extracted hyperspectral information from the surface of farmed sea bass is preprocessed in The Unscrambler X software. This preprocessing method includes one or more combinations of Savitzky-Golay smoothing, standard normal transformation, multivariate scattering correction, detrending, first derivative processing, and second derivative processing.
[0028] Smooth the hyperspectral information of farmed sea bass and reduce noise interference caused by different data acquisition environments.
[0029] (3) Using Python 3.9 (including NumPy, Pandas, PyTorch and Scikit-learn libraries) combined with preprocessed hyperspectral image data of farmed sea bass and measured muscle texture parameters of sea bass, a live farmed sea bass muscle quality evaluation model was constructed.
[0030] One example is: using preprocessed hyperspectral image data and measured muscle texture parameters of sea bass in Python 3.9 to construct a live cultured sea bass muscle quality evaluation model, which includes one or more of the following: partial least squares regression, support vector machine regression, random forest regression, convolutional neural network, and backpropagation neural network.
[0031] In this embodiment of the invention, step 1 specifically includes: Using a portable land-based hyperspectral imaging system, hyperspectral image data of the target sea bass body surface were acquired according to the procedure described in (1).
[0032] An example: Take a live farmed sea bass to be tested, anesthetize it using MS-222, place it with its right side facing up in a dark room on a portable hyperspectral imaging system, and set the land-based hyperspectral imager 45cm directly above the right side of the farmed sea bass to acquire hyperspectral image data of the live farmed sea bass body surface to be tested.
[0033] In this embodiment of the invention, step 2 specifically includes: 1) Obtain hyperspectral imager reflectance calibration data, specifically: During the acquisition of hyperspectral image data, a reflection reference plate is set up so that both the target area and the reflection reference plate are placed within the viewfinder of the hyperspectral imager during shooting, thereby acquiring reflection reference plate image data used for reflectivity calibration of the hyperspectral image data.
[0034] 2) Preprocessing of hyperspectral data from the surface of the spotted sea bass, specifically: The hyperspectral image data of the live cultured sea bass was calibrated using ENVI 5.3 software and reflectance reference plate image data. Then, the extracted hyperspectral information of the cultured sea bass was preprocessed in The Unscrambler X software to obtain preprocessed hyperspectral information.
[0035] 3) Input the preprocessed hyperspectral image data of the cultured sea bass body surface into the cultured sea bass muscle quality evaluation model to obtain the digital distribution map of the textural characteristic parameters of the live cultured sea bass muscle to be measured, specifically: By inputting the preprocessed hyperspectral image data into the cultured sea bass muscle quality evaluation model built in Python 3.9, the predicted values of the textural properties parameters of the live cultured sea bass muscle to be tested can be calculated.
[0036] In this embodiment of the invention, the method further includes: obtaining a digital distribution map of the textural properties parameters of the live cultured sea bass muscle based on the inversion spectral image data output by the constructed cultured sea bass muscle quality evaluation model.
[0037] Example 1: Visual inversion of muscle texture parameters based on hyperspectral data of live farmed sea bass in Doumen District, Zhuhai City: 1) Obtain hyperspectral images of the surface of live cultured sea bass: This example is designed as follows: Figure 2 The portable land-based hyperspectral imaging system shown was specifically deployed at a sea bass farming base in Doumen District, Zhuhai City. Doumen District is a key production area for pond-farmed sea bass in my country, with its annual production of "Baijiao sea bass" accounting for more than half of the national total. First, a portable land-based hyperspectral imaging system was constructed. Live sea bass farmed in Doumen District, Zhuhai City, were anesthetized with MS-222 and placed in the portable land-based hyperspectral imaging system in a darkroom to acquire hyperspectral images of the sea bass's body surface and reflectance reference plate.
[0038] 2) Preprocessing of hyperspectral image data: The acquired high-resolution hyperspectral images were applied to ENVI software to extract hyperspectral information, and hyperspectral information preprocessing was performed in TheUnscrambler X software to smooth the spectral information and reduce noise interference caused by different data acquisition environments on the hyperspectral information of the cultured sea bass.
[0039] 3) The preprocessed hyperspectral image data and measured muscle texture parameters of sea bass were used in Python 3.9 to construct a live cultured sea bass muscle quality evaluation model, and the prediction accuracy of different preprocessing methods and models for the muscle texture parameters of cultured sea bass was evaluated. The optimal prediction model for the muscle texture parameters of cultured sea bass is shown in Table 1.
[0040] Table 1. Optimal Prediction Model for Texture Parameters of Cultured Spotted Sea Bass Muscle .
[0041] 4) Based on the optimal model of the textural properties of cultured sea bass muscle, obtain digital images of the textural properties of cultured sea bass muscle: Digital images of the texture parameters of cultured sea bass muscle obtained by inversion are shown below. Figure 3 The corresponding digital data of the visualization map can be used to extract muscle texture characteristic parameter data for any target range or point, so as to obtain a digital visual distribution map of muscle texture characteristic parameters that is superior to traditional measurement methods in a non-destructive and rapid manner.
[0042] Example 2: Visual inversion of muscle texture parameters based on hyperspectral data of live farmed perch in Fuding City, Ningde: 1) Obtain hyperspectral images of the surface of live cultured sea bass: This example is designed as follows: Figure 2 The portable land-based hyperspectral imaging system shown was specifically deployed at the sea bass farming base of Minwei Industrial Group in Fuding City, Ningde City. Minwei Industrial Group is a key sea bass farming area in my country's nearshore cage culture, and its "Tongjiang Sea Bass" enjoys a high reputation nationwide. First, a portable land-based hyperspectral imaging system was constructed. Live sea bass farmed by Minwei Industrial Group were anesthetized with eugenol and placed in the portable land-based hyperspectral imaging system in a darkroom to acquire hyperspectral images of the surface of live farmed sea bass in Fuding City, Ningde City, as well as images of the reflectance reference plate.
[0043] 2) Preprocessing of hyperspectral image data: The acquired high-resolution hyperspectral images were applied to ENVI software to extract hyperspectral information, and hyperspectral information preprocessing was performed in TheUnscrambler X software to smooth the spectral information and reduce noise interference caused by different data acquisition environments on the hyperspectral information of the cultured sea bass.
[0044] 3) Input the preprocessed hyperspectral image data and measured texture parameters of sea bass muscle into the pre-built quality evaluation model of farmed sea bass muscle in Python 3.9.
[0045] 4) Based on the optimal model of the texture parameters of cultured sea bass muscle, a digital image of the texture parameters of cultured sea bass muscle is obtained.
[0046] 5) Comparison of measured values and estimated textural properties of sea bass muscle using spectral models: As shown in Figure 4, the regression determination coefficient R0 between the estimated and measured values of hardness, chewiness, adhesiveness, cohesion, and elasticity obtained from the fishery sea bass muscle quality evaluation model is shown in Figure 4. 2 A coefficient of determination greater than 0.85, a mean percentage error (MAPE) less than 10%, and a model determination coefficient greater than 2.5 indicate that the model established by the method of this invention is practical and can be used to accurately determine the textural properties of cultured sea bass muscle.
Claims
1. A method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images, characterized in that, The method includes: Acquire hyperspectral images of the body surface of the target live spotted bass; The hyperspectral image data of the target live sea bass body surface is input into a pre-constructed sea bass muscle quality evaluation model to obtain the target fish muscle texture parameters and the digital distribution map of the target fish muscle texture parameters.
2. The method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images according to claim 1, characterized in that, The method for determining the quality evaluation model for farmed sea bass muscle is as follows: By collecting hyperspectral image data of live spotted bass from different spotted bass farming areas, and simultaneously measuring the muscle texture parameters of corresponding parts of farmed spotted bass; Based on the hyperspectral characteristic spectral information of the live cultured sea bass, the characteristic bands of the hyperspectral image data of the target live sea bass are extracted, and the characteristic bands include spectral information in the spectral range of 400-1000 nm. The characteristic bands are preprocessed to eliminate the influence of environmental noise. Using machine learning methods, an evaluation model for the quality of farmed sea bass muscle is constructed by combining the preprocessed hyperspectral images and measured muscle texture parameter data.
3. The method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images according to claim 2, characterized in that, Hyperspectral image data of the lateral surface of live farmed sea bass were collected at a distance of 30-60 cm.
4. The method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images according to claim 3, characterized in that, The method of simultaneously measuring the textural properties of corresponding muscle parts in farmed sea bass using a texture analyzer specifically includes: Acquire hyperspectral data of one side of the muscle in farmed sea bass; The muscle obtained above is then removed, along with the surface skin and fascia. The muscles with the surface skin and fascia removed were cut into multiple small muscle cubes; The multiple muscle cubes were placed on the texture analyzer platform, and the texture characteristics parameters of the corresponding parts of the farmed sea bass were measured using the full texture analysis mode, including hardness, elasticity, adhesion, cohesiveness, adhesiveness, chewiness, and resilience. The average value of the texture parameters collected from the multiple muscle cubes is taken to represent the muscle texture parameters of the corresponding target cultured sea bass.
5. The method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images according to claim 4, characterized in that, The extraction of feature bands from the hyperspectral image data of the target live spotted sea bass based on the hyperspectral feature spectral information of the live sea bass body surface specifically includes: Acquire the image data of the reflective reference plate; The hyperspectral image data of the live cultured sea bass was calibrated using the image data of the reflective reference plate, and the calibrated hyperspectral image data of the cultured sea bass was preprocessed to obtain preprocessed hyperspectral information. Based on the preprocessed hyperspectral information, the hyperspectral characteristic spectral information of the surface of farmed sea bass was obtained.
6. The method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images according to claim 5, characterized in that, The preprocessing methods include, but are not limited to, one or more combinations of Savitzky-Golay smoothing, standard normal transformation, detrending, multivariate scattering correction, first-order reciprocal processing, and second-order derivative processing.
7. The method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images according to claim 6, characterized in that, The machine learning methods include, but are not limited to, one or more of partial least squares regression, support vector machine regression, random forest regression, convolutional neural networks, and backpropagation neural networks.
8. The method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images according to claim 2, characterized in that, The acquisition of hyperspectral images of the target live farmed sea bass specifically includes: acquiring hyperspectral image data of the target sea bass's body surface using a portable land-based hyperspectral imaging system.
9. A computer device, characterized in that, The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images as described in any one of claims 1-8.
10. A computer-readable storage medium, characterized in that, The medium stores a computer program, which, when executed by a processor, implements the method for evaluating the muscle quality of live cultured sea bass based on hyperspectral images as described in any one of claims 1-8.