A method for identifying biological fermentation product categories based on image recognition
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 汉中天然谷生物科技股份有限公司
- Filing Date
- 2026-03-26
- Publication Date
- 2026-06-02
Smart Images
Figure CN121904489B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of bio-fermentation detection technology, and in particular to a method for identifying the types of bio-fermentation products based on image recognition. Background Technology
[0002] In biofermentation, the chemical composition, concentration gradient, and intrinsic physical properties of fermentation products are contained within the refractive index distribution field and interfacial adhesion dynamics of the test medium. Existing product identification technologies primarily utilize spectral analysis or traditional image feature matching to identify product types. However, when faced with high-turbidity environments and complex interfacial interactions, existing identification systems exhibit the following failure mechanisms:
[0003] Mechanism of Feature Contribution Weight Mismatch Caused by Scattering Interference: The refractive index gradient of bio-fermentation products is an important spatial feature characterizing the product properties. Existing visual recognition technologies mostly rely on the geometric distortion of the background image to deduce the refractive index distribution. However, as the fermentation process progresses, the density of microbial cells and the content of metabolic byproducts in the test medium increase, inducing severe multiple scattering phenomena. Multiple scattering causes irregular dispersion of the optical path, resulting in a significant decrease in the feature recognition accuracy of the background image. Because existing recognition schemes lack a mechanism to dynamically adjust the spatial feature contribution weights based on the environmental optical quality, they cannot adaptively reduce the recognition weights of unreliable spatial signals when scattering interference intensifies, leading to systematic biases in the recognition conclusions.
[0004] The lack of perception mechanism for interfacial adhesion kinetics: The residual retention behavior of fermentation products on the inner wall of the observation window is closely related to the surface tension, viscosity, and intermolecular forces of the products. Different types of bio-fermentation products exhibit differentiated evolutionary characteristics in their dissipation process on the inner wall after liquid level fluctuations. Traditional identification schemes often focus on extracting transient static visual features, lacking quantitative tracking methods for the evolution of residue coverage over time. Because existing monitoring systems lack necessary constraint mechanisms at the level of interfacial adhesion evolution, and given the limitations of spatial optical features, they cannot provide kinetic verification support based on the intrinsic physical properties of the products for the identification process, leading to unstable identification conclusions under complex flow field conditions.
[0005] Nonlinear mismatch mechanism in heterogeneous feature coupling process: Spatial features formed by optical path deflection and temporal features formed by adhesion attenuation together constitute the product identification space. However, the optical quality fluctuations of the fermentation environment and the evolution of the product's adhesion characteristics exhibit a nonlinear coupling relationship in the spatiotemporal plane. Existing analytical models typically use a fixed scaling factor to simply superimpose heterogeneous features, lacking the ability to adaptively adjust to environmental fluctuations. When the turbidity of the medium changes abruptly, the lack of dynamic correction methods for the mapping relationship between the input signal and the actual physical fingerprint leads to logical deviations between the identification values generated by the identification system and the actual properties of the product.
[0006] Based on the above analysis of physical characteristics, due to the interference of scattering noise on signal weights, the lack of extraction of interface dynamic features, and the linearization defects of heterogeneous feature coupling logic, the existing schemes are unable to form a stable and reproducible recognition closed loop at the engineering implementation level, resulting in the overall failure of the steady-state characteristics of the recognition system under complex working conditions. Summary of the Invention
[0007] This invention provides a method for identifying the types of bio-fermentation products based on image recognition. This method addresses the signal distortion between the spatial visual features acquired by the image sensor and the intrinsic properties of the product caused by multiple scattering path interference generated inside the fermentation broth during high-density fermentation. Furthermore, existing identification schemes lack perception of the adhesion dynamics of the observation window interface, which further results in the loss of feature dimensions of the intrinsic physical properties of the product, making it difficult to maintain stable identification conclusions under complex working conditions.
[0008] In view of the above problems, the present invention provides a method for identifying the types of bio-fermentation products based on image recognition, comprising the following steps:
[0009] Step S1: Obtain a background reference image and extract the reference pixel distribution based on the background reference image;
[0010] Step S2: Acquire a real-time image of the test environment containing the fermentation medium to be tested, and generate a first feature vector field based on the pixel offset of the real-time image relative to the reference pixel distribution;
[0011] Step S3: Obtain the liquid level status information of the environment under test; in response to the liquid level fluctuation characteristics, obtain the time-series image sequence of the inner wall of the preset boundary area in the environment under test; extract the decay rate constant of the residue coverage over time based on the time-series image sequence, and generate a first feature index; at the same time, extract an environmental sampling operator characterizing the transparency of the environment based on the features of the real-time image.
[0012] Step S4: Using the first feature vector field, the first feature index, and the environmental sampling operator as input variables, perform weighted coupling calculations to generate a species identification comprehensive index, and determine the type of fermentation product to be tested based on the species identification comprehensive index.
[0013] Furthermore, the background reference image has a periodic geometric texture for tracing optical path deflection.
[0014] Further, step S2 specifically includes: using a background texture offset algorithm to convert the propagation path deflection of light when it penetrates the fermentation medium to be tested into the first feature vector field.
[0015] Further, step S3 specifically includes: fitting the residue coverage using a first-order kinetic decay model, and the resulting decay rate constant is the first characteristic index.
[0016] Furthermore, the weighted coupling calculation in step S4 includes the following logic:
[0017] The contribution weight of the first feature vector field is suppressed according to the environmental sampling operator, wherein the contribution weight decreases as the environmental sampling operator increases;
[0018] The first feature index is subjected to saturation mapping to keep its feature contribution within a preset range.
[0019] The first feature vector field after suppression processing is linearly superimposed with the first feature index after saturation mapping processing to generate the species identification comprehensive index.
[0020] Furthermore, the species identification composite index is calculated according to the following formula. :
[0021] in:
[0022] This is the global modulus statistic of the first feature vector field; This is a preset space reference value; For the environmental sampling operator; The first threshold; The first parameter; The first feature index; The reference decay rate; This is an affinity correction coefficient based on the interface material preset; and Preset weights.
[0023] Furthermore, environmental sampling operators The calculation is based on the normalized entropy of the grayscale information in the real-time image.
[0024] Furthermore, the reference decay rate It is a preset positive frequency constant.
[0025] This invention also provides a bio-fermentation product type identification system based on image recognition, comprising:
[0026] Reference unit, used to provide a background reference image with periodic geometric texture;
[0027] The image acquisition unit is used to acquire real-time images and time-series images of the fermentation medium to be tested, as well as the inner wall, within a field of view corresponding to the background reference image.
[0028] The processing unit is used to execute the above-described method.
[0029] The technical solution provided in this application has at least the following technical effects:
[0030] By performing nonlinear modulation on the contribution weight of the first feature vector field through the environmental sampling operator, the gain of the spatial visual feature can be adaptively adjusted according to the numerical change of the environmental sampling operator. This forms a dynamic shielding of the pixel displacement signal under the condition of enhanced multiple scattering interference, ensuring that the recognition system can still lock the effective refractive index distortion range based on the nonlinear constraint of the first threshold in the low contrast image environment.
[0031] By establishing an exponential regression relationship between the residue coverage decay process and the first feature index, the identification logic evolves from static image feature matching to rate constant determination based on interface adhesion dynamics. This provides a discrimination basis with physical time axis constraints when spatial optical signals are limited, ensuring that the species identification comprehensive index can reflect the intrinsic interface adhesion physical properties of fermentation products.
[0032] By performing saturation mapping on the first feature index and weight suppression on the first feature vector field, nonlinear compensation for environmental fluctuations and differences in material affinity is achieved, enabling the species identification comprehensive index to form a numerical gradient in different fermentation stages and species fingerprint databases. This improves the discrimination resolution of the identification system under complex working conditions and reduces the risk of logical misjudgment caused by data fluctuations.
[0033] The synergistic achievement of the above-mentioned technical effects solves the failure problem of existing recognition systems when there are sudden changes in the optical environment and the lack of adhesion feature perception. It unifies the spatial refractive index gradient characterization and the temporal interface dynamics characterization within a nonlinear coupling framework based on environmental transparency feedback, and finally forms a stable and reproducible recognition closed loop at the level of industrial fermentation production. Attached Figure Description
[0034] Figure 1 A flowchart of a method for identifying types of bio-fermentation products based on image recognition, provided in an embodiment of the present invention;
[0035] Figure 2 This is an architecture diagram of a bio-fermentation product type identification system based on image recognition, provided for an embodiment of the present invention. Detailed Implementation
[0036] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.
[0037] For examples, please refer to Figure 1 and Figure 2 This invention provides a method for identifying the types of bio-fermentation products based on image recognition, comprising the following steps:
[0038] Step S1: Obtain a background reference image and extract the reference pixel distribution based on the background reference image;
[0039] Step S2: Acquire a real-time image of the test environment containing the fermentation medium to be tested, and generate a first feature vector field based on the pixel offset of the real-time image relative to the reference pixel distribution;
[0040] Step S3: Obtain the liquid level status information of the environment under test; in response to the liquid level fluctuation characteristics, obtain the time-series image sequence of the inner wall of the preset boundary area in the environment under test; extract the decay rate constant of the residue coverage over time based on the time-series image sequence, and generate a first feature index; at the same time, extract an environmental sampling operator characterizing the transparency of the environment based on the features of the real-time image.
[0041] Step S4: Using the first feature vector field, the first feature index, and the environmental sampling operator as input variables, perform weighted coupling calculations to generate a species identification comprehensive index, and determine the type of fermentation product to be tested based on the species identification comprehensive index.
[0042] The bio-fermentation identification and processing center initiated the hardware environment construction process. The image acquisition unit was installed within the field of view corresponding to the background reference image. The optical axis of the image acquisition unit was perpendicular to the viewing mirror surface of the fermentation observation window. The background reference image was fixed on the side of the fermentation observation window facing away from the image acquisition unit. To construct a stable optical measurement field, a constant-power supplementary light source was arranged around the perimeter of the fermentation observation window. The light emitted by the supplementary light source, after reflection, illuminated the background reference image.
[0043] The background reference image features a periodic geometric texture used to trace the deflection of light paths. This periodic texture employs a non-uniform mesh generated by a fractal algorithm. The topological features of this non-uniform mesh break the translational symmetry of space. When light passes through a medium with a changing refractive index, the amount of deflection of the light path results in pixel displacement through the non-uniform mesh.
[0044] After the hardware environment was built, the bio-fermentation identification and processing center guided the system into the baseline initialization phase. The image acquisition unit captured a static image when the medium inside the fermenter was clear, and used this static image as the background reference image. To eliminate ambient astigmatism and random thermal noise from the sensor in the background reference image, the bio-fermentation identification and processing center performed median filtering on the background reference image. Median filtering used a 5×5 pixel window. The grayscale values of the pixels within the 5×5 pixel window were sorted, and the median value after sorting was taken as the new value for the center pixel.
[0045] The background reference image, after median filtering, is transmitted to the bio-fermentation identification and processing center. The center uses an edge detection operator to identify feature points of the periodic geometric texture in the background reference image. The edge detection operator identifies the centroid coordinates or intersection coordinates of the periodic geometric texture. The spatial coordinates of each identified feature point are recorded in a reference matrix in memory, thus generating a pixel-level reference coordinate distribution. This pixel-level reference coordinate distribution defines the initial physical position of each grid point in the background reference image. This pixel-level reference coordinate distribution serves as a zero-point reference for global displacement measurement, used to calibrate the offset vector of the real-time image relative to the background reference image during subsequent identification processes.
[0046] The acquisition frequency of the background reference image is consistent with the sampling period of the fermentation process. The bio-fermentation identification and processing center controls the triggering action of the image acquisition unit through a clock synchronization signal. After the pixel-level reference coordinate distribution is extracted, the baseline initialization data output is stored in a non-volatile storage unit. The pixel-level reference coordinate distribution provides the necessary pixel displacement calculation baseline for the subsequent calculation of the first feature vector field. After confirming that the baseline initialization is complete, the bio-fermentation identification and processing center transfers the logical control of the system to the subsequent spatial feature extraction process.
[0047] The spatial feature mapping stage begins after the baseline initialization stage. The bio-fermentation identification and processing center controls the image acquisition unit to capture real-time images of the test environment containing the fermentation medium to be tested. The real-time images are transmitted to the bio-fermentation identification and processing center for pixel offset calculation. The calculation process is executed through a point-by-point matching algorithm. The point-by-point matching algorithm compares the non-uniform grid feature points in the real-time image with the reference coordinate distribution in the memory baseline matrix. The two-dimensional coordinates of each non-uniform grid feature point in the real-time image are subtracted from the corresponding initial coordinates in the reference coordinate distribution to generate the pixel offset vector of each feature point. When light penetrates the fermentation medium to be tested, the propagation path of the light is deflected due to the uneven refractive index distribution caused by the product concentration gradient within the fermentation medium. The propagation path deflection is converted into a pixel offset on the image plane. The physical mapping relationship between the pixel offset and the propagation path deflection constitutes the data basis for generating the first feature vector field.
[0048] After the pixel offset vectors of the entire field are generated, the bio-fermentation identification and processing center initiates the aggregation calculation of the first feature vector field. The aggregation calculation is performed by taking the modulus of the pixel offset vectors of all feature points within the entire field and calculating the arithmetic mean. This arithmetic mean is defined as the global modulus statistic of the first feature vector field. The global modulus statistic of the first feature vector field quantifies the intensity of the refractive index field perturbation caused by biological metabolic activities in the fermentation medium under test. The global modulus statistic of the first feature vector field, as a spatial dimension feature representation of the first feature vector field, is stored in a temporary variable storage area, awaiting subsequent feature coupling calculations.
[0049] The temporal feature quantification stage is initiated in response to environmental liquid level fluctuation signals. The bio-fermentation identification and processing center receives a trigger signal from the liquid level monitoring unit. When the fermentation liquid level drops or fluctuates drastically, the bio-fermentation identification and processing center controls the image acquisition unit to capture a time-series image sequence on the inner wall of the pre-defined boundary area in the test environment. The inner wall of the pre-defined boundary area is located in the dynamic wetting line region of the fermentation observation window. The captured time-series image sequence records the retention state of fermentation product residues on the inner wall at fixed time intervals. The time-series image sequences are numbered sequentially along the time axis and stored in a cache.
[0050] After the time-series image sequence is acquired, the bio-fermentation identification and processing center performs real-time measurement of residue coverage. The center converts each frame of the time-series image sequence to grayscale and uses a dynamic threshold segmentation algorithm to extract the set of pixels covered by residue. The total number of pixels covered by residue in each frame is divided by the total number of pixels in a preset boundary region to obtain a residue coverage data sequence evolving over time. This residue coverage data sequence is then input into a first-order kinetic decay model. The first-order kinetic decay model calculates the rate constant of residue disappearance using log-linear regression. The generated decay rate constant is defined as the first characteristic index. The first characteristic index characterizes the kinetic adhesion performance between the tested fermentation product and the fermentation observation window interface.
[0051] Simultaneously with generating the first feature index, the bio-fermentation identification and processing center calculates the environmental sampling operator based on the global grayscale distribution of the real-time image. The calculation process is performed by calculating the grayscale entropy of the real-time image. Higher grayscale entropy indicates more severe environmental scattering interference and lower texture contrast in the real-time image. Lower grayscale entropy indicates better medium transparency and richer texture details in the real-time image. The calculated normalized entropy value is defined as the environmental sampling operator. The environmental sampling operator, the first feature vector field, and the first feature index together constitute the complete input variable set for species identification, marking the end of the feature extraction process.
[0052] The feature coupling calculation stage begins after the environmental sampling operator, the first feature vector field, and the first feature index are generated. The bio-fermentation identification and processing center inputs the first feature vector field, the first feature index, and the environmental sampling operator into the calculation logic of the species identification comprehensive index. Species identification comprehensive index Generate using the following formula:
[0053] In the formula, Representative species identification composite index. The global modulus statistic representing the first eigenvector field is used to quantify the total intensity of the refractive index gradient perturbation in the spatial domain. This represents a preset spatial reference value, used to normalize the global modulus statistics. This represents the environmental sampling operator. This represents the first threshold, serving as the critical boundary for determining whether optical interference has occurred. This represents the first parameter, used to adjust the smoothness of the weight switching process. This represents the first characteristic index, which is the decay rate constant extracted through first-order kinetic fitting. Representing the reference decay rate, it is a preset positive frequency constant whose dimensions are consistent with the first characteristic exponent, ensuring that the independent variable of the hyperbolic tangent function is a dimensionless value. This represents the affinity correction coefficient based on the interface material, used to compensate for the influence of different observation window materials on the adhesion characteristics of residues. and This represents the preset weight.
[0054] The left-hand side of the formula performs suppression processing on the first feature vector field. When the environmental sampling operator is below the first threshold, the exponent term in the denominator is small, and the denominator approaches one, so the contribution weight of the first feature vector field to the species identification composite index is normally preserved. When the environmental sampling operator increases and exceeds the first threshold, the exponent term in the denominator increases rapidly, and the contribution weight of the first feature vector field to the species identification composite index is nonlinearly suppressed. The nonlinear suppression logic automatically masks spatial visual features under low-quality optical environments.
[0055] The right-hand side of the formula performs a saturation mapping on the first feature index. The hyperbolic tangent function is used to constrain the numerical fluctuations of the first feature index within a preset stable range, and an affinity correction coefficient is combined to quantify the interface dynamics. The calculated species identification composite index is compared by the bio-fermentation identification processing center with feature thresholds in a preset fingerprint database. By calculating the deviation distance between the species identification composite index and the fingerprints of each preset species in the fingerprint database, the bio-fermentation identification processing center ultimately determines the specific type of the fermentation product to be tested. An arbitration process based on adaptive adjustment of physical environment conditions ensures the objectivity of the identification conclusions under complex working conditions.
[0056] Numerical boundary verification and parameter calibration logic is executed by the bio-fermentation identification and processing center. Preset weights. , and the first threshold The value range is calibrated through a pre-set standard sample comparison experiment. In the standard sample comparison experiment, known species of bio-fermentation products are placed in the test environment. By artificially intervening to change the turbidity of the medium and the interface adhesion state, the bio-fermentation identification and processing center records multiple sets of experimental data corresponding to different working conditions. The calibration is completed by fitting and matching the experimental data with an actual species fingerprint database. By adjusting the parameter values, the discrimination accuracy of the species identification comprehensive index on the known sample set reaches the preset target value, thereby determining the benchmark values of the weighting factor and the logical threshold. This calibration process ensures the matching degree between the algorithm logic and the actual physical scenario.
[0057] The computational stability under extreme sampling environments is maintained through parameter coordination within the formula. When the environmental sampling operator... Approaching the first threshold When, the first parameter The transition slope of the nonlinear suppression process is controlled by a preset non-zero first parameter. The species identification composite index exhibits a smooth transition rather than a step-like abrupt change in value at the critical point. This smooth transition characteristic ensures that the computational logic will not experience numerical oscillations or logical deadlocks during moments of drastic fluctuations in environmental transparency. (Baseline decay rate) As a positive frequency constant, it ensures that the calculation term of the first characteristic exponent is always within the valid domain when participating in the hyperbolic tangent operation, thus guaranteeing the continuity of the identification conclusion under extreme physical conditions.
[0058] For the spatial feature extraction process, in addition to using background texture offset, interference fringe analysis can also be used. During interference fringe analysis, a coherent beam penetrates the fermentation medium under test, and the phase difference caused by the product concentration gradient within the medium is converted into displacement data of the interference fringes. By extracting the displacement vector of the interference fringes, the bio-fermentation identification and processing center can also construct the first feature vector field. Furthermore, extracting spatial features using structured light coding is also a feasible approach. In the structured light coding embodiment, a pre-set geometric grating is projected onto the fermentation observation window, and the image acquisition unit acquires the grating image, which is geometrically distorted due to the refractive index field disturbance of the fermentation medium under test. The bio-fermentation identification and processing center calculates the first feature vector field reflecting the optical path deflection by comparing the distorted grating image with the original grating template. These physical measurement paths all extract spatial features by capturing the modulation effect of medium disturbance on the optical signal, achieving quantization of the refractive index gradient field in its physical essence.
[0059] The calculation logic of the species identification composite index can also be implemented through different combinations of mathematical operators. The suppression logic in the formula can be accomplished by configuring a monotonically decreasing function. In an alternative embodiment, a piecewise linear function is used to adjust the weights according to the numerical range of the environmental sampling operator. When the environmental sampling operator is in the low gray-level information entropy range (good medium transparency), the piecewise linear function assigns a higher slope coefficient to the first feature vector field, making full use of the spatial optical features. When the environmental sampling operator rises to the high gray-level information entropy range (severe scattering interference), the piecewise linear function switches to the low slope range, thereby reducing the contribution ratio of spatial features, consistent with the suppression logic of the main formula. In addition, using an inverse proportional function as the suppression operator can take advantage of the characteristic that as the independent variable increases, the dependent variable decreases, to achieve weight attenuation of the spatial phase distortion vector when the environmental optical quality deteriorates.
[0060] Saturation mapping can be performed using various monotonically increasing functions with upper bounds. Besides the hyperbolic tangent function, a first-order exponential saturation model can also be used to numerically constrain the first feature index. By setting a saturation upper limit, the first-order exponential saturation model ensures that the gain of the first feature index on the species identification composite index gradually levels off as adhesion strength increases. This saturation characteristic avoids overfitting of the overall identification conclusion due to single-point abrupt changes in interface dynamics. By pre-setting these alternative mathematical execution paths in the bio-fermentation identification processing center, the system can adapt to hardware environments with varying computing resources while maintaining consistency in physical judgment logic. These equivalent mathematical processing actions and core formulas share the same causal derivation process in achieving environmentally adaptive weight adjustment and feature strength normalization.
[0061] The implementation of all stages of technology together constitutes a closed-loop identification system. The bio-fermentation identification and processing center achieves automated species identification in industrial fermentation scenarios through the coordinated scheduling of hardware optical paths, initialization benchmarks, spatial and temporal characteristics, and nonlinear coupled logic. Every action in the identification process is based on quantifiable physical data and objective logical transformations. With the output of the species identification comprehensive index, the species information of the fermentation product to be tested is stored in the database, providing data support for subsequent fermentation process optimization.
[0062] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for identifying the types of bio-fermentation products based on image recognition, characterized in that, Includes the following steps: Step S1: Obtain a background reference image and extract the reference pixel distribution based on the background reference image; Step S2: Acquire a real-time image of the test environment containing the fermentation medium to be tested, and generate a first feature vector field based on the pixel offset of the real-time image relative to the reference pixel distribution; Step S3: Obtain the liquid level status information of the environment to be tested; In response to liquid level fluctuation characteristics, a time-series image sequence of the inner wall of a preset boundary area in the environment under test is acquired; based on the time-series image sequence, the decay rate constant of the residue coverage over time is extracted, and a first feature index is generated; simultaneously, an environmental sampling operator characterizing environmental transparency is extracted based on the features of the real-time image. Step S4: Using the first feature vector field, the first feature index, and the environmental sampling operator as input variables, perform weighted coupling calculation to generate a species identification comprehensive index, and determine the type of fermentation product to be tested based on the species identification comprehensive index; Step S3 specifically includes: fitting the residue coverage rate using a first-order kinetic decay model, and the resulting decay rate constant is the first characteristic index; Step S4 calculates the species identification composite index according to the following formula. : in: This is the global modulus statistic of the first feature vector field; This is a preset space reference value; For the environmental sampling operator; The first threshold; The first parameter; The first feature index; The reference decay rate; This is an affinity correction coefficient based on the interface material preset; and Preset weights.
2. The method according to claim 1, characterized in that, The background reference image has a periodic geometric texture for tracking optical path deflection.
3. The method according to claim 1, characterized in that, Step S2 specifically includes: using a background texture offset algorithm to convert the propagation path deflection of light when it penetrates the fermentation medium under test into the first feature vector field.
4. The method according to claim 1, characterized in that, The weighted coupling calculation in step S4 includes the following logic: The contribution weight of the first feature vector field is suppressed according to the environmental sampling operator, wherein the contribution weight decreases as the environmental sampling operator increases; The first feature index is subjected to saturation mapping to keep its feature contribution within a preset range. The first feature vector field after suppression processing is linearly superimposed with the first feature index after saturation mapping processing to generate the species identification comprehensive index.
5. The method according to claim 1, characterized in that, The environmental sampling operator The calculation is based on the normalized entropy of the grayscale information in the real-time image.
6. The method according to claim 1, characterized in that, The reference attenuation rate It is a preset positive frequency constant.
7. A system for identifying the types of bio-fermentation products based on image recognition, characterized in that, include: Reference unit, used to provide a background reference image with periodic geometric texture; The image acquisition unit is used to acquire real-time images and time-series images of the fermentation medium to be tested, as well as the inner wall, within a field of view corresponding to the background reference image. A processing unit for performing the method as described in any one of claims 1 to 6.