Forensic medicine pelvic adult bone age estimation method based on enhanced OfficientNetV2

Abstract

The invention provides a forensic medicine pelvic adult bone age estimation method based on enhanced OfficientNetV2. The forensic medicine pelvic adult bone age estimation method comprises the following steps: firstly, acquiring MSCT image data of a pelvis; performing 3D reconstruction on the MSCT image data to obtain 3D image data; and finally, determining a bone age estimation result of the forensic pelvic adult according to the preprocessed and data enhanced MSCT image data, the 3D image data and the enhanced OfficientNetV2 model. According to the enhanced OfficientNetV2 model, a linear layer and a nonlinear activation function are added on the basis of an OfficientNetV2 model, so that an output value of the model is mapped into a predicted value, and the predicted value is controlled within a specified range. According to the pelvic adult bone age estimation method based on the deep learning, an original OfficientNetV2 model is improved, a linear layer and a nonlinear activation function are added to map an output value of the model into a predicted value and control the predicted value in a specified range, exploratory application of deep learning in the aspect of forensic pelvic adult bone age estimation is achieved, the influence of subjective factors is effectively avoided, and the accuracy of an automatic recognition result is improved.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and in particular relates to a forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2. Background Art

[0002] Over the years, the use of some indicators in the pelvic bones has become the most widely used age inference method in forensic practice. Using the pelvis in the skeletal imaging data of corpses to infer age is an important way to achieve forensic bone age identification. In recent years, a series of studies have been carried out based on 3D reconstruction technology such as CT / MSCT, with the help of manual operation and doctor-level measurement of various indicator points or graded values ​​on the pelvic surface.

[0003] The existing technology requires manual operation and doctor-level assessment of each index point or graded value on the pelvic surface when inferring pelvic age. The results are affected by the subjectivity of manual operators and doctor-level experts, which greatly reduces the accuracy of individual age recognition results. Summary of the invention

[0004] In view of this, the present invention provides a forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2, aiming to solve the problem in the prior art that the accuracy of individual age recognition results is greatly reduced due to the subjectivity of manual operators and doctor-level experts.

[0005] A first aspect of an embodiment of the present invention provides a forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2, comprising:

[0006] Acquire MSCT image data of forensic pelvis;

[0007] Performing 3D reconstruction on the MSCT image data to obtain 3D image data;

[0008] Determine the bone age estimation result of the pelvis based on the MSCT image data, 3D image data and the enhanced EfficientNetV2 model;

[0009] Among them, the enhanced EfficientNetV2 model adds linear layers and nonlinear activation functions on the basis of the EfficientNetV2 model to map the output value of the model to the predicted value and control it within the specified range.

[0010] In one possible implementation, a linear layer is used to map the feature vector output by the EfficientNetV2 model to a predicted value of pelvic bone age;

[0011] The nonlinear activation function is the Sigmoid activation function;

[0012] The Sigmoid activation function is used to determine the probability distribution of pelvic bone age within a specified range based on the predicted values.

[0013] In one possible implementation, the enhanced EfficientNetV2 model is trained by the following steps:

[0014] A forensic pelvic sample dataset was obtained, 124 samples were randomly selected as training samples, and the remaining samples were used as validation sets.

[0015] According to the EfficientNetV2 model, linear layer and nonlinear activation function, an enhanced EfficientNetV2 model is constructed;

[0016] After adjusting the model pre-trained on ImageNet using transfer learning technology, the training set is used to train the enhanced EfficientNetV2 model and the optimal model is saved. The optimal model is then used to perform validation tests on the validation set.

[0017] The training parameters were adjusted to improve the model and make it more suitable for estimating the bone age of adults in forensic pelvis.

[0018] In a possible implementation, obtaining a pelvic sample dataset includes:

[0019] Construction of a forensic pelvic database;

[0020] Obtaining MSCT sample image data from the forensic pelvic database;

[0021] Performing 3D reconstruction on the MSCT sample image data to obtain 3D sample image data;

[0022] The MSCT sample image data and 3D sample image data are preprocessed, data enhanced and normalized to obtain a pelvic sample dataset.

[0023] In a possible implementation, preprocessing, data enhancement and normalization are performed on the MSCT sample image data and the 3D sample image data to obtain a pelvic sample data set, including:

[0024] Performing format conversion and resampling on the MSCT sample image data to obtain new MSCT sample image data; and simultaneously intercepting key positions of the 3D sample image data to obtain new 3D sample image data;

[0025] For the new MSCT sample image data and the new 3D sample image data, image affine transformation and Gaussian noise are used to perform data enhancement to expand the number of samples;

[0026] The expanded samples are normalized to obtain the pelvic sample dataset.

[0027] A second aspect of an embodiment of the present invention provides a forensic pelvic adult bone age estimation device based on enhanced EfficientNetV2, comprising:

[0028] An acquisition module, used for acquiring MSCT image data of the forensic pelvis; used for performing 3D reconstruction on the MSCT image data to obtain 3D image data;

[0029] The preprocessing, data enhancement and normalization module performs format conversion and resampling on the MSCT sample image data to obtain new MSCT sample image data; at the same time, the key positions of the 3D sample image data are intercepted to obtain new 3D sample image data; the new MSCT sample image data and the new 3D sample image data are enhanced by using image affine transformation and Gaussian noise to expand the number of samples; the expanded samples are normalized to obtain a pelvic sample data set;

[0030] The estimation module 230 uses the transfer learning technology to adjust the model pre-trained on ImageNet, uses the training set to train the enhanced EfficientNetV2 model and saves the optimal model, and uses the optimal model to perform a validation test on the validation set. The training parameters are adjusted to improve the model to make it more suitable for forensic pelvic adult bone age estimation.

[0031] Among them, the enhanced EfficientNetV2 model adds linear layers and nonlinear activation functions on the basis of the EfficientNetV2 model to map the output value of the model to the predicted value and control it within the specified range.

[0032] A third aspect of an embodiment of the present invention provides a terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 as described in the first aspect above are implemented.

[0033] The fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 as described in the first aspect above.

[0034] The embodiment of the present invention provides a forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2, firstly obtaining MSCT image data of the pelvis; then performing 3D reconstruction on the MSCT image data to obtain 3D image data; finally determining the bone age estimation result of the pelvis according to the MSCT image data, the 3D image data and the enhanced EfficientNetV2 model; wherein the enhanced EfficientNetV2 model adds a linear layer and a nonlinear activation function on the basis of the EfficientNetV2 model to map the output value of the model to a predicted value and control it within a specified range. The present invention improves the original EfficientNetV2 model, adds a linear layer and a nonlinear activation function to control the output value of the model within a specified range, realizes the application of deep learning in forensic pelvic adult bone age estimation, effectively avoids the influence of subjective factors, and improves the accuracy of automatic recognition results. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 It is a flowchart of the implementation of the forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 provided in an embodiment of the present invention;

[0037] Figure 2 It is a structural schematic diagram of a forensic pelvic adult bone age estimation device based on enhanced EfficientNetV2 provided in an embodiment of the present invention;

[0038] Figure 3 It is a schematic diagram of the structure of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0040] Figure 1: is a flowchart of the implementation of the forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 provided by an embodiment of the present invention. Figure 1 As shown, the forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 is characterized by including:

[0041] S110, acquiring MSCT image data of a forensic pelvis; performing 3D reconstruction on the MSCT image data to obtain 3D image data;

[0042] In the embodiment of the present invention, MSCT is Multi-slice Spiral CT, which is developed on the basis of traditional spiral CT. The X-ray tube rotates around the human body and the scanning bed moves at the same time, so as to perform continuous multi-layer scanning on specific parts of the human body. The detector receives the X-ray signal after passing through the human body, and finally forms an MSCT image after a series of complex conversion and processing.

[0043] Compared with traditional CT, MSCT has multiple detector rows and can acquire multi-layer image data simultaneously, greatly improving the scanning speed and image resolution.

[0044] In the embodiment of the present invention, a large amount of two-dimensional image data obtained by MSCT scanning is processed using specialized image processing software and algorithms. Common reconstruction methods include voxel-based methods, surface-based methods, etc. The voxel-based method directly processes the volume data, distinguishes different tissues by setting different thresholds and other operations, and then constructs a 3D model; the surface-based method mainly extracts the surface contour information of the tissue of interest first, and then constructs a three-dimensional surface model based on these contours.

[0045] S120, after resampling and cutting out key parts of the MSCT image data and the 3D image data to generate a new 3D image, the sizes are unified respectively, and then data enhancement is performed on the newly generated MSCT image data and the 3D image data, and then all are normalized.

[0046] S130, determining the bone age estimation result of the forensic pelvis of an adult based on the MSCT image data and the 3D image data after preprocessing and data enhancement, and then based on the enhanced EfficientNetV2 model.

[0047] After adjusting the model pre-trained on ImageNet using transfer learning technology, the training set was used to train the enhanced EfficientNetV2 model and save the optimal model, and the optimal model was used to perform validation tests on the validation set. The training parameters were adjusted to improve the model to make it more suitable for forensic pelvic adult bone age estimation.

[0048] Among them, the enhanced EfficientNetV2 model adds linear layers and nonlinear activation functions on the basis of the EfficientNetV2 model to map the output value of the model to the predicted value and control it within the specified range.

[0049] Traditional methods all require manual operation and doctor-level measurement of each index point or graded value of the pelvic surface. As a result, the accuracy of individual age recognition results is greatly reduced due to the subjectivity of manual operators and doctor-level experts. In addition, the optimization of feature extraction and classifier is carried out separately and cannot be optimized as a whole. Therefore, in addition to referring to the methods of Zhang Zhongyao, Liu Bin, etc., new standards and methods should be continuously introduced, and computer image processing and artificial intelligence technology should be used to seek breakthroughs.

[0050] Deep learning has made great progress in medical imaging disease diagnosis and image segmentation. On the one hand, deep neural networks can combine low-level features into abstract high-level features and have a strong ability to learn image features. On the other hand, medical imaging images have inherent advantages in deep learning research:

[0051] 1) Medical images are collected by specialized medical radiology equipment and have high quality;

[0052] 2) Medical images are stored in DICOM (Digital Imaging and Communications in Medicine) format, which is neat and easy to convert image formats;

[0053] 3) Professional diagnostic information from medical personnel can be used as high-quality label information;

[0054] 4) The black, white and gray levels of medical images are good research objects for image feature recognition by each layer of network structure in deep learning network algorithms.

[0055] Deep learning will be an ideal tool for studying forensic pelvic imaging data. The algorithm's powerful learning ability can fully learn pelvic features, which can not only be used to mine significant features of forensic anthropology such as gender and age, but will also promote the recognition of pelvic bone age in forensic imaging images and push the research on pelvic bone age recognition based on corpses into a new stage.

[0056] The present invention uses the EfficientNetV2-S model to estimate the bone age of forensic pelvis adults, which has the following advantages:

[0057] (1) Strong transfer learning capability.

[0058] Whether migrating to the field of forensics / pattern recognition or in situations with limited data (small samples), pre-trained weights can help improve model performance and speed up convergence.

[0059] (2) Efficient feature extraction.

[0060] The model uses depthwise separable convolution, which effectively reduces the amount of computation and parameters while retaining key feature information, making feature extraction more efficient.

[0061] (3) Good performance

[0062] EfficientNetV2-S performs well in forensic / pattern recognition tasks, has high accuracy and robustness, and can effectively capture patterns and features in complex data. It also has a low risk of overfitting.

[0063] (4) Computational efficiency

[0064] Lower inference time and resource savings.

[0065] Although EfficientNetV2-S has its advantages in feature extraction, it was originally designed for general image classification and other tasks. If the application scenario deviates greatly from its original design intention, such as when it is used for some special medical tasks with specific data distribution and task requirements, the extracted features may not fully meet the needs of subsequent conversion into accurate prediction values. In the pelvic bone age recognition scenario of the present invention, the EfficientNetV2-S model cannot map the features extracted from the model to predicted values, and control the output values ​​of the model within a specified range, that is, the model is not suitable for the regression task of pelvic bone age recognition. Therefore, the present invention improves the model, adds a linear layer and a nonlinear loss function, and forms an enhanced EfficientNetV2 model, so that it can accurately complete the regression task on the basis of the efficient feature extraction of the original model.

[0066] In some embodiments, the linear layer is used to map the feature vector output by the EfficientNetV2 model to a predicted value of pelvic bone age; the nonlinear activation function is a Sigmoid activation function used to determine the probability distribution of pelvic bone age within a specified range based on the predicted value.

[0067] In the embodiment of the present invention, by adding a linear layer to the EfficientNetV2-S model, an effective conversion from feature extraction to bone age estimation can be achieved. This linear layer can map a high-dimensional feature vector to a single output value for direct bone age prediction while maintaining the efficiency and simplicity of the model.

[0068] The Sigmoid function compresses the output of the linear layer to a probability value of a certain age in the range of [0,1], which is suitable for multi-label recognition tasks and can convert continuous pelvic age values ​​into probability distribution. Since the Sigmoid function is a nonlinear function, it can enable the enhanced EfficientNetV2 model (EE) to fit the complex bone age data distribution and help the EE model make more correct age estimation decisions.

[0069] Among them, since the bone age is normalized and preprocessed before model training, when the output of the linear layer is compressed to the range of [0,1] by the Sigmoid function, the predicted bone age needs to be remapped back to the original bone age range.

[0070] In some embodiments, the enhanced EfficientNetV2 model is trained by the following steps: obtaining a forensic pelvic sample data set; constructing an enhanced EfficientNetV2 model based on the EfficientNetV2 model, linear layers and non-linear activation functions; using transfer learning technology to adjust the model pre-trained on ImageNet, using the training set to train the enhanced EfficientNetV2 model and save the optimal model, and using the optimal model to verify the validation set; adjusting the training parameters and improving the model to make it more suitable for forensic pelvic adult bone age estimation.

[0071] In some embodiments, obtaining a forensic pelvic sample data set includes: constructing a forensic pelvic database; obtaining MSCT sample image data from the constructed forensic pelvic database; performing 3D reconstruction on the MSCT sample image data to obtain 3D sample image data; and preprocessing, data enhancement, and normalization on the MSCT sample image data and the 3D sample image data to obtain a forensic pelvic sample data set.

[0072] In the embodiment of the present invention, the forensic pelvis database is composed of 164 Chinese adult pelvis samples, each of which is composed of MSCT image data and 3D image data reconstructed from the MSCT image data.

[0073] In some embodiments, the MSCT sample image data and the 3D sample image data are preprocessed, data enhanced, and normalized to obtain a pelvic sample data set, including: format conversion and resampling of the MSCT sample image data to obtain new MSCT sample image data; simultaneously intercepting key positions of the 3D sample image data to obtain new 3D sample image data; using image affine transformation and Gaussian noise to perform data enhancement on the new MSCT sample image data and the new 3D sample image data to expand the number of samples; and normalizing the expanded samples to obtain a pelvic sample data set.

[0074] In the embodiment of the present invention, the MSCT image data and the 3D image data are preprocessed, specifically, the MSCT image data is converted into .DCM data, and a uniform size (39, 512, 512) is obtained by resampling.

[0075] In the embodiment of the present invention, the key parts of the 3D image data are intercepted to regenerate new 3D image data. Then, the generated new MSCT image data and 3D image data are enhanced by image affine transformation and Gaussian noise to expand the number of samples. Then, the new samples are normalized so that the neural network model constructed later is easier to converge.

[0076] The model trained by the above-mentioned embodiment can more accurately identify the forensic pelvic adult bone age, but different individuals have different patterns and speeds of bone growth and development due to differences in genetic background, nutritional status, living environment, etc. Because the model is usually trained based on a large number of general population samples, if the individual's bone development deviates from the normal pattern due to heredity, the recognition result may be biased. In addition, some congenital diseases and bone diseases can also affect bone growth and development, causing recognition bias. Therefore, the present invention is improved through the following embodiments.

[0077] In summary, the beneficial effects of the present invention are:

[0078] 1. Compared with the research in existing literature, the small sample forensic pelvic bone age estimation algorithm in existing literature needs to use manual operation and doctor-level measurement of each index point or graded assignment of the pelvic surface. The results are affected by the subjectivity of manual operators and doctor-level experts, which greatly reduces the accuracy of individual age recognition results and cannot automatically estimate the small sample forensic pelvic bone age. The present invention introduces deep learning and utilizes the powerful learning ability of the algorithm to fully learn pelvic features. It can not only be used for the mining of significant features of forensic anthropology such as gender and age, but also will promote the recognition of pelvic bone age in forensic imaging images and promote the research based on forensic pelvic bone age recognition into a new stage.

[0079] 2. The present invention improves the EfficientNetV2 model by adding a linear layer and a Sigmoid activation function at the end of the EfficientNetV2 model. It can convert the features extracted by the model into predicted values ​​and control the output values ​​within a specified range, which is suitable for bone age estimation tasks.

[0080] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0081] Figure 2Schematic diagram of the structure of a forensic pelvic adult bone age estimation device based on enhanced EfficientNetV2 provided by an embodiment of the present invention. Figure 2 As shown, in some embodiments, a forensic pelvic adult bone age estimation device 2 based on enhanced EfficientNetV2 includes:

[0082] The acquisition module 210 is used to acquire MSCT image data of the forensic pelvis; perform 3D reconstruction on the MSCT image data to obtain 3D image data;

[0083] The preprocessing, data enhancement and normalization module 220 generates a new 3D image by resampling and cutting out key parts of the MSCT image data and the 3D image data, and then unifies the sizes of the new 3D images. Then, the newly generated MSCT image data and the 3D image data are enhanced and then normalized.

[0084] The estimation module 230 uses the transfer learning technology to adjust the model pre-trained on ImageNet, uses the training set to train the enhanced EfficientNetV2 model and saves the optimal model, and uses the optimal model to perform a validation test on the validation set. The training parameters are adjusted to improve the model to make it more suitable for forensic pelvic adult bone age estimation.

[0085] Among them, the enhanced EfficientNetV2 model adds linear layers and nonlinear activation functions on the basis of the EfficientNetV2 model to map the output value of the model to the predicted value and control it within the specified range.

[0086] Optionally, the linear layer is used to map the feature vector output by the EfficientNetV2 model to a predicted value of pelvic bone age; the nonlinear activation function is a Sigmoid activation function; and the Sigmoid activation function is used to determine the probability distribution of pelvic bone age within a specified range based on the predicted value.

[0087] Optionally, the forensic pelvis adult bone age estimation device 2 based on enhanced EfficientNetV2 also includes a training module, which is used to: obtain a forensic pelvis sample data set; construct an enhanced EfficientNetV2 model based on the EfficientNetV2 model, linear layers and non-linear activation functions; use the transfer learning technology to adjust the model pre-trained on ImageNet, use the training set to train the enhanced EfficientNetV2 model and save the optimal model, and use the optimal model to verify the verification set; adjust the training parameters and improve the model to make it more suitable for forensic pelvis adult bone age estimation.

[0088] The present embodiment provides a pelvic bone age estimation device based on enhanced EfficientNetV2, which can be used to execute the above method embodiments. Its implementation principle and technical effects are similar, and this embodiment will not be repeated here.

[0089] Figure 3 Schematic diagram of the structure of the terminal provided by the embodiment of the present invention. Figure 3 As shown, a terminal 3 provided by an embodiment of the present invention includes: a processor 30, a memory 31, and a computer program 32 stored in the memory 31 and executable on the processor 30. When the processor 30 executes the computer program 32, the steps in the above-mentioned embodiment of the forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 are implemented, for example Figure 1 Alternatively, when the processor 30 executes the computer program 32, the functions of each module / unit in the above-mentioned system embodiments are implemented, for example Figure 2 The functions of each module are shown.

[0090] An embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps in the above-mentioned forensic pelvic adult bone age estimation method embodiment based on enhanced EfficientNetV2.

[0091] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0092] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0093] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0094] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2, characterized in that: include: Acquire MSCT image data of the pelvis; Performing 3D reconstruction on the MSCT image data to obtain 3D image data; Determine the bone age estimation result of the forensic pelvis adult according to the enhanced EfficientNetV2 model after preprocessing and data enhancement of the MSCT image data and the 3D image data; Among them, the enhanced EfficientNetV2 model adds a linear layer and a nonlinear activation function on the basis of the EfficientNetV2 model to map the output value of the model to a predicted value and control it within a specified range.

2. According to claim 1, a forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 is characterized in that: The linear layer is used to map the feature vector output by the EfficientNetV2 model into a predicted value of pelvic bone age; The nonlinear activation function is a Sigmoid activation function; The Sigmoid activation function is used to determine the probability distribution of pelvic bone age within a specified range based on the predicted value.

3. A forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 according to claim 1, characterized in that: The enhanced EfficientNetV2 model is trained by the following steps: Get the forensic pelvic sample dataset; According to the EfficientNetV2 model, linear layer and nonlinear activation function, an enhanced EfficientNetV2 model is constructed; The enhanced EfficientNetV2 model is trained according to the pelvic sample dataset.

4. A forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 according to claim 3, characterized in that: Get the pelvis sample dataset, including: Construction of a forensic pelvic database; Acquire MSCT sample image data from the constructed forensic pelvic database; Performing 3D reconstruction on the MSCT sample image data to obtain 3D sample image data; The MSCT sample image data and the 3D sample image data are preprocessed, data enhanced and normalized to obtain a forensic pelvic sample data set.

5. A forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 according to claim 4, characterized in that: Preprocessing and normalizing the MSCT sample image data and the 3D sample image data to obtain a forensic pelvic sample data set, including: Performing format conversion and resampling on the MSCT sample image data to obtain new MSCT sample image data; and simultaneously intercepting key positions of the 3D sample image data to obtain new 3D sample image data; Performing data enhancement on the new MSCT sample image data and the new 3D sample image data by using image affine transformation and Gaussian noise to expand the number of samples; The expanded samples are normalized to obtain the pelvic sample data set.

6. A terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the steps of the forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 as described in any one of claims 1 to 8 above.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the forensic pelvic adult bone age estimation method based on enhanced EfficientNetV2 as described in any one of claims 1 to 8 above.

Images