Fossil pattern identification and classification method and system based on image recognition

Abstract

The application discloses a kind of fossil pattern identification classification method and system based on image recognition, it is related to hydraulic engineering technical field, by obtaining the three-dimensional point cloud data of fossil sample, and three-dimensional point cloud data is projected as multi-view two-dimensional texture mapping chart, simultaneously obtains the surface element distribution diagram of fossil sample;The contrast, correlation and entropy value of fossil texture are calculated using the improved GLCM gray level co-occurrence matrix algorithm, the curvature of the fossil in the three-dimensional point cloud data, the normal vector distribution and the geometric parameters of biological part;The surface element distribution diagram is analyzed by PCA principal component analysis method, and the significant element combination related to species is extracted;Based on cross-modal attention mechanism, the feature data is mapped to a unified feature space, and the contribution of each modality is allocated through learnable weights, and the fossil pattern classification result is output. So that a large number of dinosaur fossil samples can be quickly and accurately classified.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of water conservancy projects, in particular to a fossil pattern identification and classification method and system based on image recognition. BACKGROUND

[0002] Traditional dinosaur fossil species identification mainly relies on artificial identification by paleontologists, and is classified by observing the morphological characteristics of the fossils and comparing them with existing specimens. However, this method has many limitations: the completeness and preservation state of the fossils have a great influence, and it is difficult to accurately determine the morphological characteristics of broken or severely weathered fossils; manual identification is low in efficiency, requires a large amount of time and manpower, and the identification results are affected by the subjective experience of experts, and there is a certain error and low identification efficiency. SUMMARY

[0003] The purpose of the present application is to solve the above problems, and a fossil pattern identification and classification method and system based on image recognition are designed.

[0004] To achieve the above purpose, the technical scheme of the present application is as follows: further, in the fossil pattern identification and classification method based on image recognition, the fossil pattern identification and classification method comprises the following steps:

[0005] Obtain three-dimensional point cloud data of the fossil sample, and project the three-dimensional point cloud data into multi-view two-dimensional texture mapping images, and obtain a surface element distribution map of the fossil sample;

[0006] The contrast, correlation and entropy value of the fossil texture in the multi-view two-dimensional texture mapping image are calculated by using an improved GLCM gray level co-occurrence matrix algorithm, the edge detail sensitivity is enhanced by introducing an LBP local binary pattern, and texture feature data are obtained;

[0007] The curvature, normal vector distribution and geometric parameters of the biological parts of the fossil in the three-dimensional point cloud data are calculated, and geometric feature data are obtained;

[0008] The surface element distribution map is analyzed by a PCA principal component analysis method, and a significant element combination related to the species is extracted, and element feature data are obtained;

[0009] The texture feature data, the geometric feature data and the element feature data are mapped to a unified feature space based on a cross-modal attention mechanism, the contribution degrees of the modes are distributed through learnable weights, and a fossil pattern classification result is output.

[0010] Further, in the fossil pattern identification and classification method based on image recognition, the three-dimensional point cloud data of the fossil sample are obtained, and the three-dimensional point cloud data are projected into multi-view two-dimensional texture mapping images, and a surface element distribution map of the fossil sample is obtained, which comprises:

[0011] A high-precision laser scanner is used to scan a dinosaur fossil sample in all directions to obtain original point cloud data, the original point cloud data is denoised to remove noise points caused by scanning environment interference and device errors, missing point cloud data is filled based on an interpolation algorithm of adjacent point cloud data, and three-dimensional point cloud data is obtained;

[0012] The processed three-dimensional point cloud data is projected at different angles to generate multi-angle two-dimensional texture mapping diagrams, including at least front view, side view and top view;

[0013] An X-ray fluorescence spectrometer is used to analyze the element composition of the fossil sample surface, and the element type and content data in each grid are collected in units of 1 centimeter, including at least Ca, P, Sr, Ti element concentration, and a surface element distribution map is obtained.

[0014] Further, in the above-mentioned image recognition-based fossil pattern recognition and classification method, the improved GLCM gray level co-occurrence matrix algorithm is used to calculate the contrast, correlation and entropy value of the fossil texture in the multi-angle two-dimensional texture mapping diagram, the LBP local binary pattern is introduced to enhance the edge detail sensitivity, and texture feature data is obtained, including:

[0015] A weighting mechanism is introduced based on the GLCM gray level co-occurrence matrix algorithm, different weights are given according to the importance of different regions in the image, and the features of key texture regions are highlighted to obtain an improved GLCM gray level co-occurrence matrix algorithm;

[0016] The multi-angle two-dimensional texture mapping diagram is converted into a gray image, the offset and gray level number of the gray level co-occurrence matrix are set, the contrast, correlation and entropy value are calculated, and feature data is obtained.

[0017] Further, in the above-mentioned image recognition-based fossil pattern recognition and classification method, the improved GLCM gray level co-occurrence matrix algorithm is used to calculate the contrast, correlation and entropy value of the fossil texture in the multi-angle two-dimensional texture mapping diagram, the LBP local binary pattern is introduced to enhance the edge detail sensitivity, and texture feature data is obtained, further including:

[0018] The LBP algorithm is applied to the grayed two-dimensional texture mapping diagram, and the sensitivity of image edge details is enhanced by setting different radius and neighborhood point number parameters;

[0019] A local binary pattern code is generated for each pixel in the image, LBP feature histograms are obtained by statistical analysis of the binary pattern codes, the LBP feature histograms are fused with the feature data, and texture feature data is obtained.

[0020] Further, in the above-mentioned fossil pattern recognition and classification method based on image recognition, the curvature, normal vector distribution and geometric parameters of the biological parts of the fossils in the three-dimensional point cloud data are calculated to obtain geometric feature data, including:

[0021] The local surface of the three-dimensional point cloud data is fitted based on the least square method to obtain the Gaussian curvature and average curvature of each point and the normal vector of the plane;

[0022] The calculated curvature and normal vector data are normalized to have a value range of [0, 1] to obtain normalized data;

[0023] Different biological parts are identified in the three-dimensional point cloud data, and the geometric parameters of each biological part are calculated using the measurement tool of the point cloud processing software to obtain geometric feature data.

[0024] Further, in the above-mentioned fossil pattern recognition and classification method based on image recognition, the surface element distribution map is analyzed by the PCA principal component analysis method to extract a significant element combination related to the species to obtain element feature data, including:

[0025] The surface element distribution map is standardized, the covariance matrix, eigenvalue and eigenvector of the data are calculated, and the principal components are selected according to the size of the eigenvalue;

[0026] The principal components with a cumulative contribution rate of 89% and a significant element combination related to the species are extracted to obtain element feature data.

[0027] Further, in the above-mentioned fossil pattern recognition and classification method based on image recognition, the texture feature data, geometric feature data and element feature data are mapped to a unified feature space based on the cross-modal attention mechanism, and the contribution of each modality is allocated through learnable weights, and the fossil pattern classification result is output, including:

[0028] Each modality data is mapped to a feature space of the same dimension through an MLP multi-layer perceptron; the fused feature data is processed using a softmax function to convert it into a probability distribution belonging to different dinosaur genera, and according to the size of the probability value, the genus with the largest probability is selected as the fossil pattern classification result.

[0029] Further, in a fossil pattern recognition and classification system based on image recognition, the fossil pattern recognition and classification system comprises the following modules:

[0030] A fossil data acquisition module for acquiring three-dimensional point cloud data of a fossil sample and projecting the three-dimensional point cloud data into a multi-view two-dimensional texture mapping map, and simultaneously acquiring a surface element distribution map of the fossil sample;

[0031] a texture feature extraction module configured to calculate the contrast, correlation and entropy of the fossil texture in the multi-view two-dimensional texture map by using an improved GLCM gray level co-occurrence matrix algorithm, and introduce an LBP local binary pattern to enhance the sensitivity of edge details, to obtain texture feature data;

[0032] a geometry feature extraction module configured to calculate the curvature, normal vector distribution and geometric parameters of biological parts of the fossil in the three-dimensional point cloud data, to obtain geometry feature data;

[0033] an element feature extraction module configured to analyze the surface element distribution map by using a PCA principal component analysis method, to extract a significant element combination related to the species, to obtain element feature data;

[0034] a fossil pattern classification module configured to map the texture feature data, the geometry feature data and the element feature data to a unified feature space based on a cross-modal attention mechanism, and to allocate the contribution of each modality by using a learnable weight, to output a fossil pattern classification result.

[0035] Further, in a fossil pattern recognition and classification system based on image recognition, the texture feature extraction module comprises the following sub-modules:

[0036] a weighting sub-module configured to introduce a weighting mechanism based on the GLCM gray level co-occurrence matrix algorithm, to assign different weights according to the importance of different regions in the image, to highlight the features of key texture regions, and to obtain an improved GLCM gray level co-occurrence matrix algorithm;

[0037] a calculation sub-module configured to convert the multi-view two-dimensional texture map into a gray image, to set the offset and the number of gray levels of the gray level co-occurrence matrix, to calculate the contrast, correlation and entropy, and to obtain feature data.

[0038] Further, in a fossil pattern recognition and classification system based on image recognition, the texture feature extraction module further comprises the following sub-modules:

[0039] an enhancement sub-module configured to apply the LBP algorithm to the two-dimensional texture map after gray scaling, to enhance the sensitivity of image edge details by setting different radius and neighborhood point number parameters;

[0040] a fusion sub-module configured to generate a local binary pattern code for each pixel in the image, to statistically analyze the binary pattern code to obtain an LBP feature histogram, to fuse the LBP feature histogram with the feature data, and to obtain texture feature data.

[0041] The beneficial effects are that the three-dimensional point cloud data of the fossil sample is obtained, and the three-dimensional point cloud data is projected as a multi-view two-dimensional texture mapping diagram, and a surface element distribution diagram of the fossil sample is obtained; the improved GLCM gray level co-occurrence matrix algorithm is used to calculate the contrast, correlation and entropy value of the fossil texture in the multi-view two-dimensional texture mapping diagram, the LBP local binary pattern is introduced to enhance the edge detail sensitivity, and texture feature data is obtained; the curvature, normal vector distribution and geometric parameters of the biological part of the fossil in the three-dimensional point cloud data are calculated, and geometric feature data is obtained; the surface element distribution diagram is analyzed by the PCA principal component analysis method, and a significant element combination related to the species is extracted, and element feature data is obtained; the texture feature data, the geometric feature data and the element feature data are mapped to a unified feature space based on the cross-modal attention mechanism, and the contribution of each mode is distributed through the learnable weight, and a fossil pattern classification result is output. 1. The texture feature reflects the texture details of the fossil surface, the geometric feature reflects the shape structure of the biological part, and the element feature reveals the material composition of the fossil, and the three are complementary. Based on the cross-modal attention mechanism, different modal data is mapped to a unified feature space, and a learnable weight is allocated, which can fully exert the advantages of each mode, significantly improve the accuracy of species identification, and reduce the risk of misjudgment caused by incomplete data. 2. Scientific curvature, normal vector calculation method and biological part geometric parameter measurement method are used to accurately quantify the three-dimensional morphological features of the fossil. In the aspect of element feature extraction, the PCA principal component analysis method can extract key information from complex element distribution data, and the application of these improved algorithms greatly enhances the extraction ability of fossil features. 3. Through programming and deep learning framework, various features are automatically calculated and classification is completed, reducing manual intervention. Compared with traditional manual identification, the research period is greatly shortened, the labor cost is reduced, and the subjective error and fatigue factors in manual operation are avoided, so that a large number of dinosaur fossil samples can be quickly and accurately classified. BRIEF DESCRIPTION OF DRAWINGS

[0042] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments with reference made to the accompanying drawings. The drawings are for purposes of illustration only and are not intended to be limiting of the application.

[0043] Figure 1 The first embodiment schematic diagram of the fossil pattern recognition and classification method based on image recognition in the embodiments of the application;

[0044] Figure 2 The second embodiment schematic diagram of the fossil pattern recognition and classification method based on image recognition in the embodiments of the application;

[0045] Figure 3Fig. 1 is a schematic diagram of a first embodiment of a fossil pattern recognition and classification system based on image recognition according to an embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0047] Those skilled in the art can understand that the singular forms "a", "an" and "the" used herein include plural forms unless specifically stated otherwise. It should be further understood that the use of the term "comprise" in the specification of the present application means that the features, integers, steps, operations, elements and / or components are present, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0048] The present application will be described in detail below with reference to the accompanying drawings, Figure 1 As shown in the drawings, a fossil pattern recognition and classification method based on image recognition includes the following steps:

[0049] Step 101, obtaining three-dimensional point cloud data of a fossil sample, and projecting the three-dimensional point cloud data into multi-view two-dimensional texture mapping images, and obtaining a surface element distribution map of the fossil sample;

[0050] Specifically, in the present embodiment, a high-precision laser scanner is used to scan a dinosaur fossil sample in all directions to obtain original point cloud data. The original point cloud data is denoised to remove noise points caused by scanning environment interference and equipment errors. The missing point cloud data is filled based on an interpolation algorithm of adjacent point cloud data to obtain three-dimensional point cloud data.

[0051] The processed three-dimensional point cloud data is projected at different viewing angles to generate multi-view two-dimensional texture mapping images, including at least front view, side view and top view.

[0052] An X-ray fluorescence spectrometer is used to analyze the element composition of the surface of the fossil sample. The element types and content data in each grid are collected in units of 1 centimeter, including at least Ca, P, Sr and Ti element concentrations, to obtain a surface element distribution map.

[0053] Specifically,

[0054] (I) three-dimensional point cloud data acquisition and multi-view projection

[0055] Data acquisition device

[0056] High-precision structured light 3D scanner (Artec Eva, accuracy up to 0.1 mm) or laser radar (Riegl VZ-400i, suitable for large fossils) is used to scan the whole surface of the dinosaur fossil sample. Before scanning, the surface of the fossil needs to be cleaned, and the high-reflective area is treated with non-reflective powder to ensure the integrity of data collection.

[0057] Small fossils (≤50 cm): Use a turntable scanner to collect one view every 15°, with a single sample scanning angle ≥24

[0058] Large fossils (>50 cm): Use zoned scanning, and integrate multi-region data through marker point stitching technology (Artec Studio automatic stitching function)

[0059] Point cloud processing flow

[0060] Scan data is imported into professional software (Geomagic Wrap) for processing:

[0061] De-noising: Use statistical filter (remove noise points outside 3 times the standard deviation of the distance neighborhood mean) and radius filter (delete isolated points with less than 5 neighborhood points)

[0062] Hole filling: Based on curvature interpolation algorithm (Curvature-based Fill), repair missing areas less than 5 cm², and mark damaged areas larger than 5 cm² for manual verification

[0063] Downsampling: Use voxel grid method (VoxelGrid, resolution 0.5-2 mm) to reduce data volume while preserving geometric features

[0064] Multi-view texture mapping generation

[0065] In Maya or Blender, establish a coordinate system with the center of the fossil as the origin, generate 6 basic viewing angles (XY / YZ / XZ plane positive / negative direction) according to the principle of orthogonal projection, and add ±15°, ±30° inclined viewing angles to each basic viewing angle, generating a total of 18 texture mapping images. During the mapping process, the RGB color information of the point cloud is extracted simultaneously to generate 8-bit depth PNG format texture images with a resolution of 2048x2048 pixels.

[0066] (2) Surface element distribution collection

[0067] The portable X-ray fluorescence spectrometer (Thermo Scientific Niton XL3t, detection accuracy ppm level) was used to scan the surface of the fossil row by row with a 2mm x 2mm spot diameter and a 5mm working distance. The scanning step was set to 1mm (lateral) x 2mm (vertical) to generate a grid element concentration matrix. After baseline correction and spectral peak fitting (based on the basic parameter method FP), the pseudo-color element distribution map was generated by ArcGIS Pro, with each pixel corresponding to 1mm2 of actual area, and the element value was normalized to 0-255 gray value (corresponding to 0-1000ppm concentration range).

[0068] Step 102, using the improved GLCM gray level co-occurrence matrix algorithm to calculate the contrast, correlation and entropy value of the texture of the fossil in the multi-view two-dimensional texture map, introducing LBP local binary pattern to enhance the sensitivity of edge details, and obtaining texture feature data;

[0069] Specifically, in this embodiment, a weighted mechanism is introduced based on the GLCM gray level co-occurrence matrix algorithm, different weights are given according to the importance of different regions in the image, and the features of key texture regions are highlighted, and an improved GLCM gray level co-occurrence matrix algorithm is obtained.

[0070] The multi-view two-dimensional texture map is converted into a gray scale image, the offset and gray scale number of the gray level co-occurrence matrix are set, the contrast, correlation and entropy value are calculated, and the feature data is obtained.

[0071] The LBP algorithm is applied to the gray-scale two-dimensional texture map, and by setting different radius and neighborhood point number parameters, the sensitivity of image edge details is enhanced.

[0072] Each pixel in the image generates a local binary pattern code, and the LBP feature histogram is obtained by statistical analysis of the binary pattern code. The LBP feature histogram is fused with the feature data to obtain the texture feature data.

[0073] Specifically,

[0074] (I) Implementation details of improved GLCM gray level co-occurrence matrix algorithm

[0075] 1. Pretreatment stage: image standardization and block strategy

[0076] Gray normalization: convert the multi-view texture map of 2048x2048 to an 8-bit gray scale image (0-255 value range), use adaptive histogram equalization (CLAHE) to enhance local contrast, and focus on improving the detail recognition of low texture regions (entropy value <3)

[0077] Dynamic block mechanism:

[0078] For smooth surface fossils (sauropod bones): 64x64 pixel base block, 30% overlap to reduce edge information loss

[0079] For high-texture fossils (ankylosaur scutes): 16x16 pixel subdivision block enabled, 70% overlap, combined with edge detection algorithm (Canny operator, threshold 0.3-0.7) to automatically identify texture-rich areas

[0080] Each texture map generates 4096 overlapping sub-blocks (2048 / 32-1=63, 63x63≈4096) to ensure full coverage of feature extraction

[0081] 2. Weighted GLCM feature calculation process

[0082] Dynamic allocation of direction weights:

[0083] Initialize weight matrix: default equal weight (0.25) for 0° (horizontal), 45°, 90° (vertical), and 135° main directions

[0084] Adaptive adjustment strategy: calculate sub-block texture entropy, when entropy > 5 (complex texture), enhance 45° / 135° direction weight (+0.15); when entropy < 3 (simple texture), strengthen 0° / 90° direction (+0.2)

[0085] Example: ankylosaur skin wrinkle texture (high entropy) 45° direction weight adjustment to 0.4, stegosaur bone plate smooth surface (low entropy) 0° direction weight increased to 0.45

[0086] Feature parameter fine calculation:

[0087] Contrast: focus on pixel pairs with gray difference ≥ 16, use weighted counting method (gray difference greater weight, formula: weight = gray difference / 255), highlight high-frequency textures such as deep grooves and protrusions

[0088] Correlation: introduce local mean normalization (sub-block gray - sub-block mean), calculate Pearson correlation coefficient with global texture, identify layered / striated periodic texture patterns

[0089] Entropy: use Shannon entropy formula (non-formal description: statistical sub-block gray distribution disorder, value greater texture more complex), set entropy > 6 sub-block as key feature area, automatically filter flat areas

[0090] 3. LBP edge detail enhancement implementation

[0091] Multi-scale LBP parameter configuration:

[0092] Base layer (R=1, P=8): Capture pixel-level edges (crack details), generate 8-dimensional raw LBP features

[0093] Enhanced layer (R=2, P=16): Cover 3x3 neighborhood, identify millimeter-level texture structure (scale edge), generate 16-dimensional rotation-invariant LBP features

[0094] Global layer (R=3, P=24): Capture centimeter-level texture patterns (bone plate arrangement), generate 24-dimensional uniform pattern LBP features

[0095] Feature concatenation: Concatenate three layers of features by channel to form a 48-dimensional edge-sensitive feature vector

[0096] Non-maximum suppression (NMS): Set gradient amplitude threshold (take the top 30% pixels), retain significant edge points (curvature mutation, texture boundary)

[0097] Binary processing: Otsu threshold segmentation of LBP feature map, generate edge mask (EdgeMask), mark texture boundary pixels (about 15%-25%)

[0098] Feature fusion: Multiply edge mask and GLCM features point by point to enhance texture feature contribution degree (weight increase 30%) in edge region

[0099] (B) Texture feature data integration scheme

[0100] Multi-view feature fusion:

[0101] Single-view processing: Extract 64x64 sub-block features from 18 texture maps respectively, generate 180-dimensional single-view global features through max pooling (retain the maximum value in the sub-block)

[0102] Cross-view fusion: Use attention mechanism weighted average (weight is dynamically allocated according to view entropy value, high entropy view weight x 1.2), finally form 3240-dimensional (180x18) texture feature vector

[0103] Quality control node:

[0104] Eliminate invalid views: Automatically filter views with entropy <2 (completely backlit surface), retain at least 12 valid views

[0105] Abnormal sub-block detection: Mark sub-blocks with contrast >200 or entropy >8 (may be noise area), manually check ≤5%

[0106] Step 103, calculate the curvature, normal vector distribution of the fossil in the three-dimensional point cloud data, and the geometric parameters of the biological part, to obtain the geometric feature data;

[0107] Specifically, in this embodiment, the LBP algorithm is applied to the gray-scale two-dimensional texture mapping graph, and by setting different radius and neighborhood point number parameters, the sensitivity of image edge details is enhanced.

[0108] Each pixel in the image is generated a local binary pattern code, and a statistical analysis is performed on the binary pattern code to obtain an LBP feature histogram. The LBP feature histogram is fused with feature data to obtain texture feature data.

[0109] Specifically,

[0110] (I) Surface geometry attribute calculation details

[0111] 1. Curvature and normal vector calculation process

[0112] Local surface fitting:

[0113] Point cloud preprocessing: K-neighbor algorithm (K=20) is used to construct the local neighborhood of each point to ensure uniform distribution of neighborhood points (deviation <15%)

[0114] Quadratic polynomial fitting: the fitting equation is \(z=ax²+by²+cxy+dx+ey+f\), and the coefficients are solved by least squares method, and the fitting error threshold is set to 0.3mm (adjustable according to the accuracy of the scanner)

[0115] Curvature parameter calculation:

[0116] Gaussian curvature (K):

[0117] Positive curvature (K>0): identify convex structures (joint head, tooth crown top), threshold K>0.005mm⁻² marked as strong convex region

[0118] Negative curvature (K<0): identify concave structures (joint socket, bone groove), threshold K<-0.003mm⁻² marked as deep concave region

[0119] Average curvature (H):

[0120] Absolute value >0.01mm⁻¹: defined as high curvature area (toe bone articular surface, rib arc surface)

[0121] Absolute value ≤0.01mm⁻¹: considered as flat area (long bone diaphysis, nail plate plane)

[0122] Normal vector estimation:

[0123] Covariance matrix method: calculate the covariance matrix of the local neighborhood point cloud, and take the eigenvector corresponding to the minimum eigenvalue as the normal vector

[0124] Direction consistency check: adjacent point normal vector angle > 20° is considered as edge point (skeletal fracture surface, weathering boundary), marked as geometric feature boundary point

[0125] 2. Biological site geometry parameter extraction specification

[0126] Anatomical landmark point positioning:

[0127] Establish a standard coordinate system: take a skull fossil as an example, set the front end of the nasal bone as the origin, the sagittal plane as the XZ plane, and the horizontal plane as the XY plane

[0128] Automatic recognition of key points: based on a deep learning model (PointNet++) to detect 21 anatomical landmark points (start of dentition, top of condyle), manual calibration error ≤ 1 mm

[0129] (2) Geometric feature data processing flow

[0130] Standardization and dimensionality reduction:

[0131] Z-score standardization: zero mean normalization of linear / angle parameters, formula description is (parameter value-mean) / standard deviation

[0132] Logarithmic transformation: curvature / vector data is compressed in dynamic range by ln(x+1) to ensure that the numerical value is distributed in the interval [-3, 3]

[0133] PCA dimensionality reduction: retain the first 5 principal components (cumulative variance > 95%), the physical meaning of the typical principal components is:

[0134] PC1: overall size feature (contribution degree 40%-50%)

[0135] PC2: surface curvature degree (contribution degree 20%-25%)

[0136] PC3: symmetry index (contribution degree 10%-15%)

[0137] Feature quality control:

[0138] Missing value processing: small range missing (<5%) uses KNN interpolation, large range missing sample is marked as to be supplemented

[0139] Outlier detection: based on DBSCAN algorithm to identify outliers (measurement error of joint angle), automatic labeling rate ≤ 3% Step 104, analyze the surface element distribution map by PCA principal component analysis method, extract the significant element combination related to the species, and obtain the element feature data.

[0140] Specifically, in the present embodiment, the LBP algorithm is applied to the gray-scale two-dimensional texture mapping graph, and by setting different radius and neighborhood point number parameters, the sensitivity of image edge details is enhanced;

[0141] A local binary pattern code is generated for each pixel in the image, a statistical analysis is performed on the binary pattern code to obtain an LBP feature histogram, and the LBP feature histogram is fused with feature data to obtain texture feature data.

[0142] Specifically,

[0143] PCA principal component analysis method application

[0144] The data of the surface element distribution map is imported into the data analysis library (NumPy and SciPy) of Python, and PCA principal component analysis method is used for dimension reduction and feature extraction of the data. First, the data is standardized to eliminate the influence of different element content data due to different dimensions. Then, the covariance matrix and eigenvalues and eigenvectors of the data are calculated, and the main components are selected according to the size of the eigenvalues (usually the main components with cumulative contribution rate of 80%-90% are selected). Through principal component analysis, the significant element combination related to the species is extracted to form element feature data.

[0145] Step 105, map the texture feature data, geometric feature data and element feature data to a unified feature space based on the cross-modal attention mechanism, and assign the contribution of each modality through learnable weights, and output the fossil pattern classification result.

[0146] Specifically, in this embodiment, the MLP multi-layer perceptron is used to map the modality data to the same dimension feature space; the fused feature data is processed by using the softmax function, and is converted into the probability distribution belonging to different dinosaur genera, and according to the size of the probability value, the genus with the largest probability is selected as the fossil pattern classification result.

[0147] Specifically,

[0148] (I) Cross-modal attention mechanism implementation

[0149] A network model based on cross-modal attention mechanism is constructed in a deep learning framework (TensorFlow or PyTorch). The texture feature data, geometric feature data and element feature data are respectively taken as the input of the model, and the multi-layer perceptron (MLP) is used to map the modality data to the same dimension feature space. The cross-modal attention module is introduced in the network, which assigns learnable weights to each modality data by learning the correlation between different modality data, so as to highlight the modality information with greater contribution to the classification result.

[0150] (II) Classification result output

[0151] At the output layer of the network model, the fused feature data is processed by a softmax function to convert it into a probability distribution belonging to different dinosaur genera. According to the size of the probability value, the genus with the maximum probability is selected as the final classification result of the fossil pattern, and the classification result and the corresponding probability value are output to evaluate the accuracy of the classification.

[0152] The beneficial effects are: 1. The texture feature reflects the texture details of the fossil surface, the geometric feature embodies the shape structure of the biological part, and the element feature reveals the material composition of the fossil, which are complementary to each other. Based on the cross-modal attention mechanism, different modal data are mapped to a unified feature space, and learnable weights are assigned, which can fully exert the advantages of each modality, significantly improve the accuracy of genus identification, and reduce the risk of misjudgment caused by incomplete data. 2. Scientific curvature, normal vector calculation method and biological part geometric parameter measurement method are used to accurately quantify the three-dimensional morphological features of the fossil. In the aspect of element feature extraction, the PCA principal component analysis method can extract key information from complex element distribution data. The application of these improved algorithms greatly enhances the ability to extract fossil features. 3. Through programming and deep learning framework, various features are automatically calculated and classification is completed, reducing manual intervention. Compared with traditional manual identification, the research period is greatly shortened, the labor cost is reduced, and the subjective error and fatigue factors in manual operation are avoided, so that a large number of dinosaur fossil samples can be quickly and accurately classified.

[0153] For reference Figure 2 In an image recognition-based fossil pattern identification and classification method, the curvature, normal vector distribution and geometric parameters of biological parts of the fossil in the three-dimensional point cloud data are calculated to obtain geometric feature data including the following steps:

[0154] Step 201, fitting the local surface of the three-dimensional point cloud data based on the least square method to obtain the Gaussian curvature and average curvature of each point, and obtaining the normal vector of the plane;

[0155] Step 202, normalizing the calculated curvature and normal vector data to unify the value range to [0, 1] to obtain normalized data;

[0156] Step 203, identifying different biological parts in the three-dimensional point cloud data, calculating the geometric parameters of each biological part using the measurement tool of the point cloud processing software to obtain geometric feature data.

[0157] Specifically,

[0158] (I) curvature and normal vector calculation

[0159] In the three-dimensional point cloud processing software, the curvature and normal vector of each point in the three-dimensional point cloud data are calculated by using the built-in geometric calculation module. For curvature calculation, a method based on least squares fitting of local surface can be used to obtain the Gaussian curvature and mean curvature of each point. The calculation of the normal vector is obtained by fitting the local plane of the point cloud data, and the normal vector of the plane is taken as the normal vector of the point. The calculated curvature and normal vector data are normalized to make the value range uniform between [0, 1], which is convenient for subsequent analysis and comparison.

[0160] (II) Calculation of biological part geometric parameters

[0161] According to the anatomical knowledge of dinosaur fossils, different biological parts such as bones and joints are identified in the three-dimensional point cloud data. The geometric parameters such as length, width, thickness and angle of each biological part are calculated by using the measurement tool of the point cloud processing software. For irregularly shaped biological parts, the volume parameter can be estimated by using the method based on point cloud volume calculation. These geometric parameters are sorted to form geometric feature data.

[0162] The above describes an embodiment of the fossil pattern recognition and classification method based on image recognition. Please refer to Figure 3 In a fossil pattern recognition and classification system based on image recognition, the fossil pattern recognition and classification system includes the following modules:

[0163] The fossil data acquisition module is used to acquire the three-dimensional point cloud data of the fossil sample, and project the three-dimensional point cloud data into multi-view two-dimensional texture mapping images, and acquire the surface element distribution map of the fossil sample;

[0164] The texture feature extraction module is used to calculate the contrast, correlation and entropy value of the fossil texture in the multi-view two-dimensional texture mapping image by using the improved GLCM gray level co-occurrence matrix algorithm, and introduce the LBP local binary pattern to enhance the edge detail sensitivity, to obtain the texture feature data;

[0165] The geometric feature extraction module is used to calculate the curvature, normal vector distribution of the fossil and the geometric parameters of the biological part in the three-dimensional point cloud data, to obtain the geometric feature data;

[0166] The element feature extraction module is used to analyze the surface element distribution map by PCA principal component analysis method, extract the significant element combination related to the species, and obtain the element feature data;

[0167] The fossil pattern classification module is used to map the texture feature data, geometric feature data and element feature data to a unified feature space based on the cross-modal attention mechanism, and allocate the contribution degree of each modality through learnable weight, to output the fossil pattern classification result.

[0168] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Various changes and improvements can be made to the present application without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A fossil pattern recognition and classification method based on image recognition, characterized in that, The fossil pattern identification and classification method includes the following steps: A high-precision laser scanner was used to perform a 3D point cloud scan on the dinosaur fossil sample to obtain raw point cloud data. The raw point cloud data was then denoised to remove noise caused by scanning environment interference and equipment errors. Missing point cloud data was filled in based on an interpolation algorithm of adjacent point cloud data to obtain 3D point cloud data. The processed 3D point cloud data was then projected from different perspectives to generate a multi-view 2D texture map, including at least a front view, a side view, and a top view. The elemental composition of the fossil sample surface was analyzed using an X-ray fluorescence spectrometer. The element types and contents in each grid were collected in 1 cm grid units, including at least the concentrations of Ca, P, Sr, and Ti, to obtain a surface elemental distribution map. The contrast, correlation, and entropy of fossil textures in the multi-view two-dimensional texture map are calculated using the improved GLCM gray-level co-occurrence matrix algorithm. The LBP local binary mode is introduced to enhance the sensitivity of edge details, and texture feature data is obtained. The local surface of the three-dimensional point cloud data is fitted using the least squares method to obtain the Gaussian curvature and average curvature of each point, and the normal vector of the plane is obtained. The calculated curvature and normal vector data are normalized so that their values ​​are uniformly between [0,1] to obtain normalized data. Different biological parts are identified in the three-dimensional point cloud data, and the geometric parameters of each biological part are calculated using the measurement tools of the point cloud processing software to obtain geometric feature data. The surface elemental distribution map was analyzed by PCA principal component analysis to extract significant elemental combinations related to species and obtain elemental characteristic data. Based on the cross-modal attention mechanism, texture feature data, geometric feature data, and element feature data are mapped to a unified feature space, and the contribution of each modality is assigned through learnable weights to output the fossil pattern classification results.

2. The fossil pattern recognition and classification method based on image recognition as described in claim 1, characterized in that, The improved GLCM gray-level co-occurrence matrix algorithm is used to calculate the contrast, correlation, and entropy of the fossil texture in the multi-view two-dimensional texture map. LBP local binary mode is introduced to enhance edge detail sensitivity, resulting in texture feature data, including: An improved GLCM gray-level co-occurrence matrix algorithm is obtained by introducing a weighting mechanism based on the importance of different regions in the image, highlighting the features of key texture regions. The multi-view two-dimensional texture map is converted into a grayscale image. The offset of the gray-level co-occurrence matrix and the number of gray levels are set. The contrast, correlation and entropy values ​​are calculated to obtain the feature data.

3. The fossil pattern recognition and classification method based on image recognition as described in claim 1, characterized in that, The method of calculating the contrast, correlation, and entropy of fossil textures in the multi-view two-dimensional texture map using the improved GLCM gray-level co-occurrence matrix algorithm, introducing LBP local binary mode to enhance edge detail sensitivity, and obtaining texture feature data also includes: The LBP algorithm is applied to the grayscale 2D texture map, and the sensitivity of image edge details is enhanced by setting different radius and neighborhood point parameters. A local binary pattern code is generated for each pixel in the image. Statistical analysis is performed on the binary pattern code to obtain an LBP feature histogram. The LBP feature histogram is then fused with the feature data to obtain texture feature data.

4. The fossil pattern recognition and classification method based on image recognition as described in claim 1, characterized in that, The analysis of the surface elemental distribution map using PCA principal component analysis extracts significant elemental combinations related to the species, yielding elemental characteristic data, including: The surface element distribution map is standardized, the covariance matrix, eigenvalues ​​and eigenvectors of the data are calculated, and principal components are selected according to the magnitude of the eigenvalues. Significant elemental combinations related to species were extracted from principal components with a cumulative contribution rate of 89%, and elemental characteristic data were obtained.

5. The fossil pattern recognition and classification method based on image recognition as described in claim 1, characterized in that, The cross-modal attention mechanism maps texture feature data, geometric feature data, and element feature data to a unified feature space, and assigns the contribution of each modality through learnable weights, outputting fossil pattern classification results, including: The MLP (Multilayer Perceptron) is used to map the data of each modality to a feature space of the same dimension. The softmax function is used to process the fused feature data and convert it into a probability distribution of different dinosaur genera and species. Based on the probability value, the genera and species with the highest probability are selected as the fossil pattern classification result.

6. A fossil pattern recognition and classification system based on image recognition, characterized in that, The fossil pattern recognition and classification system includes the following steps: The fossil data acquisition module is used to perform omnidirectional scanning of dinosaur fossil samples using a high-precision laser scanner to obtain raw point cloud data. The raw point cloud data is then denoised to remove noise caused by scanning environment interference and equipment errors. Missing point cloud data is filled in using an interpolation algorithm based on adjacent point cloud data to obtain three-dimensional point cloud data. The processed three-dimensional point cloud data is projected from different perspectives to generate a multi-view two-dimensional texture map, including at least a front view, a side view, and a top view. An X-ray fluorescence spectrometer is used to analyze the elemental composition of the fossil sample surface. Using a 1-centimeter grid, the type and content of elements within each grid are collected, including at least the concentrations of Ca, P, Sr, and Ti, to obtain a surface elemental distribution map. The texture feature extraction module is used to calculate the contrast, correlation and entropy of fossil textures in the multi-view two-dimensional texture map using the improved GLCM gray-level co-occurrence matrix algorithm, and introduces LBP local binary mode to enhance edge detail sensitivity to obtain texture feature data. The geometric feature extraction module is used to fit the local surface of the three-dimensional point cloud data based on the least squares method to obtain the Gaussian curvature and average curvature of each point, and to obtain the normal vector of the plane; the calculated curvature and normal vector data are normalized to unify their values ​​between [0,1] to obtain normalized data; different biological parts are identified in the three-dimensional point cloud data, and the geometric parameters of each biological part are calculated using the measurement tools of the point cloud processing software to obtain geometric feature data; The element feature extraction module is used to analyze the surface element distribution map using PCA principal component analysis, extract significant element combinations related to species, and obtain element feature data. The fossil pattern classification module is used to map texture feature data, geometric feature data, and element feature data to a unified feature space based on a cross-modal attention mechanism, and to allocate the contribution of each modality through learnable weights, and output the fossil pattern classification results.

7. A fossil pattern recognition and classification system based on image recognition as described in claim 6, characterized in that, The texture feature extraction module includes the following sub-modules: The weighted submodule is used to introduce a weighting mechanism based on the GLCM gray-level co-occurrence matrix algorithm. Different weights are assigned according to the importance of different regions in the image, highlighting the features of key texture regions, and thus obtaining an improved GLCM gray-level co-occurrence matrix algorithm. The calculation submodule is used to convert the multi-view two-dimensional texture map into a grayscale image, set the offset and grayscale level of the grayscale co-occurrence matrix, calculate the contrast, correlation and entropy values, and obtain feature data.

8. A fossil pattern recognition and classification system based on image recognition as described in claim 6, characterized in that, The texture feature extraction module also includes the following sub-modules: The enhancement submodule applies the LBP algorithm to the grayscale two-dimensional texture map, enhancing the sensitivity of image edge details by setting different radius and neighborhood point parameters. The fusion submodule generates a local binary pattern code for each pixel in the image, performs statistical analysis on the binary pattern code to obtain an LBP feature histogram, and fuses the LBP feature histogram with the feature data to obtain texture feature data.

Images