Firing pin state determination method, apparatus, and device

Abstract

The embodiment of the application provides a firing pin state determination method, device and equipment, the method comprises the following steps: collecting the primer firing pin mark data on the shell after firing; extracting a plurality of characteristic parameters of the primer firing pin mark data, and constructing an observation feature matrix based on the plurality of characteristic parameters; obtaining a plurality of fusion feature vectors corresponding to the plurality of characteristic parameters from the observation feature matrix, and mapping the plurality of fusion feature vectors into a preset classification model to determine the firing pin mark form of the shell; and determining the firing pin state of the fired shell according to the firing pin mark form. The method is used for comprehensively analyzing different characteristic information according to the firing pin mark data, and improves the accuracy of the firing pin state determination.

Description

Technical Field

[0001] This application relates to the field of military equipment maintenance technology, and in particular to a method, apparatus and equipment for determining the status of a firing pin. Background Technology

[0002] In the analysis of firing pin marks, existing technologies typically use single-modal data such as two-dimensional images or three-dimensional point clouds to extract features and then perform classification or depth assessment based on these features.

[0003] However, single-modal feature representation has limited dimensions, two-dimensional images are easily affected by illumination and noise and are difficult to reflect three-dimensional structures, and three-dimensional point cloud computing has a large amount of computation and depends on sensing accuracy. Furthermore, simple feature fusion and traditional modeling methods based on these limitations are susceptible to redundant features and collinearity interference, resulting in insufficient classification stability and depth assessment accuracy.

[0004] Therefore, how to improve the effectiveness of multi-source feature representation in striker mark analysis, while taking into account noise suppression, classification stability and depth assessment accuracy, has become a technical problem that needs to be solved. Summary of the Invention

[0005] This application provides a method, apparatus, and device for determining the firing pin status, which comprehensively analyzes different feature information based on firing pin mark data to improve the accuracy of firing pin status determination.

[0006] In a first aspect, embodiments of this application provide a method for determining the state of a firing pin, the method comprising:

[0007] Collect data on primer firing pin marks on the cartridge case after firing;

[0008] Multiple feature parameters are extracted from the primer striker mark data, and an observation feature matrix is ​​constructed based on the multiple feature parameters;

[0009] Multiple fused feature vectors corresponding to multiple feature parameters are obtained from the observed feature matrix, and the multiple fused feature vectors are mapped to a preset classification model to determine the firing pin mark morphology of the cartridge case;

[0010] The firing pin condition is determined based on the firing pin mark morphology.

[0011] In one possible implementation, the primer striker mark data includes: two-dimensional region data and three-dimensional voxel data;

[0012] The data on primer striker marks on the cartridge case after firing includes:

[0013] Acquire the first grayscale image and first point cloud data of the primer recess area after the firing pin of the cartridge case is fired;

[0014] Local extraction is performed on the first grayscale image and the first point cloud data to obtain the second grayscale image and the second point cloud data of the striker mark region.

[0015] The second grayscale image and the second point cloud data are binarized respectively to obtain the two-dimensional region data and three-dimensional voxel data of the striking mark region.

[0016] In one possible implementation, the feature parameters include at least: shape features, intensity features, and texture features;

[0017] The construction of the observation feature matrix based on multiple feature parameters includes:

[0018] Extract the shape features, intensity features, and texture features of the two-dimensional region data and the three-dimensional voxel data;

[0019] The observation feature matrix is ​​obtained by concatenating the shape feature, the intensity feature, and the texture feature.

[0020] In one possible implementation, the step of analyzing the observed feature matrix to obtain multiple fused feature vectors corresponding to multiple feature parameters includes:

[0021] Extract multiple effective signal components from the observation feature matrix to decompose the observation feature matrix into a low-rank signal matrix and a random noise matrix;

[0022] The weight values ​​corresponding to the multiple effective signal components are determined, and the multiple effective signal components are weighted and fused based on the weight values ​​corresponding to the multiple effective signal components to obtain multiple fused feature vectors.

[0023] In one possible implementation, extracting multiple valid signal components from the random noise matrix includes:

[0024] Analyze the observation feature matrix to determine the noise threshold of the observation feature matrix;

[0025] Based on the noise threshold, all signal components of the observed feature matrix decomposition are determined, and if a signal component is greater than the noise threshold, the signal component is determined to be a valid signal component.

[0026] Otherwise, the signal component is determined to be a noise component.

[0027] In one possible implementation, determining the weight values ​​corresponding to the plurality of effective signal components includes:

[0028] Determine the signal-to-noise ratio of the multiple effective signal components;

[0029] The weight value of each effective signal component is determined based on the signal-to-noise ratio corresponding to the multiple effective signal components.

[0030] The step of mapping multiple fused feature vectors to a preset classification model to determine the firing pin mark morphology of the cartridge case includes:

[0031] Multiple fused feature vectors are mapped to a preset linear transformation layer to obtain corresponding feature mapping units;

[0032] A feature mapping relationship is constructed based on the linear mapping unit using a nonlinear activation function;

[0033] Based on the feature mapping relationship, the deviation between the multiple fused feature vectors and the corresponding preset standard vectors in the preset classification model is determined;

[0034] Based on the deviation, the firing pin mark morphology of the cartridge case is determined.

[0035] In one possible implementation, after determining the deviations between the plurality of fused feature vectors and the corresponding preset standard vectors in the preset classification model, the method further includes:

[0036] Based on the feature mapping relationship, multiple loss factors corresponding to the preset classification model are determined;

[0037] A loss function is constructed based on multiple loss factors, and the preset classification model is updated using the loss function;

[0038] The deviation between the multiple fused feature vectors and the preset standard vector is updated using the updated preset classification model.

[0039] Secondly, embodiments of this application provide a firing pin state determination device, the device comprising:

[0040] The acquisition module is used to collect data on the primer firing pin marks on the cartridge case after firing;

[0041] An extraction module is used to extract multiple feature parameters from the primer striker mark data and construct a first observation feature matrix based on the multiple feature parameters;

[0042] The analysis module is used to analyze and obtain multiple fused feature vectors corresponding to multiple feature parameters from the first observation feature matrix, and map the multiple fused feature vectors to a preset classification model to determine the firing pin mark morphology of the cartridge case;

[0043] The determining module is used to determine the firing pin state corresponding to the firing pin mark shape based on the firing pin mark shape.

[0044] In one possible implementation, the primer striker mark data includes: two-dimensional region data and three-dimensional voxel data;

[0045] This data acquisition module is also specifically used for:

[0046] Acquire the first grayscale image and first point cloud data of the firing pin mark area of ​​the cartridge case;

[0047] Local extraction is performed on the first grayscale image and the first point cloud data to obtain the second grayscale image and the second point cloud data of the striker mark region.

[0048] The second grayscale image and the second point cloud data are binarized respectively to obtain the two-dimensional region data and three-dimensional voxel data of the striking mark region.

[0049] In one possible implementation, the feature parameters include at least: shape features, intensity features, and texture features;

[0050] This extraction module is also specifically used for:

[0051] Extract the shape features, intensity features, and texture features of the two-dimensional region data and the three-dimensional voxel data;

[0052] The observation feature matrix is ​​obtained by concatenating the shape feature, the intensity feature, and the texture feature.

[0053] In one possible implementation, the analysis module is further used for:

[0054] Extract multiple effective signal components from the observation feature matrix to decompose the observation feature matrix into a low-rank signal matrix and a random noise matrix;

[0055] The weight values ​​corresponding to the multiple effective signal components are determined, and the multiple effective signal components are weighted and fused based on the weight values ​​corresponding to the multiple effective signal components to obtain multiple fused feature vectors.

[0056] In one possible implementation, the analysis module is further used for:

[0057] Analyze the observation feature matrix to determine the noise threshold of the observation feature matrix;

[0058] Based on the noise threshold, all signal components of the observed feature matrix decomposition are determined, and if a signal component is greater than the noise threshold, the signal component is determined to be a valid signal component.

[0059] Otherwise, the signal component is determined to be a noise component.

[0060] In one possible implementation, the analysis module is further used for:

[0061] Determine the signal-to-noise ratio of the multiple effective signal components;

[0062] The weight value of each effective signal component is determined based on the signal-to-noise ratio corresponding to the multiple effective signal components.

[0063] In one possible implementation, the analysis module is further used for:

[0064] Multiple fused feature vectors are mapped to a preset linear transformation layer to obtain corresponding feature mapping units;

[0065] A feature mapping relationship is constructed based on the linear mapping unit using a nonlinear activation function;

[0066] Based on the feature mapping relationship, the deviation between the multiple fused feature vectors and the corresponding preset standard vectors in the preset classification model is determined;

[0067] Based on the deviation, the firing pin mark morphology of the cartridge case is determined.

[0068] In one possible implementation, the analysis module is further used for:

[0069] Based on the feature mapping relationship, multiple loss factors corresponding to the preset classification model are determined;

[0070] A loss function is constructed based on multiple loss factors, and the preset classification model is updated using the loss function;

[0071] The deviation between the multiple fused feature vectors and the preset standard vector is updated using the updated preset classification model.

[0072] Thirdly, embodiments of this application provide a firing pin state determination device, which is used to employ the firing pin state determination method as described in the first aspect and / or any of the various possible implementations of the first aspect.

[0073] Fourthly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0074] The memory stores the instructions that the computer executes;

[0075] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0076] Fifthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0077] In a sixth aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0078] The firing pin state determination method, apparatus, and device provided in this application collect primer firing pin mark data on the cartridge case after firing, extract multiple feature parameters, and construct an observation feature matrix, enabling a more comprehensive structured representation of the firing pin mark information. By analyzing multiple fused feature vectors corresponding to multiple feature parameters from the observation feature matrix and mapping them to a preset classification model to determine the firing pin mark morphology of the cartridge case, the method can reduce the interference of insufficient single feature representation and redundant features and collinearity factors on the determination results, thereby improving the stability and accuracy of firing pin mark morphology recognition. By determining the firing pin state of the fired cartridge case based on the firing pin mark morphology, the method can effectively determine the firing pin state, thereby improving the reliability of firing pin state analysis. Attached Figure Description

[0079] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0080] Figure 1 A flowchart illustrating the firing pin status determination method provided in this application;

[0081] Figure 2 A schematic diagram of the firing pin state determination device provided in this application;

[0082] Figure 3 A schematic diagram of the structure of the electronic device provided in this application.

[0083] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0084] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended specification.

[0085] To facilitate a clear description of the technical solutions in the embodiments of this application, the terms "first" and "second" are used in the embodiments of this application to distinguish identical or similar items with essentially the same function and purpose. For example, "first device" and "second device" are merely used to distinguish different devices and do not limit their order of execution. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that "first" and "second" do not necessarily imply that they are different.

[0086] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0087] In this application, "at least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0088] Firing pin mark analysis is a technical field that intersects with ammunition inspection, weapon condition assessment and intelligent image analysis. It is typically used in quality inspection of ammunition production lines, maintenance and repair of weapon equipment, evaluation of firing pin performance, and data interpretation in related test scenarios.

[0089] In these application scenarios, the surface of the cartridge case after firing will form a corresponding firing pin mark. The detection system needs to acquire data that can be analyzed around the firing pin mark area, and determine whether the firing pin is aging, loose, worn or firing abnormally.

[0090] In existing engineering practices, related systems typically include a cartridge case conveying or securing mechanism, a firing pin mark data acquisition unit, a feature processing unit, and a result output unit. The data acquisition unit acquires firing pin mark data on the cartridge case after firing, while the feature processing unit analyzes the acquired data and outputs firing pin mark morphology recognition results or firing pin condition assessment results. This type of technology not only relates to the reliability of ammunition firing but also directly impacts subsequent weapon maintenance decisions and quality release assessments.

[0091] For example, in mass production, if abnormal firing pin conditions cannot be identified in a timely manner, it may lead to a decrease in firing consistency in the same batch of ammunition; in equipment maintenance, if key information reflecting the performance degradation of the firing pin cannot be effectively extracted from the firing pin marks, it will also increase the risk of missed fault detection.

[0092] Current striker condition detection relies on manual comparison of striker mark morphology, which has large subjective errors and low efficiency. Traditional image processing only extracts texture / contour features, and each feature is judged independently. It cannot integrate multi-dimensional feature association information. Noise such as spots, wear, and reflection can easily interfere with the judgment, and it is difficult to accurately identify tiny striker wear and deformation defects.

[0093] To address the aforementioned issues and the pain points of high noise and difficulty in distinguishing minute defects in striker mark images, this application proposes a striker state determination method. This method constructs an observation matrix from the multidimensional features of the striker mark, introduces random matrix spectral distribution analysis to eliminate noise interference, and refines and effectively fuses feature vectors before combining them with a classification model to accurately identify the morphology of the striker mark. Ultimately, this achieves automated, high-precision, non-destructive testing of the striker state, overcoming the technical shortcomings of traditional manual inspection (large errors), traditional feature extraction (susceptibility to noise interference), and difficulty in identifying minute striker defects.

[0094] Figure 1 This is a flowchart illustrating the firing pin state determination method provided in an embodiment of this application, as shown below. Figure 1 As shown in the embodiment of this application, the firing pin state determination method includes:

[0095] S101. Collect data on primer firing pin marks on the cartridge case after firing.

[0096] Among them, the firing pin mark data is the input data that characterizes the firing pin mark formed on the surface of the cartridge case after the firing pin is fired. It is used for subsequent feature extraction and state determination. Specifically, it can be composed of image information and spatial morphology information of the firing pin mark area, and further organized into two-dimensional region data and three-dimensional voxel data, thus providing a basis for the unified extraction of features such as shape, intensity and texture. The firing pin mark area refers to the primer depression area after the firing pin is fired.

[0097] In the specific implementation process, the executing entity can be an industrial testing terminal deployed in an ammunition testing production line, weapon maintenance station, or test interpretation platform, or it can be an industrial control computer, embedded processor, or server equipment connected to the sensor. Before data acquisition, the fired cartridge case is first sent into the testing station through a conveying mechanism, or fixed in a preset posture by a clamping fixture, so that the primer area faces the acquisition unit, and the central axis of the cartridge case is adjusted to be in a predetermined relative position with the optical imaging axis and the contour measurement axis to reduce acquisition errors caused by posture deviation.

[0098] Subsequently, a two-dimensional grayscale or color image of the striker mark area is acquired through an image acquisition module, and the height distribution information of the corresponding area is acquired through a line laser sensor, a structured light module, a confocal displacement sensor, or a macro 3D scanning unit. In one possible embodiment, the image acquisition module and the line laser sensor are triggered uniformly by an embedded microcontroller or a synchronous controller, and the sampling of image frames and contour frames is completed within the same timing window to ensure that the two-dimensional texture information and the three-dimensional shape information of the same striker mark area have a corresponding relationship in spatial coordinates.

[0099] During the data acquisition process, coarse positioning can be performed first based on the outer circle of the primer, the bottom edge contour, or the center reference hole position. Then, the firing pin action area can be locked based on local edge enhancement and center fitting methods. Subsequently, data containing the firing pin mark and its surrounding area of ​​interest can be extracted. For two-dimensional data, the effects of ambient light fluctuations, lens distortion, and surface reflections can be eliminated through noise reduction filtering, brightness normalization, and distortion correction. For three-dimensional data, a stable local depth distribution can be obtained through point cloud registration, outlier removal, surface plane fitting, and height normalization. The spatial morphology of the firing pin mark can then be converted into three-dimensional voxel data or regular elevation grid data according to the voxel resolution.

[0100] The resulting primer striker mark data not only reflects the striker mark contour boundary, indentation depth, and bottom undulation, but also reflects the texture changes and local stress marks on the material surface, providing complete input for subsequent multi-feature joint analysis. Based on the above analysis, it can be seen that by simultaneously acquiring striker mark images and 3D morphology, the problem of accurately characterizing striker mark depth variations by relying solely on single image information can be avoided. It also reduces the impact of insufficient detail texture recognition when using only single depth information, thereby improving the stability and repeatability of subsequent state determination at the data source level. It should be understood that the above example is merely illustrative and not limiting.

[0101] S102. Extract multiple feature parameters from the primer striker mark data and construct an observation feature matrix based on the multiple feature parameters.

[0102] The observation feature matrix is ​​used to carry multiple feature parameters extracted from the primer striker mark data. It is the basic input for subsequent fusion analysis. Specifically, it can be obtained by processing two-dimensional region data and three-dimensional voxel data respectively to obtain shape features, intensity features and texture features, and then splicing them together according to a unified index relationship to form a high-dimensional feature matrix.

[0103] The extraction of multiple feature parameters can be performed after the region of interest is located. First, boundary segmentation is performed on the two-dimensional region data. The main outline of the strike mark is extracted by a combination of threshold segmentation, region growing, edge detection and morphological closing operation. Based on the outline, shape-related features such as boundary perimeter, region area, major and minor axis length, roundness, eccentricity and boundary roughness are calculated. At the same time, intensity-related features such as gray-level mean, gray-level variance, gray-level gradient, Laplacian divergence mean, central region brightness change rate and edge gray-level transition width are extracted based on the gray-level distribution and gradient changes within the outline.

[0104] For texture-related features, a gray-level co-occurrence matrix can be constructed on a two-dimensional image to calculate parameters such as energy, entropy, contrast, correlation, and inverse difference moment. Alternatively, high-frequency texture responses and directional texture components at different scales can be extracted based on wavelet transform.

[0105] Meanwhile, for three-dimensional voxel data or height grid data, parameters reflecting the spatial stress form, such as maximum depth, average depth, depth standard deviation, pit bottom curvature, volume estimate, maximum inscribed circle radius, minimum circumscribed circle radius, sidewall slope, bottom flatness, and depth distribution skewness coefficient, are further extracted, so that two-dimensional features and three-dimensional features form complementary representations on the same object.

[0106] After extracting various features, feature parameters from different sources and with different physical meanings can be arranged into feature vectors according to a preset field order, forming a row or column record for the same cartridge case sample. Repeating this process for multiple samples yields an observation feature matrix corresponding to the number of samples and the feature dimensions. To avoid bias in subsequent analysis due to differences in dimensionality, standardization can be performed before constructing the matrix. For example, range normalization, Z-score standardization, or robust scaling can be used to unify area, depth, grayscale, and texture statistics to a comparable scale. For feature terms with missing values, neighborhood imputation, median imputation, or regression estimation based on historical samples can be used to correct them, ensuring the integrity of the matrix structure.

[0107] After the observation feature matrix is ​​formed, each dimension of the matrix corresponds to a specific physical feature source and also retains the joint change relationship between different features, facilitating subsequent low-rank decomposition, noise suppression, and fusion characterization. Based on the above analysis, by extracting multiple feature parameters covering contour morphology, surface strength, and spatial texture from primer firing pin mark data and organizing them into a unified observation feature matrix, the problem that a single feature cannot comprehensively reflect the working state of the firing pin can be solved. This allows different manifestations of firing pin marks, such as shallow pits, eccentricity, collapsed edges, trailing, and wear deformation, to have a calculable and comparable structured expression, thus establishing a reliable foundation for high-precision classification and determination. It should be understood that the above example is merely illustrative and not limiting.

[0108] S103. Analyze the observed feature matrix to obtain multiple fused feature vectors corresponding to multiple feature parameters, and map the multiple fused feature vectors to the preset classification model to determine the firing pin mark morphology of the cartridge case.

[0109] In this embodiment, the fused feature vector is used to characterize the striker mark features after fusion processing. It is a key intermediate result input to the classification model. Specifically, it can be obtained by analyzing multiple effective signal components from the observed feature matrix and then weighting and fusing them according to the contribution of each signal component to the target discrimination task. The preset classification model is used to determine the striker mark morphology based on the fused feature vector. It is the core model for completing the classification decision. Specifically, it can be a pre-set and trained adaptive weighted depth metric classifier, a classification network formed by combining a convolutional network and a metric learning module, or a multilayer perceptron model with high-dimensional mapping capabilities.

[0110] In the specific implementation process, covariance estimation or spectral decomposition analysis is first performed on the observed feature matrix. The relationship between the eigenvalue distribution and the noise spectrum boundary is examined using random matrix theory to identify components in the matrix primarily dominated by noise and those primarily dominated by effective structural information. When the eigenvalue corresponding to a certain spectral component is higher than a preset noise threshold, it is determined as a valid signal component; otherwise, it is determined as a noise component and suppressed in subsequent reconstruction. The preset noise threshold can be dynamically determined based on the sample size, matrix dimension, and noise variance estimation results, for example, by approximation using the upper bound of the Marchenko-Pastul distribution, thus adapting the threshold to the current data conditions.

[0111] Subsequently, the signal-to-noise ratio, energy percentage, or class separability index is calculated for the retained effective signal components, and weight values ​​are assigned to each component based on these indicators; the larger the weight value, the higher the contribution of the component to distinguishing different striker mark morphologies. After weighted reconstruction of the low-rank signal matrix and each effective signal component based on the weight values, multiple fused feature vectors are obtained. This fusion process can compress the information scattered in the shape, intensity, and texture dimensions of the original high-dimensional features into a more discriminative representation space, and suppress redundant perturbations caused by surface contamination, acquisition jitter, material reflection, and local defects.

[0112] After obtaining the fused feature vector, it is input into a preset classification model for mapping. Taking the adaptive weighted depth metric classifier as an example, the fused feature vector is first mapped to a high-dimensional feature space through a fully connected projection layer. Then, a nonlinear activation function is used to enhance the nonlinear separability between samples of different morphologies. Subsequently, the distance or similarity between the current sample and the prototypes of each category is calculated. The prototypes of each category can be formed by the center vectors of samples of the same type during the training phase. They are used to represent the central distribution of different pin mark morphologies such as regular pin marks, shallow pin marks, skewed pin marks, collapsed pin marks, and worn pin marks in the feature space.

[0113] During classification, Euclidean distance, cosine distance, or Mahalanobis distance can be used to measure the degree of sample deviation. The distance results are then normalized and converted into probabilities for each category, ultimately outputting the corresponding striker mark morphology. If the current sample has high similarity to multiple abnormal categories simultaneously, a composite morphology label can also be output to describe composite anomalies such as both shallowness and eccentricity.

[0114] Based on the above analysis, it can be seen that by performing low-rank decomposition, noise suppression, and weighted fusion on the observed feature matrix, stable features related to the physical mechanism of the firing pin can be highlighted, and the interference of redundant features on the classification boundary can be reduced. Furthermore, combined with a pre-defined classification model possessing high-dimensional mapping and distance discrimination capabilities, the characterization of firing pin marks can be transformed from traditional manual experience-based judgment into a quantifiable and learnable morphological recognition process, thereby significantly improving the distinguishability and consistency of judgments among different firing pin anomalies. It should be understood that the above example is merely illustrative and not limiting.

[0115] S104. Determine the firing pin status of the cartridge case based on the firing pin mark morphology.

[0116] In this embodiment, the firing pin mark morphology is a classification result of the appearance and characteristics of the firing pin marks on the cartridge case, used to indirectly reflect the working state of the firing pin. Specifically, it can be obtained by a preset classification model after judging the fused feature vector, including the morphology category corresponding to regularity, depth, center offset, edge integrity, and local abrasion features. The firing pin state is the final output judgment result, used to indicate whether the firing pin of the cartridge case is aged, loose, or worn, etc. Specifically, it can be judged according to the correspondence between the firing pin mark morphology and the firing pin state.

[0117] Once the firing pin mark morphology is obtained, the system calls the pre-established morphology-state mapping rule library or the state inference model trained by sample annotation to determine the current working state of the firing pin.

[0118] Specifically, when the firing pin mark is regular in shape, within the normal depth range, has complete boundaries, and a stable center position, the firing pin is considered to be in normal condition. When the firing pin mark is shallow overall, has a larger contact area at the bottom of the pit, a gentler edge transition, or multiple samples show a continuous decreasing depth trend, the firing pin is considered to be aging or the spring is not working properly. When the firing pin mark is center-offset, has a tilted outline, and has one side of the pit wall that is significantly steep while the other side is significantly gentle, the firing pin is considered to be loose, misaligned, or the firing axis is offset. When the firing pin mark is rough at the edges, has an irregular bottom, has local trailing, or exhibits multi-peak indentation characteristics, the firing pin tip is considered to be worn, chipped, or damaged on the surface.

[0119] Optionally, a corresponding firing pin condition characterization model can be constructed based on the primer firing pin marks. By acquiring images of primer firing pin marks, the depth, diameter, edge morphology, and three-dimensional firing pin mark parameters are extracted to construct the firing pin condition characterization model. Specifically, the firing pin mark features extracted from the current sample can be input into this characterization model and cross-referenced with the firing pin mark morphology results output by the classification model. The combined output provides a complete firing pin condition judgment result, including the degree of firing pin wear, damage type, and assembly eccentricity, further reducing the probability of misjudgment caused by a single feature dimension and improving the comprehensiveness and accuracy of the firing pin condition assessment.

[0120] In one possible embodiment, to enhance the robustness of the state determination, the morphological results of a single cartridge case can be jointly analyzed with the statistical results of historical samples in the same batch. For example, the proportion of abnormal morphology, the average depth change, and the eccentricity trend in N consecutive cartridge cases can be statistically analyzed. When the proportion and trend exceed a preset threshold, an alarm for abnormal firing pin is output, thereby avoiding misjudgment caused by differences in individual cartridge case materials or occasional acquisition noise.

[0121] The output results can include status category, confidence level, corresponding morphological evidence, and suggested maintenance labels, and can be recorded and fed back through display terminal, host computer software, or production line control system; in scenarios that require closed-loop control, the results can also trigger shutdown for re-inspection, workstation alarm, or equipment maintenance work order generation.

[0122] Based on the above analysis, it can be seen that there is a clear physical relationship between the morphology of the firing pin marks and the state of the firing pin. The depth, symmetry, edge integrity, and bottom texture of the firing pin marks actually reflect the geometric state of the firing pin tip, assembly stability, and firing energy transfer. Therefore, by mapping the morphology classification results to the firing pin state, it is possible to indirectly but effectively identify abnormalities such as firing pin aging, loosening, and wear, avoiding the problems of low efficiency, strong subjectivity, and insufficient consistency caused by relying on disassembly inspection or pure manual visual comparison.

[0123] This application provides a method for determining the firing pin state. By collecting firing pin marks data on the primer casing after firing, extracting multiple feature parameters, and constructing an observation feature matrix, the method can provide a more comprehensive and structured representation of the firing pin mark information. By analyzing the observation feature matrix to obtain multiple fused feature vectors corresponding to multiple feature parameters, and mapping them to a preset classification model to determine the firing pin mark morphology of the casing, the method can reduce the interference of insufficient single feature representation, redundant features, and collinearity factors on the determination results, thereby improving the stability and accuracy of firing pin mark morphology recognition. By determining the firing pin state of the fired casing based on the firing pin mark morphology, the method can effectively determine the firing pin state, thereby improving the reliability of the firing pin state analysis.

[0124] This application embodiment simultaneously acquires two-dimensional region information and three-dimensional spatial morphology information of firing pin marks, and further constructs an observation feature matrix for multi-feature joint analysis. Combined with a low-rank decomposition, noise suppression, and fusion feature vector generation mechanism based on random matrix theory, as well as a pre-defined classification model with high-dimensional mapping and metric discrimination capabilities, a complete processing chain is achieved from raw firing pin mark data to firing pin mark morphology recognition and firing pin status determination. This not only improves the accuracy of identifying firing pin aging, loosening, wear, and firing anomalies, but also enhances the consistency of judgment under different acquisition conditions, different batches of cartridge cases, and different firing pin mark appearances. This allows the method to stably serve application scenarios such as ammunition production quality inspection, weapon maintenance and repair, and firing pin performance evaluation. It should be understood that the above examples are merely illustrative and not limiting.

[0125] Next, in Figure 1Based on the embodiments, the firing pin condition determination method of this application is described in detail. When the prescribed number of firings / fixed maintenance cycle of the firearm is reached, discarded cartridge cases from recently fired firearms are collected for firing pin condition determination. Alternatively, if the firearm exhibits faults such as weak firing, frequent misfires, primer perforation, or abnormal primer deformation of the cartridge case, the cartridge case used in the current firing is extracted, and the inspection and determination process is initiated to check for defects such as wear, chipping, or eccentricity in the firing pin. Furthermore, after the new firing pin is assembled, test firing tests can be conducted, and test cartridge cases can be collected to initiate a batch determination process to verify the quality of the firing pin's machined end face morphology. Alternatively, during long-term storage and regular safety inspections of stockpiled firearms, a portion of the firearms can be randomly selected for test firing to collect cartridge case data and initiate a determination process to assess the corrosion and deformation state of the firing pin after long-term storage.

[0126] In some optional embodiments, the primer firing pin mark data includes: two-dimensional region data and three-dimensional voxel data; for the firing pin marks on the cartridge case, image acquisition and / or sensors can be used to obtain its data dimensions, specifically through:

[0127] Acquire the first grayscale image and first point cloud data of the firing pin mark area of ​​the cartridge case;

[0128] Local extraction is performed on the first grayscale image and the first point cloud data respectively to obtain the second grayscale image and the second point cloud data of the striker mark area;

[0129] The second grayscale image and the second point cloud data are binarized to obtain two-dimensional region data and three-dimensional voxel data of the striking pin mark area.

[0130] The first grayscale image represents the two-dimensional raw visual information of the firing pin mark region, typically acquired by a line scan camera, area scan camera, or other grayscale imaging device. The first point cloud data represents the three-dimensional spatial contour information of the same region, typically acquired by structured light, laser scanning, or a depth camera. The second grayscale image is a local image of the firing pin mark cropped from the first grayscale image, and the second point cloud data is a local point set of the firing pin mark extracted from the first point cloud data; both retain the effective area of ​​the firing pin mark. The two-dimensional region data is used to represent the shape boundary of the firing pin mark on the plane, and the three-dimensional voxel data is used to represent the depth undulation and spatial occupancy of the firing pin mark. Binarization processing can distinguish the firing pin mark region from the background region, making subsequent feature extraction more stable.

[0131] Specifically, the spent cartridge case can be fixed onto the testing fixture first, and the relative positions of the imaging unit and the acquisition unit can be adjusted so that the firing pin mark area (i.e., the primer depression area after firing) is completely within the acquisition field of view. The acquired first grayscale image and first point cloud data can be first aligned and the region located, and then the local extraction window can be determined based on the center position of the firing pin mark area to obtain the second grayscale image and second point cloud data.

[0132] During local extraction, the window size can be set according to the cartridge case caliber, typical size of the firing pin mark, and acquisition resolution to ensure the complete preservation of the firing pin mark edge and minimize background interference. Subsequently, threshold segmentation processing is performed on the second grayscale image and the second point cloud data, distinguishing pixels and point cloud voxels with grayscale values ​​higher or lower than a set threshold as foreground and background. The foreground region is then mapped into two-dimensional region data and three-dimensional voxel data. The voxel data can be generated by discretizing the point cloud, and the voxel side length can be set according to the detection accuracy requirements. For example, a discrete scale in the range of 0.05mm to 0.5mm can be used to balance computational efficiency and spatial resolution. In practical applications, other models of this component can also be selected; this embodiment does not limit this.

[0133] This embodiment allows for the simultaneous preservation of the planar contour and spatial morphology of firing pin marks. Two-dimensional information is beneficial for reflecting the opening boundary, offset, and contour integrity of firing pin marks, while three-dimensional information is beneficial for reflecting the depth, local collapse, and surface undulation differences of firing pin marks. Both types of data provide the basic input for subsequent feature extraction. By first extracting locally and then binarizing, the influence of other areas of the cartridge case, surface reflection, and acquisition noise on the results can be effectively reduced, thereby improving the stability and distinguishability of primer firing pin mark data.

[0134] This implementation method enables the simultaneous acquisition of two-dimensional and three-dimensional representations of firing pin marks at the acquisition end, enhancing the completeness of the firing pin mark information and providing a more reliable data foundation for subsequent observation feature matrix construction, fusion analysis, and firing pin status determination. Because the firing pin mark region is precisely extracted and separated through binarization, background interference is significantly suppressed, thus improving the accuracy of firing pin mark recognition and the consistency of status assessment, thereby enhancing the engineering applicability of cartridge case detection and firing pin status analysis.

[0135] Furthermore, based on the data acquisition process, multiple feature parameters can be extracted from the primer striker mark data, and an observation feature matrix can be constructed based on these feature parameters. The feature parameters include at least shape features, intensity features, and texture features. The process of constructing the observation feature matrix can be as follows:

[0136] Extract shape, intensity, and texture features from two-dimensional region data and three-dimensional voxel data;

[0137] The observation feature matrix is ​​obtained by splicing together shape features, intensity features, and texture features.

[0138] In this embodiment, shape features are used to describe the geometric shape of the striker mark, typically determined by the contour boundary in the two-dimensional region data and the spatial undulations in the three-dimensional voxel data. Specifically, these features may include perimeter, area, inscribed circle radius, and circumscribed circle radius, characterizing the opening size, edge extension, and overall approximate roundness of the striker mark. Intensity features are used to describe the grayscale or response intensity distribution of the striker mark region, jointly characterized by grayscale variations in the two-dimensional region data and height variations in the three-dimensional voxel data. Specifically, parameters such as grayscale gradient and mean Laplacian divergence can reflect local light-dark transitions and abrupt changes in shape. Texture features are used to describe the detailed distribution of the striker mark surface. Based on the grayscale co-occurrence matrix, the grayscale co-occurrence relationship between adjacent pixels can be statistically analyzed, and indicators such as contrast and correlation can be extracted to reflect subtle roughness differences and structural consistency on the striker mark surface.

[0139] In practical implementation, after entering the feature extraction unit, 2D region data and 3D voxel data undergo normalization, noise suppression, and boundary correction processes respectively to reduce the impact of acquisition errors on feature calculation. Subsequently, geometric quantities of planar contours are extracted from the 2D region data, and spatial geometric quantities of voxel connected domains are calculated from the 3D voxel data. These two are then fused to form a shape feature vector. Simultaneously, local gradient analysis and morphological enhancement are performed on both types of data to obtain intensity feature vectors. A gray-level co-occurrence matrix is ​​then constructed based on local neighborhood gray-level relationships, and a texture feature vector is output. These feature vectors are concatenated in a preset order to form an observation feature matrix with consistent dimensions. The rows or columns of this matrix correspond to different data sources or different feature types, facilitating subsequent fusion analysis and classification.

[0140] The observation feature matrix can simultaneously preserve the geometric morphology, local response strength, and texture differences of firing pin marks, enabling subsequent models to identify changes in firing pin mark morphology from multiple perspectives. Since two-dimensional region data focuses on planar contour information, while three-dimensional voxel data focuses on spatial volume information, jointly stitching together their shape, intensity, and texture features avoids information loss caused by a single data source and improves the ability to identify minor wear, eccentric firing, and abnormal impact marks. Using this method, the observation feature matrix has stronger representational integrity and anti-interference capabilities, providing a stable data foundation for firing pin mark morphology determination, thereby improving the accuracy and consistency of firing pin condition analysis.

[0141] In one specific embodiment, the acquired two-dimensional grayscale image and three-dimensional point cloud data are first preprocessed to extract the firing pin impact region on the primer and perform binarization processing to obtain two-dimensional region data and three-dimensional voxel data of the firing pin impact region on the primer. Then, at least a portion of the following shape features are extracted from the firing pin impact region on the primer as shape feature vectors, including but not limited to: boundary perimeter, area of ​​the firing pin impact region on the primer, maximum inscribed circle radius, minimum circumscribed circle radius, target circularity, number of corner points, compactness, sphericity, principal axis ratio, and convexity.

[0142] Specifically:

[0143] Perimeter of the boundary That is, the regional boundary Arc length, extracting the geometry of the region boundary point sequence: Calculate the Euclidean distance between adjacent points and sum them to get: ;in, For the first Coordinates of the boundary points; This represents the total number of boundary points;

[0144] For the primer firing pin impact area on a two-dimensional plane Area above Its calculation formula is ,in, The number of foreground pixels, The side length of a single pixel. The physical size of the pixel (if the physical scale is not a concern, s=1 can be used);

[0145] Maximum inscribed circle radius Its calculation formula is ,in, Let p be the Euclidean distance from the nearest boundary point. For any pixel;

[0146] Minimum circumcircle radius Its calculation formula is By obtaining the center and radius of the smallest enclosing circle, where: The optimal center;

[0147] Target circularity It refers to a feature quantity defined using all boundary points of the target region R, and its definition is: , where the average radius is The position of the boundary point from the centroid is The standard deviation of the radius is Image region re- for ;

[0148] Number of corner points These are shape features that describe the degree of local geometric abrupt changes at the target boundary, used to characterize the number of sharp bends or discontinuous changes in the target contour. At that time, the judgment of the first Each boundary point is a corner point, and the total number of corner points is defined as follows: ,in, For the curvature threshold, the first Discrete curvature at boundary points Defined as The boundary corner is The indicator function is ;

[0149] Compactness It is a global shape feature used to measure the boundary complexity of a target region given an area, reflecting whether the target approaches the "most compact" geometry. Its expression is: ;

[0150] spherical It can describe both two-dimensional and three-dimensional targets, and is defined as follows: When describing a two-dimensional target, This represents the radius of the tangent circle within the target area. This represents the radius of the circumcircle of the target region, with the centers of both circles located at the centroid of the region.

[0151] Spindle ratio Used to describe the stretching and orientation of the target shape, reflecting whether the target is slender, flat, or approximately circular, with a principal axis ratio of [missing information]. ,in, For the extension intensity along the main direction, For the extension strength along the secondary direction, the principal axis length The maximum dimension of the firing pin impact area in the principal direction, and the secondary axis length. The scale of the impact area of ​​the primer firing pin in the vertical principal direction.

[0152] Convexity Used to describe the proximity between a target shape and its convex hull, it is an important indicator for measuring whether a target has depressions, gaps, or irregular boundaries. , Among them, The area of ​​the target region. The area of ​​the convex hull of the target region.

[0153] Further extract feature data corresponding to the three-dimensional voxel data; specifically, for closed volumes in three-dimensional space... , The original image specifically includes:

[0154] volume ,in, This represents the total number of foreground voxels (the number of voxels belonging to the firing pin impact area on the primer). Let the voxel side length be denoted as 'side length'.

[0155] Surface area ,in, The total number of exposed surface faces.

[0156] Mean surface-center distance The overall size and average indentation radius of the firing pin impact area on the primer are described by the following formula: Among them, the surface voxel set of the impact region is The regional center is defined as , No. The coordinates of each surface voxel are , No. The distance from each surface voxel to the center is ;

[0157] Standard deviation of surface-center distance The expression describing the dispersion of the surface of the firing pin impact area from the center is as follows: ;

[0158] Distance between the circumscribed cuboids Its value describes the degree to which the voxels in the firing pin impact area of ​​the primer are uniformly distributed within the circumscribed rectangle. The calculation expression is: In the formula, The center of the 12 intersection lines of the 6 cross-sections of the rectangular prism outside the impact area of ​​the primer firing pin.

[0159] Secondly, the intensity feature data of the primer striker mark data is extracted, specifically including:

[0160] grayscale gradient (From inside to outside) Used to detect the contour and uniform separation area of ​​the firing pin impact zone on the primer, its expression is: ,in, It refers to the number of times the firing pin strikes the area of ​​the primer to perform corrosion operations; For the first Average gray level after secondary etching;

[0161] Mean of Laplace divergence The formula used to distinguish the firing pin impact area on the primer from areas with similar grayscale values ​​is: ,in, For the Laplace operator; For the first Each pixel's grayscale value;

[0162] Laplace range of variation The formula describes the range of divergence in the firing pin impact region of the primer, improving the identification of impact formation characteristics. The calculation expression is as follows: ,in, For the maximum Laplace response; It is the minimum Laplace response.

[0163] Furthermore, texture feature data of the primer firing pin mark data is extracted. In this embodiment, this is based on the gray-level co-occurrence matrix of the firing pin impact region on the primer. Extract texture features.

[0164] Specifically, the grayscale measurement of the firing pin impact area on the primer is performed, with the grayscale measurement level set to a predetermined value (e.g., 32 levels), and a grayscale co-occurrence matrix is ​​constructed in multiple directions. The following texture features are extracted from the grayscale co-occurrence matrix as texture feature vectors: contrast. Correlation ; second moment of angle Inverse difference moment ;

[0165] Gray-level co-occurrence matrix Represented as grayscale With gray level The probability of them occurring together under a given spatial displacement, where... The gray level of the reference pixel. G represents the gray level of neighboring pixels with a given spatial relationship, and G is the number of gray level quantizations. Standard deviation is , , The grayscale mean is , .

[0166] Subsequently, depth feature data of the primer striker mark data was extracted, specifically including:

[0167] Maximum depth is used to describe the deepest indentation in the firing pin impact area on the primer, and it is defined as follows: ,in, The total number of sampling points. For the first The depth value at each point, with the maximum depth directly reflecting the magnitude of the impact energy of the firing pin and the degree of local wear;

[0168] The average depth is used to describe the overall level of indentation in the firing pin impact area on the primer, and it is defined as follows: The average depth can be used to reflect the overall deformation of the impact area.

[0169] The standard deviation of depth is used to describe the dispersion of depth distribution, and its calculation expression is: The larger the standard deviation of depth, the more uneven the depth distribution in the impact area.

[0170] Depth volume is used to characterize the overall indentation in the firing pin impact area on the primer, and it is defined as follows: ,in, For a single sampling unit, the area and depth volume can comprehensively reflect the overall degree of plastic deformation in the impact area.

[0171] Depth uniformity is used to evaluate whether the depth distribution of the firing pin impact area on the primer is uniform; it is defined as follows: ,in, To prevent extremely small constants with a denominator of zero, when the depth uniformity is close to 1, it indicates that the depth distribution is relatively uniform; when it is small, it indicates that the local depressions are significantly different.

[0172] The depth gradient change rate is used to describe the drasticness of changes in the depth field, and its expression is: ,in, for Orientation depth gradient; for The directional depth gradient and the rate of change of the depth gradient can reflect the steepness of the edge of the impact zone and the surface change characteristics.

[0173] Depth skewness reflects the displacement of an impact crater in a certain direction, and its expression is: ;

[0174] Depth kurtosis reflects the sharpness of an impact crater, and its expression is: .

[0175] Based on the above construction steps, the observation feature matrix for striking mark determination can be obtained. In an optional embodiment, in order to suppress the influence of noise in the determination process, it is necessary to analyze the observation feature matrix and extract multiple effective signal components in the observation feature matrix to decompose the observation feature matrix into a low-rank signal matrix and a random noise matrix.

[0176] The weight values ​​corresponding to multiple effective signal components are determined, and the multiple effective signal components are weighted and fused based on the weight values ​​to obtain multiple fused feature vectors.

[0177] Among them, the low-rank signal matrix is ​​used to characterize the dominant effective information directly related to the physical structure of the firing pin mark, and centrally carries the global stable morphological features such as the overall contour morphology of the firing pin mark, the gradient change of the depth of the pit bottom, and the continuity of the edge contour; the random noise matrix is ​​used to characterize various redundant disturbances and unstable interference components generated during the acquisition and imaging transmission of firing pin mark data, specifically including irrelevant interference information such as image gray quantification error, local reflective spots on the surface of the cartridge case, Gaussian noise in the image sensor sampling, and interference from tiny rust particles on the surface of the cartridge case.

[0178] Multiple effective signal components are the main feature components with the value of characterizing the morphology of the striker mark, which are obtained by performing spectral distribution analysis on the observed feature matrix based on random matrix theory. The selection criteria can be any one or more combinations of feature amplitude threshold, local window energy, and component signal-to-noise ratio, so as to accurately retain fine-grained features that can characterize the overall structure and local minor damage of the striker mark, and avoid the key morphological information used to determine the morphology of the striker mark being filtered out by noise.

[0179] The weight values ​​are used to quantify the contribution ratio of each effective signal component to the final fused feature vector. The weight allocation rule comprehensively considers three indicators: data stability, feature peak intensity, and local signal-to-noise ratio of each effective signal component. Higher weights are assigned to components that have a stronger ability to distinguish between normal and abnormal wear patterns of hammer marks, thereby amplifying the proportion of highly discriminative fine-grained features in the fused vector.

[0180] Specifically, the observation feature matrix can be constructed by splicing and fusing the shape features, grayscale intensity features, and surface texture features of the striker mark. The observation feature matrix is ​​then separated into a low-rank signal matrix and a random noise matrix using matrix decomposition. Robust principal component analysis or a matrix decomposition model with low-rank regularization constraints can be used to complete the decomposition. After model constraints, the low-rank signal matrix retains complete information related to the main structure of the striker mark, while the random noise matrix accommodates interference components such as imaging disturbances and surface noise that cannot be explained by the main structure of the striker mark.

[0181] Based on the spectral distribution threshold of random matrix theory, each effective signal component is selected from the observed feature matrix; the corresponding weight value is calculated based on the normalized signal-to-noise ratio of each effective signal component, and the normalized signal-to-noise ratio value is mapped to the weighting coefficient of each component.

[0182] Subsequently, the main features corresponding to the low-rank signal matrix and the selected effective signal components are combined with their respective weighting coefficients to perform weighted summation or feature dimension splicing operations. Then, the overall features after splicing and fusion are normalized to generate a standardized fusion feature vector, which is then input into the preset classification model to complete the needle mark morphology discrimination.

[0183] After adopting the above feature fusion processing scheme, the globally stable low-rank main features of the striker mark and the selected and retained fine-grained effective morphological features can jointly characterize the complete morphological details of the striker mark. While fully preserving the basic features such as the regular contour and depth gradient of the striker mark, it improves the distinguishability of weak defect features such as local defects, edge burrs, and fine wear marks of the striker mark. It effectively suppresses the recognition interference caused by various noises during the acquisition and imaging process, improves the robustness of striker mark morphology recognition under complex working conditions, and provides feature support with better signal-to-noise ratio and more complete information dimensions for subsequent accurate inference of the working state of striker wear, chipping, and eccentric impact.

[0184] It should be noted that the following description is merely an exemplary implementation of the technical solution of this application and does not constitute a limitation on the technical solution of this application. In another optional embodiment, the step of extracting effective signal components may include: analyzing the observation feature matrix and determining the noise threshold of the observation feature matrix;

[0185] Based on the noise threshold, all signal components of the observed feature matrix decomposition are determined. If a signal component is greater than the noise threshold, it is determined to be a valid signal component.

[0186] Otherwise, the signal component is determined to be a noise component.

[0187] The noise threshold is a critical discrimination value obtained by statistical analysis of the spectral distribution and singular value features of the observed feature matrix based on random matrix theory. It is used to distinguish between effective signal components and pure noise components within the observed feature matrix. The signal components are all candidate feature components obtained after decomposing the observed feature matrix. Whether each candidate component is retained is determined by the relationship between its amplitude, feature intensity, and the noise threshold. The noise components are interference components that do not have the value of characterizing the morphology of the strike mark. If the feature intensity does not exceed the noise threshold, it is determined to be a noise component and is removed or its weight is weakened in the subsequent feature fusion process.

[0188] In the specific implementation process, matrix decomposition is first performed on the observation feature matrix to separate the low-rank signal matrix and the random noise matrix; the noise threshold is calculated based on the overall characteristic energy, singular value amplitude or spectral distribution frequency domain statistical characteristics of the observation feature matrix.

[0189] All candidate signal components generated by the decomposition of the observation feature matrix are traversed and compared one by one with a noise threshold: if the feature intensity of a candidate signal component is greater than the noise threshold, it is determined that the component carries the morphological structure information corresponding to the striker mark and is retained as a valid signal component; if the feature intensity of a candidate signal component is not greater than the noise threshold, it is determined to be a worthless noise component. For all the selected valid signal components, a subsequent weighted fusion operation is performed based on the corresponding weights of each component to further improve the completeness and recognizability of the extraction of the true morphological features of the striker mark.

[0190] In one specific embodiment, after extracting the shape, intensity, and texture features of the firing pin impact area on the primer, all candidate features are standardized, and effective features are automatically selected through contribution analysis of random matrix theory. Multi-source features are then uniformly modeled and fused.

[0191] First, the extracted shape features, intensity features, and texture features are combined according to feature dimensions to construct the original feature vector of a single primer striker impact region sample. For the first... The original feature vector of each primer firing pin impact region sample is represented as follows: ,in, This represents the total dimension of the features. Indicates the first The first sample Each feature component.

[0192] By stacking the feature vectors of all samples, an observation feature matrix is ​​constructed: : ,in: To collect the number of samples from the impact area of ​​the firing pin on the primer; The feature dimension is defined as follows: each row of the matrix corresponds to a sample of the firing pin impact area on the primer, and each column corresponds to a feature attribute, which fully describes the distribution of all samples in the multidimensional feature space.

[0193] Since the observation feature matrix not only contains the true structural information reflecting the firing pin state, but is also affected by factors such as image acquisition error, point cloud measurement error, environmental disturbance, and sensor noise, the observation matrix is ​​represented as a superposition of the latent signal matrix and the random noise matrix: ,in, This is the potential signal matrix, used to characterize the true structural information of the firing pin impact area on the primer; This is a random noise matrix used to characterize random interference during the data acquisition process.

[0194] To extract effective structural information and suppress random noise from the observation matrix, singular value decomposition is performed on the observation matrix: ,in, It is a left singular vector matrix. It is a right singular vector matrix. Let be a singular value matrix. A singular value matrix can be represented as: And satisfy: ,in, Indicates the first There are several singular values. The magnitude of the singular value reflects the information energy carried by the corresponding feature direction. Larger singular values ​​usually correspond to the true structural information of the firing pin impact area on the primer, while smaller singular values ​​mainly reflect random noise components.

[0195] To separate effective structural information from random noise in the observation feature matrix, singular value spectra are analyzed based on random matrix theory. Suppose that the observation matrix, after singular value decomposition, yields a sequence of singular values. The upper bound of the theoretical spectrum corresponding to random noise is calculated based on Marchenko–Pastur theory. When singular values Greater than the upper limit of the spectrum When the singular value is found to contain valid structural information, the potential signal component corresponding to that singular value is determined to contain valid structural information; when the singular value is found to contain valid structural information, the potential signal component corresponding to that singular value is determined to contain valid structural information. Less than or equal to the upper limit of the spectrum When this occurs, it is considered to primarily reflect random noise components. Among them,

[0196] This determines the number of valid singular values. and utilize the previous The singular vectors corresponding to the effective singular values ​​can be used to reconstruct the latent signal matrix: ,in, This represents the number of valid latent signal components retained. The latent signal matrix is ​​reconstructed using the retained singular values ​​and their corresponding singular vectors. ,in, and They represent the first The left and right singular vectors corresponding to each singular value. The reconstructed latent signal matrix can effectively retain information related to the structure of the firing pin impact region on the primer, while filtering out most of the random noise components.

[0197] This discrimination mechanism relies on a unified noise threshold to separate effective morphological information from irrelevant disturbances in all candidate components of the observed feature matrix. The retained effective signal components accurately reflect the global structure and local minor damage features of the striker mark. By filtering out noise components with low information contribution through the noise threshold, the negative impact of random interference such as imaging noise, surface scratches, and illumination disturbances on the construction process of the fused feature vector can be significantly reduced. This improves the stability and consistency of the multi-dimensional feature fusion results, providing reliable feature inputs with higher signal-to-noise ratio and less interference for the subsequent striker mark morphology classification and recognition and striker working status inference.

[0198] Based on the aforementioned embodiments, in order to accurately characterize the physical structure of the firing pin mark, it is necessary to perform weighting processing on the effective signal components. The weighting parameters can be determined based on the signal-to-noise ratio of each signal component. Specifically, the signal-to-noise ratios of multiple effective signal components are determined.

[0199] The weight value of each effective signal component is determined based on the signal-to-noise ratio corresponding to multiple effective signal components.

[0200] The signal-to-noise ratio (SNR) measures the reliability and purity of the effective signal component relative to the noise background, characterizing the ratio of signal strength to corresponding noise strength. A higher SNR indicates richer information about the true morphology of the striker mark carried by the effective signal component. The effective signal component is a feature component obtained by spectral distribution filtering of the observed feature matrix based on random matrix theory. It contains key information such as the overall morphology of the striker mark, the texture of the pit bottom, and local minor defects. Targeted filtering removes a large number of irrelevant random disturbances, reducing the interference of noise on subsequent feature fusion.

[0201] In the specific implementation process, after all effective signal components are obtained, the mean amplitude, energy, or variance of each effective signal component is calculated. Simultaneously, the background noise energy and noise variance corresponding to that component are estimated. The signal-to-noise ratio (SNR) is then calculated based on the ratio of signal power to noise power. For discrete effective signal components, the signal power can be solved by the sum of squares of its sample sequence. Combined with blank noise samples or a pre-set noise estimation model, the noise power is obtained, generating SNR parameters that match each effective signal component one-to-one. Furthermore, to adapt to computational requirements, the SNR can be normalized, constraining the values ​​to a unified preset range, enabling horizontal comparisons of effective signal components at different dimensions and scales.

[0202] After obtaining the signal-to-noise ratio (SNR) parameters corresponding to each effective signal component, the matching weight value is calculated based on the relative proportion of the SNR of a single component in the entire SNR set. In one possible implementation, the SNR of a single effective signal component can be divided by the sum of the SNRs of all effective signal components to obtain the normalized weight; another implementation is to first perform a nonlinear mapping transformation on the SNR and then perform a normalization operation to amplify the contribution proportion of high SNR components in feature fusion.

[0203] After processing by this weight allocation rule, effective signal components with higher signal-to-noise ratio and stronger information credibility will be assigned greater weights and occupy a higher proportion of expression when weighted and fused with the low-rank signal matrix; components with lower signal-to-noise ratio will be assigned smaller weights to suppress the interference of residual noise on the fused feature vector.

[0204] In practical applications, the weight values ​​can also be calculated by the processor retrieving pre-stored weight allocation rules from memory, or they can be generated in real time by software programs after the signal-to-noise ratio of each component is statistically analyzed. Adaptive weight allocation based on signal-to-noise ratio allows the fusion result of multiple effective signal components to match the actual information carrying capacity of each component, improving the ability of the fused feature vector to represent abnormal features such as fine textures, local wear, and edge chipping of striker marks. This provides input features with higher signal-to-noise ratio and stronger stability for backend striker mark morphology classification and striker working status determination.

[0205] Specifically, considering that different potential signal components have different abilities to characterize the firing pin state, this application embodiment further uses a signal-to-noise ratio estimation method to evaluate the importance of each potential signal component.

[0206] Let the first The signal-to-noise ratio corresponding to each potential signal component is: The signal-to-noise ratio (SNR) is used to characterize the relative proportion between effective information and random noise in a signal component. A higher SNR indicates that the component contains richer effective structural information and contributes more to the identification of the firing pin status.

[0207] Based on the singular value energy distribution relationship between the signal subspace and the noise subspace, the signal-to-noise ratio parameter corresponding to each potential signal component is calculated. : ;

[0208] According to the corresponding signal-to-noise ratio estimation results Assign weights to each potential signal component : ,satisfy: ,in, Indicates the first The proportion of each potential signal component in the final fusion result. This method allows high signal-to-noise ratio features to receive greater fusion weight, thereby enhancing the ability to express effective information.

[0209] Subsequently, the potential signal components are weighted and fused to obtain the fused feature matrix. : This matrix integrates effective information from shape features, intensity features, and texture features, and significantly reduces the impact of random noise on feature representation.

[0210] Since the fusion matrix still has high dimensionality, to further improve the computational efficiency and generalization ability of subsequent classification models, the feature components in the fusion matrix are sorted according to their contribution, and the top d fusion components with the highest contribution are retained to construct the final fusion feature vector: ,in, Indicates the fusion feature dimension. Indicates the first This vector is a fused feature component. It retains the most discriminative feature representation of the structural information of the firing pin impact region on the primer, while achieving feature dimensionality reduction and noise suppression.

[0211] The final result Using the dimensional fusion feature vector as input to the subsequent firing pin state classification model reduces feature redundancy while ensuring feature discrimination capability, thereby improving the classification model's recognition accuracy and robustness for different firing pin states.

[0212] In one possible implementation, based on the obtained fused feature vectors, the firing pin mark morphology of the cartridge case can be determined by a preset classification model. For example, multiple fused feature vectors are mapped to a preset linear transformation layer to obtain the corresponding feature mapping unit.

[0213] A nonlinear activation function is used to construct a feature mapping relationship based on a linear mapping unit;

[0214] Based on the feature mapping relationship, the deviation between multiple fused feature vectors and the corresponding preset standard vectors in the preset classification model is determined;

[0215] Based on the deviation, determine the shape of the firing pin mark on the cartridge case.

[0216] The system comprises a pre-defined linear transformation layer that linearly maps the fused feature vectors to form an intermediate representation suitable for subsequent classification. The feature mapping unit carries this intermediate representation and serves as the basis for deviation calculation. A non-linear activation function introduces non-linear expressive power into the linear mapping result, making the mapped feature distribution more easily compared with a pre-defined standard vector. The pre-defined standard vector represents the reference feature center for each strike mark morphology category in the pre-defined classification model, while the deviation represents the degree of difference between the fused feature vector and the corresponding standard vector.

[0217] In specific implementation, multiple fused feature vectors can come from the previous feature fusion results and are input into a preset linear transformation layer to perform matrix multiplication and bias superposition operations, obtaining feature mapping units corresponding to each category. Subsequently, a nonlinear activation function is applied based on the feature mapping units to form a feature mapping relationship with class separability. Then, the distance, difference, or similarity between the fused feature vector and each preset standard vector in the preset classification model is calculated based on this mapping relationship to obtain the deviation. The category with the smallest deviation or the largest similarity is output as the firing pin mark morphology of the cartridge case. The preset classification model can be constructed using a multi-class discriminant model, and its parameters can be obtained by training with known firing pin mark samples. In practical applications, other model forms can also be selected for this classification model, which is not limited in this embodiment.

[0218] In one possible implementation, after adaptive weighted fusion based on random matrix theory and noise suppression processing, the final result is obtained. 3D fusion feature vector: The system comprehensively characterizes multi-source information such as the shape, texture, intensity, and depth features of the firing pin impact area on the primer, and has completed random noise suppression and redundant feature elimination.

[0219] To improve the separability between different types of firing pin impact areas on the primer corresponding to different firing pin states, this invention employs an adaptive weighted depth metric classification strategy guided by category prototypes to construct a firing pin state classification model for the firing pin impact area on the primer.

[0220] First, feature space mapping is completed using a fully connected projection layer combined with a non-linear activation function: ,in, Here is the projection layer weight matrix; It is the bias vector; It is a non-linear activation function; This represents the high-dimensional discriminative features after mapping. This mapping process can further enhance the distinguishing ability between samples of different categories, enabling the fused features to have better clustering properties in the new metric space.

[0221] Then, a category prototype is constructed using the training samples, for the first... Class sample set The category prototype is calculated using an adaptive weighting method. ,in, Indicates the first The class prototype corresponding to the firing pin impact area on the primer.

[0222] Unlike traditional mean-based prototypes, the embodiments of this application introduce sample-level adaptive weights. : In the formula, For the sample fitting residuals, This is the weight decay coefficient. This mechanism can reduce the impact of noisy samples, outliers, and samples with blurred boundaries on the class prototype construction process, making the class prototype closer to the true class center.

[0223] Since the depth variation of the firing pin impact area on the primer directly reflects the firing pin impact energy variation, depth feature constraints are introduced during the construction of the category prototype. Let the... The deep feature vectors corresponding to each sample are: The depth feature vector is calculated from 3D point cloud data and includes parameters such as maximum depth, average depth, depth standard deviation, depth volume, depth uniformity, and depth gradient change rate, which are used to describe the degree of indentation and spatial distribution characteristics of the firing pin impact area on the primer.

[0224] For the Class samples build deep prototypes: The depth prototype is used to describe typical patterns of the depth distribution of the firing pin impact area on the primer under different firing pin conditions.

[0225] After completing the prototype construction, calculate the distance relationship between the sample to be classified and the prototypes of each category: ,in, It is a learnable metric matrix used to describe the importance and correlation between different feature dimensions. The distance function is updated using the backpropagation algorithm during training and constrained to a positive semi-definite matrix to ensure that the distance function satisfies the metric space property. Indicates the sample and the first Metric distance between class prototypes.

[0226] Simultaneously calculate depth distance: To utilize both morphological and depth features simultaneously, a joint distance function is constructed: ,in, This represents the depth constraint weight. This joint distance can simultaneously measure the similarity of samples in both morphological and depth spaces, thereby improving the sensitivity of the classification results of the firing pin impact region on the primer to changes in the firing pin state.

[0227] After obtaining the joint distance, the class assignment probability is calculated using probability normalization: ,in, This indicates that the current sample belongs to the first... The probability of a class.

[0228] Based on the cumulative number of firing pins, wear measurement results, and expert annotation information, this invention establishes a set of firing pin status categories: ,in, Indicates a normal state; Indicates a state of slight wear; Indicates moderate wear. This indicates a state of severe wear. Samples for each category are obtained from actual detection results or manual annotation results and are used as training labels for the classification model.

[0229] The final classification result is defined as: That is, the category with the highest probability is selected as the final classification result of the current primer firing pin impact area.

[0230] Furthermore, to further evaluate the reliability of the classification results, a credibility assessment mechanism is introduced. Credibility is used to evaluate the degree of matching between the sample and its corresponding category prototype. Higher credibility indicates that the sample is closer to the center of its category in the feature space; lower credibility indicates that the sample is in the category boundary region or has anomalous features, and its classification result has significant uncertainty. Specifically, let... ,in, When the confidence level is close to 1, it indicates that the current sample has a high degree of consistency with the corresponding category prototype, and the classification result is reliable. When the confidence level is low, it indicates that the sample is in the category boundary area or has abnormal characteristics, and a comprehensive judgment should be made in conjunction with the subsequent firing pin status assessment results. After completing the firing pin status classification of the firing pin impact area on the primer, a firing pin status mapping relationship is established based on the category label, the degree of deviation from the depth prototype, and the confidence index.

[0231] When the sample is close to the normal category prototype, has a normal depth distribution, and high confidence, it can be determined that the firing pin is in normal working condition. When the depth features gradually deviate from the normal depth prototype, and the confidence concentrates on the mild wear category, it can be determined that the firing pin has slight aging or wear. When the morphological features, texture features, and depth features all deviate significantly from the normal category prototype and are classified into the abnormal category, it indicates that the firing pin may have undergone significant wear, local deformation, or fatigue damage. When the sample maintains a large distance from all normal category prototypes, and the confidence concentrates on the severe abnormal category, it can be determined that the firing pin has severe wear or even failure risk and needs further inspection or replacement.

[0232] This allows the fused feature vectors to enter a stable classification space and form a clear comparison benchmark with the standard vectors, thereby improving the discriminative ability of pin mark morphology recognition. By making judgments based on deviations, the probability of misclassification between different pin mark morphologies can be reduced, and the ability to characterize subtle differences on the cartridge case surface can be enhanced, thus improving the accuracy and consistency of pin mark analysis results.

[0233] In order to further improve the accuracy of the strike mark determination during the above steps, in one possible implementation of this application embodiment, multiple loss factors corresponding to the preset classification model are determined according to the feature mapping relationship.

[0234] A loss function is constructed based on multiple loss factors, and the loss function is used to update the preset classification model;

[0235] Using the updated preset classification model, the deviation between multiple fused feature vectors and the preset standard vectors is updated.

[0236] The loss factor characterizes the error contribution between the fused feature vector and the preset standard vector under the current feature mapping relationship, as well as the direction of model parameter correction. It is typically determined by the deviation, class distance, and classification confidence. The loss function unifies the constraints of multiple loss factors to form an optimizable objective, driving the parameter iteration of the preset classification model. The preset classification model can consist of a linear classification layer and a nonlinear discriminant unit. Its parameter updates can be achieved through gradient descent, stochastic gradient descent, or adaptive optimization. The updated model can further refine the standard vector representation corresponding to different striker mark morphologies.

[0237] In practical implementation, the deviation obtained based on the feature mapping relationship can first be converted into deviation values ​​for each category, then combined with category weights to generate multiple loss factors, and finally weighted and summed to form a loss function. The loss function is applied to the classification parameters, mapping parameters, or normalization parameters of the preset classification model to make the model output more consistent with the distribution characteristics of the current pinprick mark samples. After the model is updated, the deviation between the fused feature vector and the preset standard vector is recalculated, which can reduce the errors caused by sample discreteness, acquisition noise, or feature drift, making the pinprick mark morphology determination results more stable.

[0238] For example, to further improve classification reliability and model generalization ability, a total loss function with multiple constraints is constructed during the training phase, which collaboratively optimizes the weighted depth metric loss, adaptive classification loss, multimodal consistency constraint loss, cross-modal contrastive loss, and class prototype smoothing regularization loss: .in, All are adjustable non-negative balance coefficients used to control the weight ratio of each loss term in the overall optimization objective. They are determined through experiments or grid search to achieve the comprehensive optimization of classification accuracy, modal consistency and model stability.

[0239] The weighted depth metric loss is: In the formula, This represents the total number of samples within the training batch. To represent the first in the training set The true category labels corresponding to the samples from the impact area of ​​the firing pin on the primer are manually labeled supervisory information. Indicates the relationship with the current sample The index of the nearest heterogeneous class in the feature space is used to construct inter-class separation constraints in the metric loss. For the first The sample to the first Distance between class prototypes.

[0240] The adaptive weighted classification loss is: In the formula, The normalized exponential function transforms distance into class probability. This is a scaling factor used to control the degree of concentration of the probability distribution.

[0241] The multimodal consistency constraint loss is: In the formula, Feature vectors extracted from branches of a two-dimensional image; Feature vectors extracted from branches of a 3D point cloud; It is the L2 norm, used to measure the magnitude of the eigenvector.

[0242] The cross-modal contrast constraint loss is: In the formula, For the first Two-dimensional and three-dimensional features of a sample; No. Three-dimensional features of each sample ( (for negative sample pairs) This is a temperature coefficient used to adjust the smoothness of the similarity distribution.

[0243] The category prototype smoothing regularization loss is: In the formula, The total number of categories for the classification task; For the first The class's prototype vector; This refers to the gradient information of the category prototype.

[0244] By backpropagating through the loss function, the projection layer parameters, category prototypes, and metric matrix are continuously updated, causing similar samples to continuously aggregate in the feature space and dissimilar samples to continuously move away, forming a category feature space with clear boundaries and regular distribution.

[0245] This processing mechanism performs closed-loop correction on the classification model through loss factors and loss functions, enabling the preset classification model to continuously adapt to changes in the firing pin mark characteristics of different batches of cartridge cases, and improving the consistency and robustness of firing pin mark morphology recognition after updating deviations. Furthermore, since deviations are recorrected with model iterations, the probability of false positives and false negatives is reduced, providing a more reliable basis for determining the firing pin status based on the firing pin mark morphology.

[0246] It should be noted that the following description is merely an exemplary implementation of the technical solution of this application and does not constitute a limitation on the technical solution of this application. The embodiments of this application can also analyze and predict the changing trend of the deepest impact area of ​​the firing pin on the primer. The regression model for the deepest impact area of ​​the firing pin on the primer does not run independently but serves as an auxiliary criterion in the firing pin status assessment module. The classification model is used to determine the status category of the firing pin, while the regression model is used to predict the changing trend of the deepest impact area of ​​the firing pin on the primer. Finally, the classification results, depth prediction results, and confidence index are combined to complete a comprehensive assessment of the firing pin status.

[0247] In this embodiment, a regression modeling method incorporating sparse constraints is constructed to model the relationship between the feature vector of the primer firing pin impact region and the true deepest depth of the primer firing pin impact region. By applying sparse constraints to the regression coefficients during the regression optimization process, the regression coefficients corresponding to features with low correlation to the deepest depth of the primer firing pin impact region are compressed to zero or close to zero, thereby reducing the interference of redundant features on the modeling results and achieving automatic feature selection.

[0248] Meanwhile, a stratified analysis was conducted on the relationship between the features and the deepest depth of the firing pin impact area within different primer pin impact area size ranges to evaluate the stability of the influence relationship between each feature and the deepest depth of the firing pin impact area under different size conditions.

[0249] By constructing a Lasso regression optimization model based on L1 regularization constraints, while ensuring the good fit between the features of the firing pin impact region on the primer and the true deepest depth, the regression coefficients are constrained to suppress redundant features and select key features closely related to the deepest depth of the firing pin impact region on the primer, thereby improving the stability and reliability of the modeling results for the deepest depth of the firing pin impact region on the primer. The Lasso regression model minimizes the following loss function: ,in, It is the target variable (dependent variable); These are model predictions; It is the regression coefficient; It is a regularization hyperparameter (penalty term strength), which controls the degree of shrinkage of the regression coefficients; It represents the number of independent variables. Its core objective function is: In the formula, This is the hyperparameter for regularization strength. Let L1 be the norm of the coefficient vector.

[0250] The striker condition determination method provided in this application involves acquiring two-dimensional grayscale images and three-dimensional point cloud data of striker marks, extracting multi-dimensional features after region cropping and binarization, and then stitching them together to construct an observation feature matrix. Based on random matrix theory, a noise threshold is calculated, and effective signal components carrying striker mark morphology information are selected from the observation matrix. Weights are adaptively assigned based on the signal-to-noise ratio of each component, and low-rank main features and fine-grained effective features are fused to obtain a high-quality fused feature vector. The fused features are then input into a classification model through linear transformation and nonlinear activation. The morphology of the striker mark is identified by calculating the deviation of the feature standard vector, and the classification model is iteratively optimized based on a loss function. Finally, the working state of the striker, such as wear and breakage, is determined based on the morphology of the striker mark. This application embodiment integrates two-dimensional and three-dimensional multi-source features, automatically filters imaging and sampling noise using random matrix theory, and improves the identification of subtle striker mark defects. Adaptive weighting amplifies the effect of high-reliability features, and combined with iterative optimization of the classification model, reduces human judgment errors, achieving high-precision, automated non-destructive testing of the striker's condition.

[0251] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0252] Based on the same inventive concept, this application also provides a firing pin state determination device for implementing the firing pin state determination method described above. The solution provided by this firing pin state determination device is similar to the solution described in the firing pin state determination method above. Therefore, the specific limitations in one or more device embodiments provided below can be found in the limitations of the firing pin state determination method described above, and will not be repeated here.

[0253] In one embodiment, such as Figure 2 As shown, a firing pin state determination device 200 is provided, the firing pin state determination device 200 includes:

[0254] The acquisition module 201 is used to acquire data on primer firing pin marks on the cartridge case after firing;

[0255] Extraction module 202 is used to extract multiple feature parameters from primer striker mark data and construct a first observation feature matrix based on the multiple feature parameters;

[0256] Analysis module 203 is used to analyze and obtain multiple fused feature vectors corresponding to multiple feature parameters from the first observation feature matrix, and map the multiple fused feature vectors to a preset classification model to determine the firing pin mark morphology of the cartridge case;

[0257] The determination module 204 is used to determine the firing pin state corresponding to the firing pin mark shape based on the firing pin mark shape.

[0258] In one possible implementation, the primer striker mark data includes: two-dimensional region data and three-dimensional voxel data;

[0259] The acquisition module 201 is also specifically used for:

[0260] Acquire the first grayscale image and first point cloud data of the firing pin mark area of ​​the cartridge case;

[0261] Local extraction is performed on the first grayscale image and the first point cloud data respectively to obtain the second grayscale image and the second point cloud data of the striker mark area;

[0262] The second grayscale image and the second point cloud data are binarized to obtain the two-dimensional region data and three-dimensional voxel data of the striking pin mark area.

[0263] In one possible implementation, the feature parameters include at least: shape features, intensity features, and texture features;

[0264] The extraction module 202 is also specifically used for:

[0265] Extract shape, intensity, and texture features from two-dimensional region data and three-dimensional voxel data;

[0266] The observation feature matrix is ​​obtained by splicing together shape features, intensity features, and texture features.

[0267] In one possible implementation, the analysis module 203 is further configured to:

[0268] Multiple effective signal components are extracted from the observation feature matrix to decompose the observation feature matrix into a low-rank signal matrix and a random noise matrix;

[0269] The weight values ​​corresponding to multiple effective signal components are determined, and the multiple effective signal components are weighted and fused based on the weight values ​​to obtain multiple fused feature vectors.

[0270] In one possible implementation, the analysis module 203 is further configured to:

[0271] Analyze the observation feature matrix and determine the noise threshold of the observation feature matrix;

[0272] Based on the noise threshold, all signal components of the observed feature matrix decomposition are determined. If a signal component is greater than the noise threshold, it is determined to be a valid signal component.

[0273] Otherwise, the signal component is determined to be a noise component.

[0274] In one possible implementation, the analysis module 203 is further configured to:

[0275] Determine the signal-to-noise ratio of multiple valid signal components;

[0276] The weight value of each effective signal component is determined based on the signal-to-noise ratio corresponding to multiple effective signal components.

[0277] In one possible implementation, the analysis module 203 is further configured to:

[0278] Multiple fused feature vectors are mapped to a preset linear transformation layer to obtain the corresponding feature mapping units;

[0279] A nonlinear activation function is used to construct a feature mapping relationship based on a linear mapping unit;

[0280] Based on the feature mapping relationship, the deviation between multiple fused feature vectors and the corresponding preset standard vectors in the preset classification model is determined;

[0281] Based on the deviation, determine the shape of the firing pin mark on the cartridge case.

[0282] In one possible implementation, the analysis module 203 is further configured to:

[0283] Based on the feature mapping relationship, determine multiple loss factors corresponding to the preset classification model;

[0284] A loss function is constructed based on multiple loss factors, and the loss function is used to update the preset classification model;

[0285] Using the updated preset classification model, the deviation between multiple fused feature vectors and the preset standard vectors is updated.

[0286] Each module in the above-mentioned device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0287] Figure 3 A schematic diagram of the structure of the electronic device provided in this application. Figure 3As shown, the electronic device 300 provided in this embodiment includes at least one processor 301 and a memory 302. Optionally, the device 300 further includes a communication component 303. The processor 301, memory 302, and communication component 303 are connected via a bus 304.

[0288] In a specific implementation, at least one processor 301 executes computer execution instructions stored in memory 302, causing at least one processor 301 to perform the above-described method.

[0289] The specific implementation process of processor 301 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0290] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0291] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0292] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0293] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0294] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0295] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0296] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0297] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0298] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0299] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0300] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0301] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0302] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A method for determining the state of a firing pin, characterized in that, The method includes: Collect data on primer and firing pin marks on the cartridge case after firing; Multiple feature parameters are extracted from the primer striker mark data, and an observation feature matrix is ​​constructed based on the multiple feature parameters; Multiple fused feature vectors corresponding to multiple feature parameters are obtained from the observed feature matrix, and the multiple fused feature vectors are mapped to a preset classification model to determine the firing pin mark morphology of the cartridge case; Based on the firing pin mark morphology, determine the firing pin state corresponding to the firing pin mark morphology.

2. The method according to claim 1, characterized in that, The primer striker mark data includes: two-dimensional region data and three-dimensional voxel data; The data on primer striker marks on the cartridge case after firing includes: Acquire the first grayscale image and first point cloud data of the firing pin mark area of ​​the cartridge case; Local extraction is performed on the first grayscale image and the first point cloud data respectively to obtain the second grayscale image and the second point cloud data of the striker mark region; The second grayscale image and the second point cloud data are binarized respectively to obtain the two-dimensional region data and three-dimensional voxel data of the striking mark region.

3. The method according to claim 2, characterized in that, The feature parameters include at least: shape features, intensity features, and texture features; The construction of the observation feature matrix based on multiple feature parameters includes: Extract the shape features, intensity features, and texture features of the two-dimensional region data and the three-dimensional voxel data; The observation feature matrix is ​​obtained by concatenating the shape feature, the intensity feature, and the texture feature.

4. The method according to claim 1, characterized in that, The step of analyzing the observed feature matrix to obtain multiple fused feature vectors corresponding to multiple feature parameters includes: Extract multiple effective signal components from the observation feature matrix to decompose the observation feature matrix into a low-rank signal matrix and a random noise matrix; The weight values ​​corresponding to the multiple effective signal components are determined, and the multiple effective signal components are weighted and fused based on the weight values ​​corresponding to the multiple effective signal components to obtain multiple fused feature vectors.

5. The method according to claim 4, characterized in that, The extraction of multiple effective signal components from the observed feature matrix includes: Analyze the observation feature matrix to determine the noise threshold of the observation feature matrix; Based on the noise threshold, all signal components of the observed feature matrix decomposition are determined, and if a signal component is greater than the noise threshold, the signal component is determined to be a valid signal component. Otherwise, the signal component is determined to be a noise component.

6. The method according to claim 4, characterized in that, Determining the weight values ​​corresponding to the multiple effective signal components includes: Determine the signal-to-noise ratio of the multiple effective signal components; The weight value of each effective signal component is determined based on the signal-to-noise ratio corresponding to the multiple effective signal components.

7. The method according to claim 1, characterized in that, The step of mapping multiple fused feature vectors to a preset classification model to determine the firing pin mark morphology of the cartridge case includes: Multiple fused feature vectors are mapped to a preset linear transformation layer to obtain corresponding feature mapping units; A feature mapping relationship is constructed based on the linear mapping unit using a nonlinear activation function; Based on the feature mapping relationship, the deviation between the multiple fused feature vectors and the corresponding preset standard vectors in the preset classification model is determined; Based on the deviation, the firing pin mark morphology of the cartridge case is determined.

8. The method according to claim 7, characterized in that, After determining the deviations between the multiple fused feature vectors and the corresponding preset standard vectors in the preset classification model, the method further includes: Based on the feature mapping relationship, multiple loss factors corresponding to the preset classification model are determined; A loss function is constructed based on multiple loss factors, and the preset classification model is updated using the loss function; The deviation between the multiple fused feature vectors and the preset standard vector is updated using the updated preset classification model.

9. A firing pin state determination device, characterized in that, The device includes: The acquisition module is used to collect data on the primer firing pin marks on the cartridge case after firing; An extraction module is used to extract multiple feature parameters from the primer striker mark data and construct a first observation feature matrix based on the multiple feature parameters; The analysis module is used to analyze and obtain multiple fused feature vectors corresponding to multiple feature parameters from the first observation feature matrix, and map the multiple fused feature vectors to a preset classification model to determine the firing pin mark morphology of the cartridge case; The determining module is used to determine the firing pin state corresponding to the firing pin mark shape based on the firing pin mark shape.

10. A firing pin state determination device, characterized in that, The firing pin status determination device is used to employ the firing pin status determination method as described in any one of claims 1-8.

Images