Cwhale morphology photogrammetry method based on endogenous biological scaleplate coupling segmented weighted integrated model

Abstract

The invention provides a cetacean morphology photogrammetry method based on an endogenous biological scale coupled segmented weighted integrated model, and relates to the technical field of ecological monitoring. According to the method, the head hole length and the body length pixel length in the cetacean back image are extracted, the ratio of the head hole length to the body length is calculated, the developmental stages are judged according to the ratio, the initial calculation values of the two body lengths are obtained by combining the statistics of the head hole length in the stages and the regression model, and the weighting coefficient is determined based on the error of the standard sample in each developmental stage. And carrying out segmented weighted integration on the two results to obtain a final calibration body length. According to the method, self-adaptive segmented modeling and error weighted fusion with the endogenous biological scale as the core are achieved, high-precision body size measurement can be achieved without external calibration, and a unified, stable and extensible measurement system is provided for whale form quantification and ecological monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ecological monitoring, in particular to a whale morphometric photogrammetry method based on an endogenous biological scale coupled with a segmented weighted integrated model. BACKGROUND

[0002] Nowadays, under the background of ecological protection, as an indicator species of ecological environment, the measurement and acquisition of wild individual body size data of whale animals play a crucial role in population nutritional health status evaluation and species protection. For the work of obtaining morphometric data of wild whales through videogrammetry, a reference object with known length must exist in the picture to achieve it.

[0003] Existing whale morphometric measurement relies on exogenous reference objects (such as ships, laser scales, etc.), which has problems such as great implementation difficulty, many error sources, and great disturbance to animals. Especially for small and sensitive species such as Yangtze finless porpoises, the applicability of exogenous methods is poor. SUMMARY

[0004] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a whale morphometric photogrammetry method based on an endogenous biological scale coupled with a segmented weighted integrated model, which realizes high-precision whale body size measurement without external calibration through the collaborative design of endogenous biological scale, development segmentation modeling and error weighted integration, and provides a unified, stable and sustainable measurement system for morphometric quantification and ecological monitoring.

[0005] To achieve the above purpose, the present application provides the following scheme:

[0006] A whale morphometric photogrammetry method based on an endogenous biological scale coupled with a segmented weighted integrated model, comprising:

[0007] Extracting the pixel length corresponding to the blowhole length and body length from the obtained whale dorsal image, and calculating the proportion of the blowhole length and body length;

[0008] According to the corresponding relationship between the proportion of the blowhole length and body length and the development stage, determining the development stage to which the individual belongs, and determining the matched segmental blowhole length statistic in the development stage;

[0009] Taking the segmental blowhole length statistic corresponding to the development stage as the endogenous biological scale, converting the image pixel scale into the actual scale to obtain a first body length preliminary calculation value;

[0010] Substituting the proportion of the blowhole length and body length into a preset regression model to obtain a second body length preliminary calculation value;

[0011] In the development stage, local errors or confidence of the first body length preliminary calculation value and the second body length preliminary calculation value are evaluated according to a standard sample, and a weighted coefficient of the development stage is determined;

[0012] According to the weighted coefficient of the development stage, the first body length preliminary calculation value and the second body length preliminary calculation value are segmented and weighted integrated to obtain a final calibrated body length.

[0013] Preferably, it further comprises:

[0014] A conversion relationship between pixel length and actual length is established with a pixel length corresponding to the final calibrated body length as the uniform image scale, and scale anchoring of the morphological parameter measurement is completed.

[0015] Preferably, it further comprises:

[0016] The acquired back image is checked for posture flatness, shielding condition, imaging clarity and uniformity of illumination.

[0017] Preferably, the establishment and application process of the corresponding relationship of head hole length and body length proportion-development stage comprises:

[0018] Based on standard samples, threshold values or distribution intervals of the "head hole length and body length proportion" of the development stage are respectively counted to determine the proportion determination boundary of each development stage, and the corresponding relationship of mapping the proportion to the development stage is formed; the development stage includes juvenile, youth and adult stages;

[0019] When processing a new individual, the development stage determination is performed according to the proportion determination boundary in the corresponding relationship; the proportion determination boundary is incrementally updated with the introduction of new standard samples, and the corresponding relationship is synchronously updated.

[0020] Preferably, the in-segment head hole length statistic is at least one of in-segment mean, in-segment robust mean or in-segment quantile.

[0021] Preferably, the preset regression model is at least one of polynomial regression, segmented regression or regression with regularization constraint.

[0022] Preferably, the determination process of the weighted coefficient of the development stage comprises:

[0023] In the development stage, the error indicators of the first body length preliminary calculation value and the second body length preliminary calculation value are calculated respectively by using the measured body length of the standard sample;

[0024] According to the rule that "the smaller the error, the greater the weight", the corresponding weight is determined, and after normalization processing, each weight is non-negative and the sum of the weights is one, to obtain the weighted coefficient of the development stage.

[0025] When a new standard sample is introduced, the error index is recalculated and the weighting coefficient is updated accordingly.

[0026] Preferably, after the unified image scale is established, the morphological parameters of selected parts are measured according to the unified image scale, structured measurement results are generated, and the structured measurement results are associated with image numbers, development stage identifiers and the final calibrated body length and output.

[0027] Preferably, the correspondence between the head length and the body length ratio-development stage is segmented and determined by using the following ratio threshold value:

[0028] When the head length and the body length ratio is greater than 9.5%, it is determined to be juvenile;

[0029] When the head length and the body length ratio is between 8.7% and 9.5%, it is determined to be young;

[0030] When the head length and the body length ratio is less than or equal to 8.7%, it is determined to be adult.

[0031] Preferably, the calculation formula of the final calibrated body length is:

[0032] Wherein, the calculation formula of the weighting coefficient of the development stage is:

[0033] In the formula, is the final calibrated body length at the development stage ; is the first body length preliminary calculation value converted based on the head length statistics in the segment at the development stage ; is the second body length preliminary calculation value obtained based on the regression model of the head length and the body length ratio at the development stage ; is the weighting coefficient of the development stage , and the value range is 0 to 1; is the average absolute error of the first body length preliminary calculation value at the development stage , based on the measured body length of the standard sample; is the average absolute error of the second body length preliminary calculation value at the development stage , based on the measured body length of the standard sample.

[0034] The present application discloses the following technical effects:

[0035] ​​This invention uses the ratio of head aperture length to body length as an intrinsic scale parameter to extract inherent individual characteristics from images, enabling scale conversion without relying on external scales, buoys, or reference objects. This design avoids the influence of water condition changes, shooting angle, and distance on calibration accuracy, making the measurement process more stable and safer. It is particularly suitable for non-invasive monitoring scenarios of cetaceans in the wild or protected waters.

[0036] Existing photogrammetry methods often use a single regression model to cover all individuals, making it difficult to adapt to differences in body proportions at different growth stages. This invention automatically determines an individual's developmental segment by establishing a correspondence between "head aperture length and body length ratio—developmental stage," and within each segment, determines the head aperture length statistic as a benchmark parameter. This achieves precise model adaptation across juvenile, adolescent, and adult individuals, thereby improving measurement accuracy and structural fit.

[0037] Traditional single-channel regression methods neglect body structure information and are susceptible to noise samples. This invention simultaneously calculates a first preliminary body length value based on intra-segment cephalomum length statistics and a second preliminary body length value based on a regression model of cephalomum length to body length ratio within each developmental stage. Through complementary fusion of dual paths, it balances the stability of physiological characteristics with the generalization ability of statistical regression, ensuring that the model is both biologically sound and data-adaptive.

[0038] Existing technologies often employ empirical weights or fixed coefficients, making it difficult to dynamically respond to differences in model accuracy at different stages. This invention determines the weighting coefficients for developmental stages by calculating the mean absolute error of standard samples. Based on the rule of "the smaller the error, the greater the weight," it performs segmented weighted integration, thereby reducing the impact of single-model bias at the source and achieving error self-correction. This ensures that the final calibrated body length maintains high accuracy output under different samples and environments.

[0039] This invention establishes a unified image scale after obtaining the final calibrated body length, integrating the scale results with image numbering, developmental stage identification, and body length results for unified output, thus constructing a standardized morphological measurement system that can be sustainably reused. This scale can be used not only for body length statistics but also extended to the quantification of multidimensional morphological parameters such as body width and dorsal fin height, providing a repeatable and comparable measurement basis for individual cetacean monitoring, population growth assessment, and ecological conservation research.

[0040] In summary, this invention establishes a cetacean morphological measurement model with adaptive, updatable, and high-precision characteristics through a system design that incorporates endogenous biological scales, developmental segmentation, dual-path estimation, and error-weighted integration. This model overcomes the limitations of traditional photogrammetry methods that rely on external scales and single regression models, and has significant application value in ecological research and conservation management. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0043] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0044] 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.

[0045] The purpose of this invention is to provide a cetacean morphological photogrammetry method based on an endogenous biological scale coupled with a segmented weighted ensemble model. By deeply integrating the endogenous biological characteristics of individual cetaceans with the segmented weighted ensemble model, a body size measurement method that combines physiological rationality and mathematical accuracy is constructed, realizing the automated, standardized, and high-precision acquisition of cetacean morphological parameters in complex marine environments.

[0046] 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.

[0047] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model, comprising:

[0048] Step 100: Extract the pixel lengths corresponding to the head aperture length and body length from the acquired whale back image, and calculate the ratio of head aperture length to body length.

[0049] Step 200: Based on the correspondence between the ratio of head aperture length to body length and developmental stage, determine the developmental stage to which the individual belongs, and determine the corresponding intra-segment head aperture length statistics within the developmental stage.

[0050] Step 300: Using the cephalic length statistics within the segment corresponding to the developmental stage as the endogenous biological scale, the image pixel scale is converted into the actual scale to obtain the preliminary calculated value of the first body length.

[0051] Step 400: Substitute the ratio of cephalometric length to body length into the preset regression model to obtain the preliminary calculated value of the second body length;

[0052] Step 500: During the developmental stage, assess the local error or confidence level of the preliminary calculated values ​​of the first and second body lengths based on standard samples, and determine the weighting coefficients for the developmental stage;

[0053] Step 600: According to the weighting coefficients of the developmental stage, perform segmented weighted integration of the preliminary calculated values ​​of the first and second body lengths to obtain the final calibrated body length.

[0054] Specifically, this embodiment calculates the ratio between the head aperture length and body length based on the actual measured body length and head aperture length data of 82 finless porpoises:

[0055] ×100%

[0056] Based on the reproductive and developmental characteristics of the Yangtze finless porpoise and the research team's experience in artificial breeding, juveniles complete nutritional weaning around one year of age and begin self-sufficient hunting and feeding, reaching sexual maturity around 4.5 years of age (using the age of late-maturing males as the cutoff, excluding sex differences). Therefore, in this embodiment, individuals not yet fully weaned before one year of age are considered juveniles; those who begin self-sufficient hunting and do not rely on breast milk between one and 4.5 years of age, but are not yet sexually mature, are considered juveniles; and sexually mature individuals older than 4.5 years are considered adults, based on the Yangtze finless porpoise age-body-length formula: ; In this embodiment, 82 Yangtze finless porpoises were divided into three stages according to their body length and developmental stage: juvenile (age < 1.0 y, body length < 115.3 cm); young adult (1.0 y ≤ age < 4.5 y, 115.3 cm ≤ body length < 140.0 cm); and adult (age ≥ 4.5 y, body length ≥ 140.0 cm), where y is an abbreviation for year.

[0057] Furthermore, the descriptive statistics of the overall cephalomium length data of the Yangtze finless porpoise in this embodiment show that the overall variation of the data is extremely small, with a standard deviation of 1.2 cm and a coefficient of variation of only 9.72%, indicating that the data is highly concentrated around the mean of 11.8 cm. The middle 50% of the data ranges only 2.0 cm (11.0-13.0 cm), and the standard error of 0.127 and the 95% confidence interval (11.6-12.1 cm) both show that the data dispersion is extremely low and the overall performance is stable.

[0058] The cephalometric length data were divided into three groups based on age: juvenile (11.0±0.8 cm), adolescent (11.7±1.2 cm), and adult (12.3±1.0 cm). The coefficient of variation (CV) within each group showed a low dispersion level of ≤10.1%, with the juvenile group exhibiting the lowest dispersion (standard deviation SD=0.8, CV=7.33%). The 95% confidence intervals for the means (μ) of all three groups were less than 1.1 absolute units (with a confidence half-width of only ±0.35 for the adult group). The core distribution (interquartile range IQR 11.0–13.0) of the box plots closely matched the median. Furthermore, the means showed only a slight increasing trend across different age groups, and ANOVA analysis revealed no significant differences between adjacent age groups. Therefore, in this embodiment, the arithmetic mean within each age group can be used as a typical parameter for subsequent biological model calculations.

[0059] Based on the body length (x-axis) and age group grouping of the Yangtze finless porpoise, a system was constructed. , The scatter plot and corresponding regression curves are shown. The allometric growth model for the cephalic aperture length is as follows: Its allometric growth index is 0.231. Here, BHL is the head pore length, which is the linear distance from the snout tip to the posterior edge of the blowhole in cetaceans, and is an important parameter used in this method to establish an endogenous biological scale; BL represents body length, which is the overall length from the snout tip to the end of the tail fin, and is the main independent variable used for modeling; f represents the function mapping, used to describe the regression relationship between the two, that is, the functional expression of BHL as a function of BL. is the significance probability, used to measure the statistical significance of the regression relationship in the model. When The results indicate that the correlation between cephalometric aperture length and body length is significant at the 99% confidence level, meaning the model is reliable. The coefficient of determination reflects the goodness of fit of the model. This indicates the number of samples, specifically the number of whale samples used in the regression calculation.

[0060] Therefore, the application mechanism of this method is as follows: for a photograph of a finless porpoise with a straight posture and an intact back, the proportion of the cranial aperture can be calculated first to determine the age group of the porpoise—when When the percentage is greater than 9.5%, the finless porpoise can be identified as juvenile (age < 1.0 y, body length < 115.3 cm), with an average head aperture length of 11.0 cm; when the percentage is less than 8.7%, the finless porpoise can be identified as juvenile (age < 1.0 y, body length < 115.3 cm), with an average head aperture length of 11.0 cm. When the percentage is ≤9.5%, the finless porpoise can be identified as a juvenile (1.0 y ≤ age < 4.5 y, 115.3 cm ≤ body length < 140.0 cm), and its average head aperture length is 11.7 cm; when If the percentage is ≤8.7%, the finless porpoise can be identified as an adult (age ≥4.5 years, body length ≥140.0 cm), with an average head aperture length of 12.3 cm. After determining the average head aperture length of the finless porpoise in the photograph to its age group and using it as a scale, morphological data represented by body length can be measured. This method is called the "average head aperture length method". It represents the percentage of head aperture length to body length.

[0061] Specifically, building Scatter plot and trinomial regression model Comparing the linear regression model (AIC=201), binomial regression model (AIC=203), and trinomial regression model (AIC=197), the necessity of introducing a cubic term was ultimately determined. The trinomial regression model is as follows: . The good fit of the regression model suggests that it can be directly constructed. A regression model is used to directly predict the body length of the finless porpoise using data on the proportion of its cranial aperture length. An "automatic model selector" is built using R language to select the optimal model based on the AIC value. .in, To represent the F-statistic and its degrees of freedom, F is the statistic for the analysis of variance (ANOVA) or the overall significance test of regression, used to determine whether the independent variables in the regression model have a significant impact on the dependent variable. (3,78) represents two degrees of freedom, where 3 represents the number of independent variables in the regression model (or the model's degrees of freedom); and 78 represents the degrees of freedom of the residual term (sample size minus the number of parameters). To adjust the coefficient of determination; The total body length is the complete body length of a cetacean individual from the snout to the end of the tail fin; AIC is the Akaike Information Criterion, used for model optimization.

[0062] As body length increases, the proportion of head aperture length generally decreases, but there are certain differences between different age groups. The curve slope is steepest in the juvenile stage, with a head aperture length proportion of 10.6±1.2%, showing a rapid decline; in the juvenile stage, the head aperture length proportion is 8.9±0.9%, showing a gentle decline; after entering the adult stage, the head aperture length proportion is 8.1±0.9%, and the decline trend with increasing body length becomes more pronounced.

[0063] Substituting the body length thresholds categorized by age group (juveniles: BL < 115.3 cm; adolescents: 115.3 cm ≤ BL < 140.0 cm; adults: BL ≥ 140.0 cm) into the regression equation, the corresponding theoretical values ​​for the head aperture length ratio were calculated and used as characteristic nodes for the head aperture length ratio at different developmental stages: Juveniles: >9.5%; Youth: 8.7% < ≤9.5%; Adults: ≤8.7%.

[0064] Furthermore, in order to integrate the "head hole length mean method" and the "head hole length proportion regression method" and improve the accuracy of the calculation results of the two methods, this embodiment decides to adopt a segmented weighted ensemble model, which is optimized by dynamic weight allocation based on the local confidence of standard samples.

[0065] Based on the piecewise weighted ensemble model formula, the mean absolute error of the head hole length method for the juvenile group was calculated ( =6.2, Mean absolute error of the head hole length ratio regression method ( =12.2, weighting coefficient =0.66, final calculated body length =109±12.1 cm (sample size n=11); Youth group =10.9、 =3.8, weighting coefficient =0.26, final calculated body length =135±10.5 cm (sample size n=36); Adult group =11.2、 =10.0, weighting coefficient =0.47, final calculated body length =148±14.0 cm (sample size n=35); therefore, the final calculated body length of the population is... =137±17.6 cm (sample size n=82).

[0066] This embodiment employs four statistical analysis methods—error index, paired t-test, regression analysis, and Bland-Altman analysis—to comprehensively evaluate... =138±22.4 cm (n=82) =136±13.1 cm (n=82) and =137±17.6 cm (n=82) Three sets of data and The statistical differences between the values ​​of 136.5 ± 17.3 cm (n = 82) were statistically significant.

[0067] The results show: Mean error (ME) = 1.57 cm, mean absolute error (MAE) = 10.36 cm, root mean square error (RMSE) = 12.8 cm; : ME = 0 cm, MAE = 9.33 cm, RMSE = 11.25cm; : ME = 0.65 cm, MAE = 8.72 cm, RMSE = 10.46 cm. In ME, The optimal state is one without systematic bias. Secondly, The valuation is clearly overestimated. In the MAE, absolute error ratio 6.5% lower than 15.8% lower, significantly better. In RMSE, The dispersion is the lowest (compared to) 7% lower than (18% lower), indicating better robustness to outliers. Comprehensive analysis... It outperforms other methods across the board in both MAE and RMSE, and has only a slight positive bias in ME (0.65), making it the most practical.

[0068] In the paired t-test results, (p = 0.2691) (p = 0.9977) and (p = 0.5775) and There were no significant differences between the groups.

[0069] The results of the linear regression analysis show that, R² = 0.678, slope = 0.636, intercept = 48.7; R² = 0.573, slope = 1, intercept = 0.015; R² = 0.671, slope = 0.806, intercept = 25.936. and The values ​​are close and all higher than This results in better collinearity.

[0070] Bland-Altman analysis results Mean deviation = 1.57 (95% consensus limit: -23.48 to 26.62); Mean deviation = 0 (95% consensus limits: -22.2 to 22.19); Mean deviation = 0.65 (95% consensus limit: -19.94 to 21.24). The average deviation is slightly higher than However, it is still within an acceptable range, with the narrowest consistency limit, indicating a significant reduction in data volatility. The stability and consistency of the measurement results are the best among the three groups, with reductions of 17.8% and 7.2% compared to the initial method group.

[0071] Based on the above results, it can be seen that by combining the two methods to dynamically allocate weights in each age group, the body length calculation data has been optimized, and a segmented weighted ensemble model has been successfully established.

[0072] Optionally, the established segmented weighted ensemble model is applied to predict the body length of the 62 collected finless porpoise image samples. The predicted body length is calculated using the "mean head opening length method" based on the proportion of head opening length presented in the images. =138±18.8 cm; predicted using the "head hole length ratio regression method". =136±11.0 cm; Using a piecewise weighted ensemble model, combined with existing weighting coefficients for each age group. (Juvenile / Youth / Adult) standards are used to calculate the final body length. =137±15.1 cm.

[0073] This embodiment employs five statistical analysis methods—error index, paired t-test, regression analysis, correlation analysis, and Bland-Altman analysis—to comprehensively evaluate... (n=62) and actual body length Whether the differences between them are significant.

[0074] The error index results show that the mean deviation (ME) is only 0.04 cm and the mean relative error (MRE) is 0.37%, indicating that from a global perspective, right There was no obvious systematic overestimation or underestimation trend in the predicted true values. The mean absolute error (MAE=6.69 cm) and root mean square error (RMSE=7.75 cm) showed that the average absolute deviation of individual predicted values ​​from true values ​​ranged from 6.69 to 7.75 cm, with a mean absolute deviation of 7.2 cm. The mean absolute percentage error (MAPE=5.02%) and root mean square relative error (RMSRE=5.97%) showed that the average relative error between the calculated value and the true value in a single measurement was about 5%, which is within an acceptable range in the context of finless porpoise body length measurement.

[0075] The paired t-test results showed p=0.965, t=-0.04, indicating that there was no statistically significant difference between the two groups, supporting the conclusion that... right The overall mean was consistent; the effect size (Cohen's d=0.01) was extremely small, further proving that there was only a very slight difference between the two groups of data.

[0076] The regression analysis results show that the slope of the regression line (0.984) is close to the ideal value of 1, and the intercept (2.108) is small, indicating that the predicted values ​​can retain the relative relationship of the original numerical scale relatively completely; the coefficient of determination (R²=0.785) indicates that the model can explain about 78.5% of the variation in the true value.

[0077] The Bland-Altman consensus analysis showed an average bias of only 0.37%, but the 95% consensus threshold was relatively wide ([-11.40%, 12.13%]). This indicates that in approximately 95% of the measurement points, the relative error between the calculated and the true values ​​may fluctuate within the range of ±11%-12%. This wide range suggests a significant risk of individual variation, and attention should be paid to the potential impact of extreme bias cases on practical applications.

[0078] for right Both sets of data, the mean deviation (ME) and paired t-test results indicate that the predicted values ​​have no overall directional bias; the slope of the regression model is close to 1 and the intercept is close to 0, which meets the theoretical requirements for predictive ability; however, the relative error of individual measurements fluctuates to some extent (approximately ±12%) (RMSRE>MAPE), and coupled with the relatively wide consistency limit in the Bland-Altman analysis, it suggests that there may be a significant risk of misjudgment in extreme cases. The piecewise weighted ensemble model performs reliably in overall estimation and can be applied to the calculation of body length data from complete images of the Yangtze finless porpoise's back, where high average level requirements are needed (the impact of individual variability needs to be assessed in conjunction with the specific application scenario), and can be used to determine the image scale, laying a necessary benchmark foundation for the calculation of other morphological data.

[0079] like Figure 2 As shown, the technical approach of this embodiment includes the following steps:

[0080] Step 1: Calculate the head hole length ratio.

[0081] Obtain an image of the Yangtze finless porpoise with its body straight and intact. Extract two key parameters from the image: head aperture length (BHL, the length from the snout to the posterior edge of the blowhole) and body length (BL, the straight-line length from the snout to the tail). Then, use the formula: head aperture length ratio (BHL / BL). = (Cephalopore length ÷ Body length) × 100% to calculate the cephalopore length percentage of the finless porpoise, which will serve as the core basic parameter for subsequent age group determination and body length prediction.

[0082] As an optional implementation, before extracting the pixel lengths corresponding to the head aperture length and body length, the following preprocessing sub-steps are performed on the finless porpoise dorsal image:

[0083] The acquired back image is subjected to color space transformation and brightness histogram quantile calculation to obtain highlight candidate regions and guided repair using neighborhood non-reflective textures to obtain a highlight-suppressed image.

[0084] Frequency domain bandstop filtering and spatiotemporal consistency detection are performed on the specular suppression image to generate an occlusion mask, which is used to limit the participating area for subsequent contour extraction.

[0085] An energy optimization model was constructed under the constraint of an occlusion mask to obtain the ridgeline, and affine correction was performed along the principal direction of the ridgeline to obtain the preprocessed back image, ridgeline, occlusion mask, and pose correction parameters. The ridgeline extraction satisfies the following conditions:

[0086] ;

[0087] in, The optimal spine line; To normalize the arc length Parameterized candidate ridge lines; The image shows the back after highlight suppression; The gradient vector of the image at the position on the curve; Let be the curvature of the curve at that point; This is a smoothing tradeoff coefficient.

[0088] In one embodiment, a resolution of is selected. Pixelated image of the back of a Yangtze finless porpoise After suppressing highlights and reflections, the ridgeline is solved using an energy optimization model: the arc length parameter is... In the interval The sampling points are evenly spaced at 200 points, and the Sobel operator is used for calculation. The curvature is approximated by a second-order difference with a pixel spacing of 1. Smoothing tradeoff coefficients are taken Numerical optimization (30 iterations in total) yielded the optimal ridgeline. Based on this, affine pose normalization is completed, and preprocessed images, ridgelines, occlusion masks, and pose correction parameters are output; after conversion using a unified image scale, Its length is 1.46 meters and its maximum curvature is The residual tilt angle after attitude correction is This meets the quality requirements for subsequent extraction of head hole length and body length pixel length. Among them, This is the normalized arc length parameter; The geometric curvature of the curve; The smoothing tradeoff coefficient for the curvature term is taken in this embodiment. .

[0089] Step 2: Determine the age group of the finless porpoise and the corresponding average head aperture length.

[0090] 2.1 Age group determination criteria.

[0091] Based on the established correlation between age group and head pore length of the Yangtze finless porpoise, and combined with the calculations in step 1, the following results were obtained. The value is used to determine the age group of the finless porpoise:

[0092] like If the percentage is greater than 9.5%, the finless porpoise is considered to be juvenile (age < 1.0 y, corresponding to a body length range BL < 115.3 cm).

[0093] If 8.7% < If the percentage of finless porpoises is ≤9.5%, then the porpoise is considered to be a juvenile (1.0 y ≤ age < 4.5 y, corresponding to a body length range of 115.3 cm ≤ BL < 140.0 cm).

[0094] like If the percentage is ≤8.7%, the finless porpoise is considered an adult (age ≥4.5 years, corresponding to a body length range BL ≥140.0 cm).

[0095] 2.2 Extract the average length of the cephalometric hole for the corresponding age group.

[0096] Based on the age groups of the finless porpoises determined above, the average head aperture length corresponding to each age group is extracted as a key parameter for subsequent calculation of body length using the "average head aperture length method":

[0097] Juveniles: The average length of the cephalomena is 11.0 cm;

[0098] Youth: The average length of the cephalium is 11.7 cm;

[0099] Adults: The average length of the cephalomena is 12.3 cm.

[0100] Therefore, by using the length of the cephalothorax as a scale, a preliminary calculation of the body length can be obtained. )

[0101] Meanwhile, body length was initially predicted using the "cephalic aperture length ratio regression method," combined with the corresponding regression model: Substitute It is worthwhile to make a preliminary calculation of the body length ( ).

[0102] Step 3: Determine the weighting coefficients for each age group ( ).

[0103] 3.1 Clarify the calculation logic of the weighting coefficients.

[0104] Weighting coefficients ( , where k represents the age group: childhood / adolescence / adulthood, is used to balance the calculation errors of the "mean head hole length method" and the "head hole length percentage regression method." Its core calculation logic is: based on the mean absolute error (MAE) of the two methods in the standard samples of each age group, the contribution of the method with smaller error is highlighted by dynamically allocating weights. The formula is as follows: ,in:

[0105] Within a certain age group (k), the average absolute error when calculating body length using the "average head hole length method" (i.e., the average absolute deviation between the calculated body length and the actual body length, defined as the first preliminary body length calculation value).

[0106] The average absolute error (defined as the preliminary calculated value of the second body length) when calculating body length using the "head hole length ratio regression method" within a certain age group (k).

[0107] 3.2 Substitute the error data to calculate the weighting coefficients.

[0108] MAE data for each age group (childhood:) obtained from previous standard sample statistics. =6.2 cm, =12.2 cm; Youth: =10.9 cm, =3.8 cm; Adult: =11.2cm, =10.0 cm), substituting these values ​​into the above formula, we obtain the weighting coefficients for each age group:

[0109] childhood: =6.2÷(6.2+12.2)=0.66;

[0110] youth: =10.9÷(10.9+3.8)=0.26;

[0111] adult: =11.2÷(11.2+10.0)=0.47.

[0112] Step 4: Calculate the final calibrated body length ( And determine the image scale.

[0113] 4.1 Calculate the final calibration length.

[0114] Based on the preliminary calculated body length obtained from the two methods in step 2 ( Body length is calculated using the "average cephalometric length method" with the average cephalometric length of the corresponding age group as the benchmark. The preliminary predicted body length obtained through the "cephalic length ratio regression method" is combined with the weighted coefficients for the corresponding age group determined in step 3. Substituting into the core formula of the piecewise weighted ensemble model: The final calibrated body length of the finless porpoise was calculated. ).

[0115] 4.2 Determine the image scale.

[0116] Based on the above endogenous segmented weighted ensemble model, the pixel length corresponding to the calculated body length of the finless porpoise in the dorsal image measurement is obtained, and a conversion relationship of "pixel length - actual length" (i.e., image scale) is established. For example, if the body length (110.0 cm) in the image corresponds to a pixel length of 550 pixels, then the scale is "5 pixels = 1 cm". If intuitive labeling is required, the actual length of 20 cm can be used as the standard in the image, and the corresponding pixel length can be calculated according to the scale and labeled with "20 cm". This serves as a unified standard for subsequent measurement and calculation of other morphological data such as back width, body girth, and body mass of the finless porpoise.

[0117] Step 5: Extended Applications (Morphological Data and Body Mass Calculation).

[0118] Based on the image scale determined in step 4, the back width data of the Yangtze finless porpoise is automatically extracted or manually measured within the range of 0-90% of its body length (18 measurement points in 5% intervals). The lateral height data is calculated by combining the body height ratio (HW ratios), and then the body girth data is obtained. The trunk mass is calculated using the cetacean elliptical volume formula and trunk tissue density. The mass of the pectoral fin / tail fin is predicted by combining the linear regression model of pectoral fin / tail fin mass and body length. The total mass of the porpoise is obtained by adding the three together, realizing the complete measurement and calculation of body length, multi-dimensional morphological parameters, and body mass.

[0119] As an optional implementation method, the total mass is calculated as follows:

[0120] (1) During the developmental stage, the trunk volume is calculated and its mass is converted using a triaxial ellipsoid approximation combined with piecewise shape correction:

[0121]

[0122] in, This represents the volume of the torso (after shape correction). The shape correction factor for the developmental stage; These are the torso length, maximum body width, and maximum body thickness, calculated according to a unified image scale. Pi; For trunk mass; This refers to the density parameters of trunk tissues during the developmental stage.

[0123] (2) During the developmental stage, regression predictions of pectoral and caudal fin masses were established based on the final calibrated body length, and the accessory fin masses were summed to obtain the mass of the accessory fins:

[0124]

[0125] in, It is the sum of the masses of the pectoral fin and the caudal fin; The regression intercept and slope of pectoral fin mass versus final calibrated body length; The regression intercept and slope of caudal fin mass and final calibrated body length; For the final calibration of body length.

[0126] (3) The total quality is obtained by performing an error-driven weighted ensemble of the developmental stage on the "quality of the geometry-density link" and the "quality of the regression link":

[0127]

[0128] in, For the final total mass; The quality integration weighting coefficient for the developmental stage (values ​​range from 0 to 1); This is the total mass regression estimate established using the final calibrated body length.

[0129] (4) The quality integration weighting coefficients are determined according to the standard sample error of the developmental stage:

[0130]

[0131] in, The quality integration weighting coefficient for the developmental stage; The mean absolute error of the total mass regression estimate at this developmental stage; For geometric density estimation The mean absolute error at this developmental stage.

[0132] The beneficial effects of this invention are as follows:

[0133] (1) Wide applicability of citizen science: It enables reliable morphological assessment (body length, weight, physical condition) through any back image, unlocking the potential of citizen science and greatly expanding the opportunities for data collection.

[0134] (2) Non-invasive and ethical: The method is completely non-contact and conforms to the best practices of minimizing disturbance to endangered wildlife worldwide.

[0135] (3) Direct management application: The automated process developed in this invention provides key health indicators for protection managers, which are crucial for assessing individual fitness and population status.

[0136] This invention presents a cutting-edge, applied methodological advancement that directly addresses a pressing conservation problem. It transcends mere methodology, providing a practical and scalable tool for cetacean ecological management.

[0137] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0138] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model, characterized in that, include: The pixel lengths corresponding to the head aperture length and body length are extracted from the acquired whale back images, and the ratio of head aperture length to body length is calculated. Based on the correspondence between the ratio of head aperture length to body length and developmental stage, the developmental stage of an individual is determined, and the corresponding intra-segment head aperture length statistics are determined within the developmental stage. Using the cephalom length statistics within the segment corresponding to the developmental stage as an endogenous biological scale, the image pixel scale is converted into the actual scale to obtain the preliminary calculated value of the first body length. Substituting the ratio of the cephalometric length to the body length into a preset regression model, a preliminary calculated value of the second body length is obtained; During the developmental stage, the local error or confidence level of the preliminary calculated values ​​of the first body length and the second body length are evaluated based on standard samples to determine the weighting coefficient of the developmental stage; According to the weighting coefficients of the developmental stages, the preliminary calculated values ​​of the first and second body lengths are segmented and weighted to obtain the final calibrated body length.

2. The cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model according to claim 1, characterized in that, Also includes: A conversion relationship between pixel length and actual length is established based on the final calibrated body length and the pixel length corresponding to the body length, and the conversion relationship is used as a unified image scale to complete the scale anchoring of morphological parameter measurement.

3. The cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model according to claim 1, characterized in that, Also includes: The acquired back images were checked for posture straightness, occlusion, image sharpness, and illumination uniformity.

4. The cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model according to claim 1, characterized in that, The process of establishing and applying the correspondence between the cephalometric length and body length ratio and developmental stage includes: Based on standard samples, the threshold or distribution range of the "ratio of head aperture length to body length" for each developmental stage is statistically analyzed to determine the ratio judgment boundary for each developmental stage, thus forming the correspondence that maps the ratio to the developmental stage; the developmental stages include infancy, adolescence, and adulthood. When processing new individuals, the developmental stage is determined according to the proportion determination boundary in the correspondence relationship; the proportion determination boundary is incrementally updated as new standard samples are introduced, and the correspondence relationship is updated synchronously.

5. The cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model according to claim 1, characterized in that, The statistic for the head hole length within a segment is at least one of the following: the mean within the segment, the robust mean within the segment, or the quantile within the segment.

6. The cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model according to claim 1, characterized in that, The preset regression model is at least one of multinomial regression, piecewise regression, or regression with regularization constraints.

7. The cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model according to claim 1, characterized in that, The process of determining the weighting coefficients for the developmental stages includes: During the developmental stage, the error index between the preliminary calculated value of the first body length and the preliminary calculated value of the second body length was calculated using the measured body length of the standard sample. The corresponding weights are determined according to the rule of "the smaller the error, the greater the weight", and normalized to make each weight non-negative and the sum of the weights equal to one, thus obtaining the weighting coefficients for the developmental stage. When a new standard sample is introduced, the error index is recalculated and the weighting coefficients are updated accordingly.

8. The cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model according to claim 2, characterized in that, After establishing the unified image scale, the morphological parameters of the selected parts are measured according to the unified image scale to generate structured measurement results. The structured measurement results are then associated with the image number, developmental stage identifier, and the final calibrated body length and output.

9. The cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model according to claim 4, characterized in that, The correspondence between the cephalometric length and body length ratio and the developmental stage is determined segmentally using the following ratio threshold: When the ratio of the length of the cephalothorax to the length of the body is greater than 9.5%, it is considered a juvenile. A child is classified as a juvenile when the ratio of the length of the cephalothorax to the length of the body is between 8.7% and 9.5%. A person is considered an adult when the ratio of the length of the cephalothorax to the length of the body is less than or equal to 8.7%.

10. The cetacean morphological photogrammetry method based on an endogenous bioscale coupled piecewise weighted ensemble model according to claim 1, characterized in that, The formula for calculating the final calibration body length is: ； The formula for calculating the weighting coefficients for developmental stages is as follows: ； In the formula, For developmental stage The final calibration body length is as follows; For developmental stage Below is the preliminary calculated value of the first body length obtained based on the statistical conversion of the head hole length within the segment; For developmental stage Below is the preliminary calculated value of the second body length obtained based on the regression model of the ratio of cephalometric length to body length; Developmental stage The weighting coefficients range from 0 to 1. For developmental stage The mean absolute error of the first body length is calculated based on the measured body length of the standard sample. For developmental stage The average absolute error is calculated based on the measured body length of the standard sample and the preliminary calculated value of the second body length.

Images