Microfossil genus and species intelligent identification method and device and computer equipment

Abstract

The application provides a microfossil genus and species intelligent identification method and device and computer equipment, and belongs to the technical field of microfossil genus and species identification. The method comprises the following steps: acquiring images of microfossils at different shooting angles; inputting the images into a genus and species identification model to identify the genus and species of the microfossils. The genus and species identification model is used for respectively extracting features from the images of the microfossils at different shooting angles, fusing the extracted features according to a preset fusion rule, and classifying the genus and species by using the fused features. The fusion rule comprises at least one of fusion of first convolutional deep features extracted from the images of the microfossils at different shooting angles and fusion of second convolutional deep features extracted from the images of the microfossils at different shooting angles. The above method extracts features from the images of the microfossils at different shooting angles by using the genus and species identification model and classifies the features, and before classification, the features are fused according to the fusion rule, so that high-accuracy genus and species intelligent identification is realized.

Description

Technical Field

[0001] This application belongs to the field of microfossil genera and species identification technology, specifically relating to a method for intelligent identification of microfossil genera and species, a device for intelligent identification of microfossil genera and species, a computer device, and a machine-readable storage medium. Background Technology

[0002] Microfossils refer to tiny fossils of ancient organisms, invisible to the naked eye, preserved in sedimentary strata from various geological eras. Microfossils are mainly divided into skeletal fossils and trace fossils. Conodonts, an important category of microfossils, also known as conodonts, are formerly called conodontids, conodonts, conical tusks, and conical tusks. They are partial remains of ancient, unknown organisms, often appearing as sharp or serrated teeth. Traditional or conventional methods for identifying conodonts (spines) are almost never mentioned in textbooks; related content is usually presented as "classification of conodonts (spines)" (e.g., Ding Meihua, 1983; Jiang Wu et al., 1986; Wang Chengyuan, 1987). This "classification of conodonts (spines)" is the foundation and basis of identification, and must be followed by any identification method. Traditional identification methods rely on researchers using "classification standards" as guidance, observing the characteristics of fossil specimens under a microscope, comparing them with published illustrations, and making subjective judgments based on their professional expertise. Therefore, identification is a highly specialized task. Practitioners must possess sufficient professional skills and a certain amount of experience. The quality of identification is closely related to the professional level of the examiner. Identification is also a time-consuming "slow study." A small specimen may take a whole day or even several days to complete. Therefore, traditional identification methods suffer from a series of drawbacks, including low efficiency, high labor costs, and errors due to subjective human factors. Summary of the Invention

[0003] The purpose of this application is to provide a method, device, computer equipment, and machine-readable storage medium for intelligent identification of microfossil genera and species, in order to overcome the shortcomings of existing technologies, such as low efficiency, high labor costs, and low accuracy of identification results due to subjective human factors, in the manual identification of microfossils such as conodonts.

[0004] To achieve the above objectives, the first aspect of this application provides a method for intelligent identification of microfossil genera and species, comprising: Acquire images of microfossils from different shooting angles; Images of the microfossils taken from different angles are input into a pre-built genus and species identification model to identify the genus and species of the microfossils. The genus and species identification model is obtained by training and optimizing the parameters of an initial convolutional neural network. The genus and species identification model is used to extract features from images of microfossils taken from different angles and to fuse the extracted features according to a preset fusion rule, and to classify the genus and species using the fused features. The fusion rule includes at least one of the fusion of a first convolutional depth feature extracted from images of microfossils taken from different angles and the fusion of a second convolutional depth feature extracted from images of microfossils taken from different angles, wherein the convolutional layer depth used to extract the second convolutional depth feature is greater than the convolutional layer depth used to extract the first convolutional depth feature.

[0005] In a specific embodiment of this application, the genus and species identification model includes: Multiple first convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one by one, to obtain each first feature map; The first fusion layer is used to fuse the various first feature maps to obtain the second feature map; The second convolutional network is used to extract the second convolutional depth features of the second feature map to obtain the third feature map; The first classification network is used to classify microfossil genera and species using the third feature map.

[0006] In a specific embodiment of this application, the genus and species identification model includes: Multiple third convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one by one, to obtain the fourth feature maps. Multiple fourth convolutional networks are used to extract the second convolutional depth features of each fourth feature map in a one-to-one correspondence, so as to obtain each fifth feature map; The second fusion layer is used to fuse the various fifth feature maps to obtain the sixth feature map; The second classification network is used to classify microfossil genera and species using the sixth feature map.

[0007] In a specific embodiment of this application, the genus and species identification model includes: Multiple fifth convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain various seventh feature maps; Multiple third fusion layers are used to fuse the seventh feature maps in each group in a one-to-one correspondence to obtain the eighth feature maps that correspond one-to-one with each group; Multiple sixth convolutional networks are used to extract the second convolutional depth features of each eighth feature map and each remaining seventh feature map that has not been assigned to any group, in a one-to-one correspondence, to obtain each ninth feature map. The fourth fusion layer is used to fuse the various ninth feature maps to obtain the tenth feature map; The third classification network is used to classify microfossil genera and species using the tenth feature map; The grouping rules for the seventh feature image include: if the shooting angle differences of the microfossil images corresponding to multiple seventh feature images are within a first preset range, then the multiple seventh feature images are grouped into the same group.

[0008] In a specific embodiment of this application, the genus and species identification model includes: Multiple seventh convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain the eleventh feature maps; Multiple fifth fusion layers are used to fuse the eleventh feature maps in each group in a one-to-one correspondence, to obtain the twelfth feature maps that correspond one-to-one with each group; The sixth fusion layer is used to fuse the various twelfth feature maps to obtain the thirteenth feature map; The eighth convolutional network is used to extract the second convolutional depth features of the thirteenth feature map to obtain the fourteenth feature map; The fourth classification network is used to classify microfossil genera and species using the fourteenth feature map; The grouping rules for the eleventh feature map include: If the shooting angle differences of the microfossil images corresponding to multiple eleventh feature images are within a second preset range, then the multiple eleventh feature images are grouped into the same group.

[0009] In a specific embodiment of this application, the genus and species identification model includes: Multiple ninth convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain the fifteenth feature maps; Multiple tenth convolutional networks are used to extract the second convolutional depth features of each fifteenth feature map in a one-to-one correspondence, thus obtaining each sixteenth feature map; Multiple seventh fusion layers are used to fuse the sixteenth feature maps in each group in a one-to-one correspondence to obtain the seventeenth feature maps that correspond one-to-one with each group. The eighth fusion layer is used to fuse the various seventeenth feature maps to obtain the eighteenth feature map; The fifth classification network is used to classify genera and species of microfossils using the eighteenth feature map; The grouping rules for the sixteenth feature map include: If the shooting angle differences of the microfossil images corresponding to multiple sixteenth feature images are within the third preset range, then the multiple sixteenth feature images are grouped into the same group.

[0010] In a specific embodiment of this application, the first fusion layer has a convolutional kernel size of One of the following: convolutional layer, max pooling layer, or average pooling layer.

[0011] In a specific embodiment of this application, the second fusion layer is one of a fully connected layer, a max pooling layer, and an average pooling layer.

[0012] In a specific embodiment of this application, the third fusion layer has a convolutional kernel size of The fourth fusion layer is one of a convolutional layer, a max pooling layer, and an average pooling layer, and is a fully connected layer, a max pooling layer, and an average pooling layer.

[0013] In a specific embodiment of this application, both the fifth and sixth fusion layers have convolutional kernel sizes of [size missing]. One of the following: convolutional layer, max pooling layer, or average pooling layer.

[0014] In specific embodiments of this application, the seventh fusion layer and the eighth fusion layer are both one of a fully connected layer, a max pooling layer, and an average pooling layer.

[0015] In specific embodiments of this application, the images of the microfossils taken from different angles include top-view images, bottom-view images, and side-view images.

[0016] In a specific embodiment of this application, a genus and species identification model is obtained after training and parameter optimization of an initial convolutional neural network, including: The weights of the pre-trained model are used as the initial weights of the convolutional layers in the initial convolutional neural network used to extract the first convolutional depth features and the convolutional layers used to extract the second convolutional depth features, as well as the weights of the classification network in the initial convolutional neural network used for the classification of microfossil genera and species. Train the current convolutional neural network using the constructed training samples; The training results are evaluated using a preset loss function. If the training results meet the preset requirements, the next step is executed. Otherwise, the weight values ​​of each convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for the classification of microfossil genera and species are updated. The absolute value of the weight difference before and after the update of the convolutional layer used to extract the first convolutional depth features is less than the absolute value of the weight difference before and after the update of the convolutional layer used to extract the second convolutional depth features, and the absolute value of the weight difference before and after the update of the convolutional layer used to extract the second convolutional depth features is less than the absolute value of the weight difference before and after the update of the classification network used for the classification of microfossil genera and species are executed. Then, the process jumps to the previous step. The latest weight values ​​are assigned to the convolutional neural network to obtain the species identification model.

[0017] In a specific embodiment of this application, updating the weight values ​​within the convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for classifying microfossil genera and species includes: The first formula is used to update the weight values ​​of the convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for the classification of microfossil genera and species. The first formula is expressed as: , This indicates the updated weights. This indicates the weights before the update. Indicates the learning rate. Represents the loss function In weight The rate of change in direction, used to extract the first convolutional depth features during weight updates. The value is less than that of the convolutional layer used to extract the second convolutional depth feature during weight update. The value is used to extract the second convolutional depth feature during weight updates. The value is less than that of the classification network used for the classification of microfossil genera and species when updating weights. Take values, and .

[0018] In a specific embodiment of this application, the weights of each convolutional layer used to extract the depth features of images taken from different shooting angles of microfossils are updated. Different values ​​are used to update the weights of each convolutional layer in the second convolutional depth feature extraction process for images taken from different angles of microfossils. The values ​​are different.

[0019] A second aspect of this application provides a smart identification device for microfossil genera and species, comprising: The image acquisition module is used to acquire images of microfossils from different shooting angles; The genus and species identification module is used to input images of the acquired microfossils from different shooting angles into a pre-constructed genus and species identification model to identify the genus and species of the microfossils. The genus and species identification model is obtained by training and optimizing the parameters of an initial convolutional neural network. The genus and species identification model is used to extract features from images of microfossils taken from different angles and to fuse the extracted features according to a preset fusion rule, and to classify the genus and species using the fused features. The fusion rule includes at least one of the fusion of a first convolutional depth feature extracted from images of microfossils taken from different angles and the fusion of a second convolutional depth feature extracted from images of microfossils taken from different angles, wherein the convolutional layer depth used to extract the second convolutional depth feature is greater than the convolutional layer depth used to extract the first convolutional depth feature.

[0020] A third aspect of this application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the intelligent identification method for microfossil genera and species described in the first aspect of this application.

[0021] A fourth aspect of this application is a machine-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent identification method for microfossil genera and species described in the first aspect of this application.

[0022] In the above technical solution, a species identification model based on convolutional neural networks is used to extract features and classify images of microfossils taken from different angles. Before classification, the extracted features are fused according to fusion rules. The fusion rules include at least one of the fusion of first convolutional depth features extracted from images of microfossils taken from different angles and the fusion of second convolutional depth features extracted from images of microfossils taken from different angles. The convolutional layer depth used to extract the second convolutional depth features is greater than the convolutional layer depth used to extract the first convolutional depth features. This achieves intelligent identification of microfossil species, improves identification efficiency and accuracy, and saves labor costs.

[0023] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0024] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings: Figure 1 The schematic diagram illustrates a flowchart of a method for intelligent identification of microfossil genera and species according to an embodiment of this application; Figure 2 This schematically illustrates a first network structure diagram of a genus and species identification model according to an embodiment of this application; Figure 3 This schematically illustrates a second network structure diagram of the genus and species identification model according to an embodiment of this application; Figure 4 This schematically illustrates a third network structure diagram of the genus and species identification model according to an embodiment of this application; Figure 5 This schematically illustrates a fourth network structure diagram of the genus and species identification model according to an embodiment of this application; Figure 6 This schematically illustrates a fifth network structure diagram of the genus and species identification model according to an embodiment of this application; Figure 7 The diagram illustrates the network structure of a genus-species identification model in a specific application example. Figure 8 A schematic block diagram of a computer device according to an embodiment of this application is shown. Detailed Implementation

[0025] The specific embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the embodiments of this application.

[0026] If the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0027] Figure 1 A schematic flowchart illustrating the intelligent identification method for microfossil genera and species according to an embodiment of this application is shown. Figure 1 As shown, the intelligent identification method for microfossil genera and species provided in this application embodiment may include the following steps: Step 102: Obtain images of the microfossils from different shooting angles.

[0028] In this application, images of microfossils can be captured using stereomicroscopes and scanning electron microscopes. Images taken from different angles can be top-down, bottom-up, and side-view images, among which side-view images can be taken from the front, rear, left, and right sides. In particular, for some microfossils that do not have obvious top, bottom, and side surfaces, any shooting angle is acceptable.

[0029] Step 104: Input the images of the microfossils taken from different angles into a pre-constructed genus and species identification model to identify the genus and species of the microfossils. The genus and species identification model is obtained by training and optimizing the parameters of an initial convolutional neural network. The model is used to extract features from the images of the microfossils taken from different angles and to fuse the extracted features according to a preset fusion rule. The fused features are then used for genus and species classification. The fusion rule includes at least one of the fusion of a first convolutional depth feature extracted from the images of the microfossils taken from different angles and the fusion of a second convolutional depth feature extracted from the images of the microfossils taken from different angles. The convolutional layer depth used to extract the second convolutional depth feature is greater than the convolutional layer depth used to extract the first convolutional depth feature.

[0030] In this application, the network used for feature extraction in the genus and species identification model can be a feature extraction network such as RestNet, EfficientNet, MobileNet, or ViT, and the classification network used for genus and species classification in the genus and species identification model can be a multi-classification model such as a fully connected layer, support vector machine, or linear model.

[0031] In one comparative embodiment, a single microfossil image is input into a pre-constructed convolutional neural network model for the identification of microfossil genera and species. Because microfossils are three-dimensional structures, a single-angle image of a microfossil does not fully reflect the shape, outline, details, etc., which can lead to identification errors and thus result in incorrect identification of microfossil genera and species. It is evident that the accuracy of genera and species classification using a single microfossil image is poor. For example, Chinese patent application No. 202110253901.1 discloses a method, system, and application for detecting, classifying, and discovering microfossil images. The proposed method includes: establishing standards for microfossil image acquisition and capturing microfossil images; constructing a dataset with simulated microfossils; building an SSD network; adjusting the aspect ratio of pre-selected bounding boxes; loading the original weight file of the pre-trained model and training the network model for microfossil image detection; inputting the image to be detected into the trained network model and using a non-maximum suppression algorithm to select suitable detection results; and highlighting and recording the detection results of artificially simulated fossils that differ from the original known microfossils. While the above-disclosed method utilizes an SSD network to automatically filter and classify microfossils, enabling the discovery of new fossil species, the image to be detected input into the trained network model is a single image. The features extracted from this single image within the network model are not rich enough, resulting in insufficient accuracy in classification using these extracted features.

[0032] In the above embodiments of this application, feature extraction is performed using microfossil images taken from multiple angles. In order to make full use of the rich features extracted from images taken from different angles of microfossils when classifying genera and species, feature fusion is performed based on the fusion of single convolutional depth features or the fusion of multiple convolutional depth features before classifying genera and species. This improves the accuracy of genera and species classification by utilizing the rich fused features, thereby improving the accuracy of microfossil genera and species identification.

[0033] In one specific embodiment of this application, the first convolutional depth feature is a shallow, low-level feature, and the second convolutional depth feature is a deep, high-level feature. Accordingly, the convolutional layer used to extract the first convolutional depth feature is a shallow convolutional layer, and the convolutional layer used to extract the second convolutional depth feature is a deep convolutional layer. The convolutional depths of the shallow and deep convolutional layers are determined according to specific application requirements.

[0034] In one specific embodiment of this application, the genus and species identification model obtained after training and parameter optimization of an initial convolutional neural network may include the following steps: The weights of the pre-trained model are used as the initial weights of the convolutional layers used to extract the first convolutional depth features and the convolutional layers used to extract the second convolutional depth features within the initial convolutional neural network, as well as the weights of the classification network used for the classification of microfossil genera and species within the initial convolutional neural network. Train the current convolutional neural network using the constructed training samples; The training results are evaluated using a preset loss function. If the training results meet the preset requirements, the next step is executed. Otherwise, the weight values ​​of each convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for the classification of microfossil genera and species are updated. The absolute value of the weight difference before and after the update of the convolutional layer used to extract the first convolutional depth features is less than the absolute value of the weight difference before and after the update of the convolutional layer used to extract the second convolutional depth features, and the absolute value of the weight difference before and after the update of the convolutional layer used to extract the second convolutional depth features is less than the absolute value of the weight difference before and after the update of the classification network used for the classification of microfossil genera and species are executed. Then, the process jumps to the previous step. The latest weight values ​​are assigned to the convolutional neural network to obtain the species identification model.

[0035] For example, when updating weights using the gradient descent method, updating the weight values ​​within the convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for classifying microfossil genera and species may include the following steps: The first formula is used to update the weight values ​​of the convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for the classification of microfossil genera and species. The first formula is expressed as: , This indicates the updated weights. This indicates the weights before the update. Indicates the learning rate. Represents the loss function In weight The rate of change in direction, used to extract the first convolutional depth features during weight updates. The value is less than that of the convolutional layer used to extract the second convolutional depth features during weight updates. The value is used to update the weights of the convolutional layer that extracts the second convolutional depth features. The value is less than that of the classification network used for the classification of microfossil genera and species when updating weights. Take values, and . In a comparative embodiment, a species identification model is obtained by training and optimizing the parameters of an initial convolutional neural network based on transfer learning, which may include the following steps: The weights of the pre-trained model are used as the initial weights of the convolutional layers used to extract the first convolutional depth features and the convolutional layers used to extract the second convolutional depth features within the initial convolutional neural network, as well as the weights of the classification network used for the classification of microfossil genera and species within the initial convolutional neural network. The training results are evaluated using a preset loss function. If the training results meet the preset requirements, the next step is executed. Otherwise, the weight values ​​of each convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for the classification of microfossil genera and species are updated using the first formula, and the process jumps to the previous step. The latest weight values ​​are assigned to the convolutional neural network to obtain the genus and species identification model. The first formula is expressed as: ,in, This indicates the updated weights. This indicates the weights before the update. Indicates the learning rate. Represents the loss function In weight The rate of change in direction, used in the classification network for the classification of microfossil genera and species during weight updates. A value of 1 indicates that the weights of layers other than the classification network within the convolutional neural network are updated accordingly. All values ​​are 0.

[0036] In the comparative embodiments described above, only the classification network at the end of the convolutional neural network is retrained, while all other layers within the convolutional neural network are frozen. Although this improves training efficiency through transfer learning, pre-trained models are typically trained and optimized using a single image as input. The weights obtained from pre-training cannot adequately meet the feature extraction requirements of images of microfossils taken from different angles. In the comparative embodiments described above, freezing the weights within the convolutional layers used to extract the first and second convolutional depth features during training as the weights of the pre-trained model can lead to poor transfer learning performance and may also increase training time. In the embodiments of this application, when updating the weights using the gradient descent method, the shallow layers of the convolutional neural network... The value is smaller than that of the deeper layers of the convolutional neural network. The method involves freezing the network weights based on depth optimization, gradually unfreezing them as the depth of the convolutional neural network increases. While increasing the depth of the convolutional network expands the field of view of the convolutional kernels, it also leads to a decline in training performance. If, as in the comparative embodiment described above, the weights of all layers except the classification network are frozen to pre-trained weights, it is difficult to achieve good training results. However, by freezing the network weights based on depth optimization, this deficiency can be overcome, resulting in better training performance and thereby improving the recognition accuracy of the constructed species identification model.

[0037] For example, as a preferred embodiment, the weight updates of each convolutional layer used to extract the depth features of images taken from different angles of microfossils are performed as follows: Different values ​​are used to update the weights of each convolutional layer in the second convolutional depth feature extraction process for images taken from different angles of microfossils. The values ​​differ, meaning that as the network gradually unfreezes with increasing depth, the weights of each convolutional layer at the same depth are updated using different values. The values ​​are different. Because when extracting the first and second convolutional depth features from microfossil images taken from different angles, under non-ideal conditions, the weight values ​​of the convolutional layers corresponding to microfossil images taken from different angles should be different; undifferentiated settings will lead to poor training results. Based on this, in the preferred embodiment above, based on... The different values ​​are used to reflect the differences in weight values ​​within each convolutional layer of the same convolutional depth, thereby achieving better training results and improving the recognition accuracy of the constructed species identification model.

[0038] Figure 2 A schematic diagram illustrating a first network structure of a genus and species identification model according to an embodiment of this application is shown. For example... Figure 2 As shown, the genus and species identification model includes multiple first convolutional networks, a first fusion layer, a second convolutional network, and a first classification network, wherein: Multiple first convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one by one, to obtain each first feature map; The first fusion layer is used to fuse the various first feature maps to obtain the second feature map; The second convolutional network is used to extract the second convolutional depth features from the second feature map to obtain the third feature map; The first classification network is used to classify microfossil genera and species using the third feature map.

[0039] In this application, the first fusion layer may be a convolutional kernel with a size of It is one of the following: convolutional layer, max pooling layer, or average pooling layer. Here, the max pooling layer performs max pooling in the channel direction, and the average pooling layer performs average pooling in the channel direction.

[0040] In the above embodiment, feature fusion is performed before the second convolutional depth features are extracted using the second convolutional network, which is a feature "early fusion" method.

[0041] For example, suppose that images of microfossils taken from different angles all generate feature map tensors of dimension (H, W, C) after passing through the first convolutional network. These feature maps are then concatenated along the channel directions to obtain (H, W, C) The feature map of dimension (H, W, C) is obtained after the fusion operation of the first fusion layer. The feature map of dimension (H, W, C) is then fed into the second convolutional network and the subsequent first classification network.

[0042] Figure 3 A second network structure diagram of the genus and species identification model according to an embodiment of this application is illustrated schematically. For example... Figure 3 As shown, the genus and species identification model includes multiple third convolutional networks, multiple fourth convolutional networks, a second fusion layer, and a second classification network, wherein: Multiple third convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one by one, to obtain the fourth feature maps. Multiple fourth convolutional networks are used to extract the second convolutional depth features of each fourth feature map in a one-to-one correspondence, so as to obtain each fifth feature map; The second fusion layer is used to fuse the various fifth feature maps to obtain the sixth feature map; The second classification network is used to classify microfossil genera and species using the sixth feature map.

[0043] In this application, the second fusion layer can be one of a fully connected layer, a max pooling layer, and an average pooling layer. Here, the max pooling performed by the max pooling layer is max pooling in the K-direction, and the average pooling performed by the average pooling layer is average pooling in the K-direction, where the K-direction refers to the direction with the shooting angle as the dimension.

[0044] In the above embodiment, feature fusion is performed after extracting the second convolutional depth features using the second convolutional network, which is a feature "late fusion" method.

[0045] For example, suppose that images of microfossils taken from different angles are processed by a third and a fourth convolutional network, respectively, generating feature vectors of length L. These feature vectors are then concatenated to obtain a feature vector of length L. The feature vector, where K represents the number of shooting angles, is fused into a feature vector of length L after the fusion operation of the second fusion layer. This feature vector of length L is then fed into the second classification network. The fusion operation performed by the second fusion layer can be: 1) The input and output sizes are respectively... 1) Fully connected layer; 2) eigenvector transpose After the two-dimensional tensor, in Perform max pooling or average pooling in the direction of the pool.

[0046] Figure 4 A third network structure diagram of the genus and species identification model according to an embodiment of this application is illustrated schematically. For example... Figure 4 As shown, the genus and species identification model includes multiple fifth convolutional networks, multiple third fusion layers, multiple sixth convolutional networks, a fourth fusion layer, and a third classification network, wherein: Multiple fifth convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain various seventh feature maps; Multiple third fusion layers are used to fuse the seventh feature maps in each group in a one-to-one correspondence to obtain the eighth feature maps that correspond one-to-one with each group; Multiple sixth convolutional networks are used to extract the second convolutional depth features of each eighth feature map and each remaining seventh feature map that has not been assigned to any group, in a one-to-one correspondence, to obtain each ninth feature map. The fourth fusion layer is used to fuse the various ninth feature maps to obtain the tenth feature map; The third classification network is used to classify microfossil genera and species using the tenth feature map; The grouping rules for the seventh feature image include: if the shooting angle differences of the microfossil images corresponding to multiple seventh feature images are within a first preset range, then the multiple seventh feature images are grouped into the same group.

[0047] For example, in one specific embodiment, the microfossil images input to the genus and species identification model are captured from top, bottom, and side angles. The side angles differ significantly. Therefore, the seventh feature maps extracted from the microfossil images captured from the top and bottom are grouped together and fused through the third fusion layer. Because the side angles differ significantly, the seventh feature maps extracted from the microfossil images captured from each side are directly fed into the respective sixth convolutional networks. Figure 4 In the first preset range, there are multiple differences in shooting angles. The seventh feature maps corresponding to these shooting angles are grouped together. There is a seventh feature map corresponding to one shooting angle that is not grouped into this group.

[0048] In this application, the third fusion layer can be a convolutional kernel with a size of The third fusion layer can be any of the following: a convolutional layer, a max-pooling layer, or an average-pooling layer. The max-pooling layer, used as the third fusion layer, performs max-pooling along the channel direction, and the average-pooling layer, used as the third fusion layer, performs average-pooling along the channel direction. The fourth fusion layer can be any of the following: a fully connected layer, a max-pooling layer, or an average-pooling layer. The max-pooling layer, used as the fourth fusion layer, performs max-pooling along the K-direction, and the average-pooling layer, used as the fourth fusion layer, performs average-pooling along the K-direction. The K-direction refers to the direction where the shooting angle is used as the dimension.

[0049] For example, given K images of microfossils taken from different angles and input into the genus and species identification model, the differences in shooting angles among the images of microfossils taken from different angles are grouped within a first preset range. The images of Zhangwei fossils are grouped together. The images of Zhang Wei fossils were processed by a fifth convolutional network to extract features, resulting in seven feature maps. These seven feature maps were then fused by the third fusion layers to obtain feature vectors U. The remaining... After passing through the fifth and sixth convolutional networks, the Zhang Wei fossil image yields a set of feature vectors. Then, the fourth fusion layer is used to process this. The feature vectors are fused together, and the resulting tenth feature map is then fed into the third classification network.

[0050] As in the above embodiment, the features extracted from images of microfossils taken from different shooting angles are fused twice. The first fusion involves the fusion of some first convolutional depth features, and the second fusion involves the fusion of second convolutional depth features. Compared with the first and second network structures of the genus and species identification model provided in this application embodiment, the number of fusions is increased, and the fusion of different convolutional depth features is involved at the same time. In addition, the group fusion based on the shooting angle difference fuses the first convolutional depth features extracted from each microfossil image with shooting angle differences within a first preset range, which enhances the expression of the first convolutional depth features, thereby providing more accurate classification features to the third classification network, and thus improving the accuracy of genus and species identification.

[0051] Figure 5 A fourth network structure diagram of the genus and species identification model according to an embodiment of this application is illustrated schematically. For example... Figure 5 As shown, the genus and species identification model includes multiple seventh convolutional networks, multiple fifth fusion layers, a sixth fusion layer, an eighth convolutional network, and a fourth classification network, wherein: Multiple seventh convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain the eleventh feature maps; Multiple fifth fusion layers are used to fuse the eleventh feature maps in each group in a one-to-one correspondence, to obtain the twelfth feature maps that correspond one-to-one with each group; The sixth fusion layer is used to fuse the various twelfth feature maps to obtain the thirteenth feature map; The eighth convolutional network is used to extract the second convolutional depth features of the thirteenth feature map to obtain the fourteenth feature map; The fourth classification network is used to classify microfossil genera and species using the fourteenth feature map; The grouping rules for the eleventh feature image include: if the shooting angle differences of the microfossil images corresponding to multiple eleventh feature images are within a second preset range, then the multiple eleventh feature images are grouped into the same group.

[0052] In this application, both the fifth and sixth fusion layers have convolutional kernels with a size of [missing value]. It is one of the following: convolutional layer, max pooling layer, or average pooling layer. Here, the max pooling layer performs max pooling in the channel direction, and the average pooling layer performs average pooling in the channel direction.

[0053] For example, based on the differences in the shooting angle of the microfossil images, they are divided into q groups. For each group, the microfossil images of each group are generated by each seventh convolutional network. The feature map tensor (eleventh feature map). The feature map tensors are obtained by concatenating them along the channel direction. Feature map tensor, This indicates the number of shooting angles within the group, which is then processed through the fifth fusion layer to obtain... The feature map tensor (twelfth feature map), q groups generate q elements. The feature map tensors are obtained by concatenating them along the channel direction. The feature map, after passing through the sixth fusion layer, yields... The feature map tensor (the thirteenth feature map) will be used to... The feature map tensors are fed into the eighth convolutional network and the subsequent fourth classification network. Figure 5 The diagram shown is of the network structure with q=2.

[0054] As in the above embodiment, the first convolutional depth features extracted from images of microfossils taken from different shooting angles are grouped and fused based on the differences in shooting angles, which is a variation of the first network structure of the genus and species identification model provided in this application embodiment. The fusion operation performed within each group can be flexibly varied. By flexibly selecting, the genus and species identification effect based on this network structure can be optimized, and it is more flexible than the first network structure.

[0055] Figure 6 A fifth network structure diagram of a genus and species identification model according to an embodiment of this application is illustrated schematically. For example... Figure 6 As shown, the genus and species identification model includes multiple ninth convolutional networks, multiple tenth convolutional networks, multiple seventh fusion layers, eighth fusion layers, and a fifth classification network, wherein: Multiple ninth convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain the fifteenth feature maps; Multiple tenth convolutional networks are used to extract the second convolutional depth features of each fifteenth feature map in a one-to-one correspondence, thus obtaining each sixteenth feature map; Multiple seventh fusion layers are used to fuse the sixteenth feature maps in each group in a one-to-one correspondence to obtain the seventeenth feature maps that correspond one-to-one with each group. The eighth fusion layer is used to fuse the various seventeenth feature maps to obtain the eighteenth feature map; The fifth classification network is used to classify genera and species of microfossils using the eighteenth feature map; The grouping rules for the sixteenth feature map include: If the shooting angle differences of the microfossil images corresponding to multiple sixteenth feature images are within the third preset range, then the multiple sixteenth feature images are grouped into the same group.

[0056] In this application, both the seventh and eighth fusion layers are one of a fully connected layer, a max pooling layer, and an average pooling layer. Here, the max pooling performed by the max pooling layer is K-direction max pooling, and the average pooling performed by the average pooling layer is K-direction average pooling. The K-direction refers to the direction with the shooting angle as the dimension.

[0057] For example, based on the differences in the shooting angles of the microfossil images, they are divided into q groups. For each group, each microfossil image in the group is processed by the ninth and tenth convolutional networks to generate a feature vector of length L (the sixteenth feature map). The feature vectors of length L generated in the group are concatenated to obtain a feature vector of length nL, where n represents the number of shooting angles in the group. After fusion operation, a feature vector of length L is obtained. The q feature vectors of length L generated in group q are concatenated to obtain a feature vector of length qL. After fusion operation again, a feature vector of length L is obtained. The feature vector of length L obtained at this time is fed into the fifth classification network. Figure 6 The diagram shows the network structure with q=2.

[0058] As in the above embodiment, the second convolutional depth features extracted from images of microfossils taken from different shooting angles are then grouped and fused based on the differences in shooting angles. This is a variation of the second network structure of the genus and species identification model provided in this application embodiment. The fusion operation performed within each group can be flexibly varied. By flexibly selecting the appropriate method, the genus and species identification effect based on this network structure can be optimized, making it more flexible than the second network structure.

[0059] Conodonts, as a special type of microfossil, typically have a principal axis in their structure. Therefore, when using the microfossil species identification method provided in this application to identify the species of conodonts, the image of the conodont input into the species identification model can be captured from the top, bottom, or side. The side view can be from the front, back, left, or right. In a specific application example for conodonts, the network structure of the species identification model is as follows: Figure 7 As shown, the system includes six convolutional networks A1, three convolutional networks A2, one fusion layer C1, one fusion layer C2, and a classifier. Convolutional network A1 extracts shallow features, convolutional network A2 extracts deep features, and fusion layer C1 has a convolutional kernel of... The convolutional layer C2 is one of the following: a channel-oriented max pooling layer, a channel-oriented average pooling layer, and a fully connected layer with a convolutional kernel, or a K-oriented max pooling layer, or a K-oriented average pooling layer. Figure 7 In the process, there are a total of 6 shooting angles, namely, the images used as input to the model include the top view image of the conodont taken from the top, the bottom view image of the conodont taken from the bottom, the front view image of the conodont taken from the front, the rear view image of the conodont taken from the rear, the left view image of the conodont taken from the rear, and the right view images of the conodont taken from the rear, the left, and the right views ... fusion layer C1 is used for fusion.

[0060] This application also provides a microfossil species identification device, including: The image acquisition module is used to acquire images of microfossils from different shooting angles; The genus and species identification module is used to input images of the acquired microfossils taken from different angles into a pre-configured... The established genus and species identification model identifies the genus and species of the microfossils; The genus and species identification model is achieved by training and optimizing the parameters of an initial convolutional neural network. The resulting model is used to extract features from images of microfossils taken from different angles and to fuse the extracted features according to a preset fusion rule. The fused features are then used for species classification. The fusion rule includes at least one of the fusion of a first convolutional depth feature extracted from images of microfossils taken from different angles and the fusion of a second convolutional depth feature extracted from images of microfossils taken from different angles. The convolutional layer depth used to extract the second convolutional depth feature is greater than the convolutional layer depth used to extract the first convolutional depth feature.

[0061] In a specific embodiment of this application, the genus and species identification model includes: Multiple first convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one by one, to obtain each first feature map; The first fusion layer is used to fuse the various first feature maps to obtain the second feature map; The second convolutional network is used to extract the second convolutional depth features of the second feature map to obtain the third feature map; The first classification network is used to classify microfossil genera and species using the third feature map.

[0062] In a specific embodiment of this application, the genus and species identification model includes: Multiple third convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one by one, to obtain the fourth feature maps. Multiple fourth convolutional networks are used to extract the second convolutional depth features of each fourth feature map in a one-to-one correspondence, so as to obtain each fifth feature map; The second fusion layer is used to fuse the various fifth feature maps to obtain the sixth feature map; The second classification network is used to classify microfossil genera and species using the sixth feature map.

[0063] In a specific embodiment of this application, the genus and species identification model includes: Multiple fifth convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain various seventh feature maps; Multiple third fusion layers are used to fuse the seventh feature maps in each group in a one-to-one correspondence to obtain the eighth feature maps that correspond one-to-one with each group; Multiple sixth convolutional networks are used to extract the second convolutional depth features of each eighth feature map and each remaining seventh feature map that has not been assigned to any group, in a one-to-one correspondence, to obtain each ninth feature map. The fourth fusion layer is used to fuse the various ninth feature maps to obtain the tenth feature map; The third classification network is used to classify microfossil genera and species using the tenth feature map; The grouping rules for the seventh feature image include: if the shooting angle differences of the microfossil images corresponding to multiple seventh feature images are within a first preset range, then the multiple seventh feature images are grouped into the same group.

[0064] In a specific embodiment of this application, the genus and species identification model includes: Multiple seventh convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain the eleventh feature maps; Multiple fifth fusion layers are used to fuse the eleventh feature maps in each group in a one-to-one correspondence, to obtain the twelfth feature maps that correspond one-to-one with each group; The sixth fusion layer is used to fuse the various twelfth feature maps to obtain the thirteenth feature map; The eighth convolutional network is used to extract the second convolutional depth features of the thirteenth feature map to obtain the fourteenth feature map; The fourth classification network is used to classify microfossil genera and species using the fourteenth feature map; The grouping rules for the eleventh feature map include: If the shooting angle differences of the microfossil images corresponding to multiple eleventh feature images are within a second preset range, then the multiple eleventh feature images are grouped into the same group.

[0065] In a specific embodiment of this application, the genus and species identification model includes: Multiple ninth convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain the fifteenth feature maps; Multiple tenth convolutional networks are used to extract the second convolutional depth features of each fifteenth feature map in a one-to-one correspondence, thus obtaining each sixteenth feature map; Multiple seventh fusion layers are used to fuse the sixteenth feature maps in each group in a one-to-one correspondence to obtain the seventeenth feature maps that correspond one-to-one with each group. The eighth fusion layer is used to fuse the various seventeenth feature maps to obtain the eighteenth feature map; The fifth classification network is used to classify genera and species of microfossils using the eighteenth feature map; The grouping rules for the sixteenth feature map include: If the shooting angle differences of the microfossil images corresponding to multiple sixteenth feature images are within the third preset range, then the multiple sixteenth feature images are grouped into the same group.

[0066] In a specific embodiment of this application, the first fusion layer has a convolutional kernel size of One of the following: convolutional layer, max pooling layer, or average pooling layer.

[0067] In a specific embodiment of this application, the second fusion layer is one of a fully connected layer, a max pooling layer, and an average pooling layer.

[0068] In a specific embodiment of this application, the third fusion layer has a convolutional kernel size of The fourth fusion layer is one of a convolutional layer, a max pooling layer, and an average pooling layer, and is a fully connected layer, a max pooling layer, and an average pooling layer.

[0069] In a specific embodiment of this application, both the fifth and sixth fusion layers have convolutional kernel sizes of [size missing]. One of the following: convolutional layer, max pooling layer, or average pooling layer.

[0070] In specific embodiments of this application, the seventh fusion layer and the eighth fusion layer are both one of a fully connected layer, a max pooling layer, and an average pooling layer.

[0071] In specific embodiments of this application, the images of the microfossils taken from different angles include top-view images, bottom-view images, and side-view images.

[0072] In a specific embodiment of this application, a genus and species identification model is obtained after training and parameter optimization of an initial convolutional neural network, including: The weights of the pre-trained model are used as the initial weights of the convolutional layers in the initial convolutional neural network used to extract the first convolutional depth features and the convolutional layers used to extract the second convolutional depth features, as well as the weights of the classification network in the initial convolutional neural network used for the classification of microfossil genera and species. Train the current convolutional neural network using the constructed training samples; The training results are evaluated using a preset loss function. If the training results meet the preset requirements, the next step is executed. Otherwise, the weight values ​​of each convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for the classification of microfossil genera and species are updated. The absolute value of the weight difference before and after the update of the convolutional layer used to extract the first convolutional depth features is less than the absolute value of the weight difference before and after the update of the convolutional layer used to extract the second convolutional depth features, and the absolute value of the weight difference before and after the update of the convolutional layer used to extract the second convolutional depth features is less than the absolute value of the weight difference before and after the update of the classification network used for the classification of microfossil genera and species are executed. Then, the process jumps to the previous step. The latest weight values ​​are assigned to the convolutional neural network to obtain the species identification model.

[0073] In a specific embodiment of this application, updating the weight values ​​within the convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for classifying microfossil genera and species includes: The first formula is used to update the weight values ​​of the convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for the classification of microfossil genera and species. The first formula is expressed as: , This indicates the updated weights. This indicates the weights before the update. Indicates the learning rate. Represents the loss function In weight The rate of change in direction, used to extract the first convolutional depth features during weight updates. The value is less than that of the convolutional layer used to extract the second convolutional depth feature during weight update. The value is used to extract the second convolutional depth feature during weight updates. The value is less than that of the classification network used for the classification of microfossil genera and species when updating weights. Take values, and .

[0074] In a specific embodiment of this application, the weights of each convolutional layer used to extract the depth features of images taken from different shooting angles of microfossils are updated. Different values ​​are used to update the weights of each convolutional layer in the second convolutional depth feature extraction process for images taken from different angles of microfossils. The values ​​are different.

[0075] Figure 8A schematic block diagram of a computer device according to an embodiment of the present application is shown. In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown below. Figure 8 As shown in the figure, the computer device includes a processor A01, a network interface A02, a display screen A04, an input device A05, and a memory (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A06. The non-volatile storage medium A06 stores an operating system B01 and a computer program B02. The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 stored in the non-volatile storage medium A06. The network interface A02 is used for communication with external terminals via a network connection. When the computer program is executed by the processor A01, it implements a method for identifying microfossil species. The display screen A04 can be a liquid crystal display (LCD) or an e-ink display. The input device A05 can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0076] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0077] In one embodiment, the microfossil species identification device provided in this application can be implemented as a computer program, which can be implemented in the form of, for example... Figure 8 The device operates on the computer shown. The computer's memory can store various program modules that make up the microfossil species identification device. The computer program, composed of these program modules, causes the processor to execute the steps in the microfossil species identification methods of the various embodiments of this application described in this specification.

[0078] In one embodiment, this application also provides a machine-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the microfossil species identification method in the above embodiments.

[0079] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0080] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0081] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for intelligent identification of genera and species of microfossils, characterized in that, include: Acquire images of microfossils from different shooting angles; Images of the microfossils taken from different angles are input into a pre-built genus and species identification model to identify the genus and species of the microfossils. The genus and species identification model is obtained by training and optimizing the parameters of an initial convolutional neural network. The genus and species identification model is used to extract features from images of microfossils taken from different angles and to fuse the extracted features according to a preset fusion rule, and to classify the genus and species using the fused features. The fusion rule includes at least one of the fusion of a first convolutional depth feature extracted from images of microfossils taken from different angles and the fusion of a second convolutional depth feature extracted from images of microfossils taken from different angles, wherein the convolutional layer depth used to extract the second convolutional depth feature is greater than the convolutional layer depth used to extract the first convolutional depth feature.

2. The method according to claim 1, characterized in that, The genus and species identification model includes: Multiple first convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one by one, to obtain each first feature map; The first fusion layer is used to fuse the various first feature maps to obtain the second feature map; The second convolutional network is used to extract the second convolutional depth features of the second feature map to obtain the third feature map; The first classification network is used to classify microfossil genera and species using the third feature map.

3. The method according to claim 1, characterized in that, The genus and species identification model includes: Multiple third convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one by one, to obtain the fourth feature maps. Multiple fourth convolutional networks are used to extract the second convolutional depth features of each fourth feature map in a one-to-one correspondence, so as to obtain each fifth feature map; The second fusion layer is used to fuse the various fifth feature maps to obtain the sixth feature map; The second classification network is used to classify microfossil genera and species using the sixth feature map.

4. The method according to claim 1, characterized in that, The genus and species identification model includes: Multiple fifth convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain various seventh feature maps; Multiple third fusion layers are used to fuse the seventh feature maps in each group in a one-to-one correspondence to obtain the eighth feature maps that correspond one-to-one with each group; Multiple sixth convolutional networks are used to extract the second convolutional depth features of each eighth feature map and each remaining seventh feature map that has not been assigned to any group, in a one-to-one correspondence, to obtain each ninth feature map. The fourth fusion layer is used to fuse the various ninth feature maps to obtain the tenth feature map; The third classification network is used to classify microfossil genera and species using the tenth feature map; The grouping rules for the seventh feature image include: if the shooting angle differences of the microfossil images corresponding to multiple seventh feature images are within a first preset range, then the multiple seventh feature images are grouped into the same group.

5. The method according to claim 1, characterized in that, The genus and species identification model includes: Multiple seventh convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain the eleventh feature maps; Multiple fifth fusion layers are used to fuse the eleventh feature maps in each group in a one-to-one correspondence, to obtain the twelfth feature maps that correspond one-to-one with each group; The sixth fusion layer is used to fuse the various twelfth feature maps to obtain the thirteenth feature map; The eighth convolutional network is used to extract the second convolutional depth features of the thirteenth feature map to obtain the fourteenth feature map; The fourth classification network is used to classify microfossil genera and species using the fourteenth feature map; The grouping rules for the eleventh feature map include: If the shooting angle differences of the microfossil images corresponding to multiple eleventh feature images are within a second preset range, then the multiple eleventh feature images are grouped into the same group.

6. The method according to claim 1, characterized in that, The genus and species identification model includes: Multiple ninth convolutional networks are used to extract the first convolutional depth features of images of microfossils taken from different shooting angles, one-to-one, to obtain the fifteenth feature maps; Multiple tenth convolutional networks are used to extract the second convolutional depth features of each fifteenth feature map in a one-to-one correspondence, thus obtaining each sixteenth feature map; Multiple seventh fusion layers are used to fuse the sixteenth feature maps in each group in a one-to-one correspondence to obtain the seventeenth feature maps that correspond one-to-one with each group. The eighth fusion layer is used to fuse the various seventeenth feature maps to obtain the eighteenth feature map; The fifth classification network is used to classify genera and species of microfossils using the eighteenth feature map; The grouping rules for the sixteenth feature map include: If the shooting angle differences of the microfossil images corresponding to multiple sixteenth feature images are within the third preset range, then the multiple sixteenth feature images are grouped into the same group.

7. The method according to claim 2, characterized in that, The first fusion layer has a convolutional kernel size of One of the following: convolutional layer, max pooling layer, or average pooling layer.

8. The method according to claim 3, characterized in that, The second fusion layer is one of a fully connected layer, a max pooling layer, and an average pooling layer.

9. The method according to claim 4, characterized in that, The third fusion layer has a convolutional kernel size of [missing information]. The fourth fusion layer is one of a convolutional layer, a max pooling layer, and an average pooling layer, and is a fully connected layer, a max pooling layer, and an average pooling layer.

10. The method according to claim 5, characterized in that, The fifth and sixth fusion layers both have convolutional kernel sizes of [missing information]. One of the following: convolutional layer, max pooling layer, or average pooling layer.

11. The method according to claim 6, characterized in that, The seventh and eighth fusion layers are both one of the following: fully connected layer, maximum pooling layer, and average pooling layer.

12. The method according to claim 1, characterized in that, The images of the microfossils taken from different angles include top-down, bottom-up, and side-view images.

13. The method according to claim 1, characterized in that, After training and parameter optimization of the initial convolutional neural network, a genus and species identification model is obtained, including: The weights of the pre-trained model are used as the initial weights of the convolutional layers in the initial convolutional neural network used to extract the first convolutional depth features and the convolutional layers used to extract the second convolutional depth features, as well as the weights of the classification network in the initial convolutional neural network used for the classification of microfossil genera and species. Train the current convolutional neural network using the constructed training samples; The training results are evaluated using a preset loss function. If the training results meet the preset requirements, the next step is executed; otherwise, the weight values ​​of each convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for the classification of microfossil genera and species are updated. The absolute value of the weight difference before and after the update of the convolutional layer used to extract the first convolutional depth features is less than the absolute value of the weight difference before and after the update of the convolutional layer used to extract the second convolutional depth features, and the absolute value of the weight difference before and after the update of the convolutional layer used to extract the second convolutional depth features is less than the absolute value of the weight difference before and after the update of the classification network used for the classification of microfossil genera and species are executed. Then, the process jumps to the previous step. The latest weight values ​​are assigned to the convolutional neural network to obtain the species identification model.

14. The method according to claim 13, characterized in that, Update the weights in the convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for classifying microfossil genera and species, including: The first formula is used to update the weight values ​​of the convolutional layer used to extract the first convolutional depth features, the convolutional layer used to extract the second convolutional depth features, and the classification network used for the classification of microfossil genera and species. The first formula is expressed as: , This indicates the updated weights. This indicates the weights before the update. Indicates the learning rate. Represents the loss function In weight The rate of change in direction, used to extract the first convolutional depth features during weight updates. The value is less than that of the convolutional layer used to extract the second convolutional depth feature during weight update. The value is used to extract the second convolutional depth feature during weight updates. The value is less than that of the classification network used for the classification of microfossil genera and species when updating weights. Take values, and .

15. The method according to claim 14, characterized in that, During weight updates, the first convolutional depth features of each convolutional layer used to extract images of microfossils from different shooting angles are updated. Different values ​​are used to update the weights of each convolutional layer in the second convolutional depth feature extraction process for images taken from different angles of microfossils. The values ​​are different.

16. A smart identification device for microfossil genera and species, characterized in that, include: The image acquisition module is used to acquire images of microfossils from different shooting angles; The genus and species identification module is used to input images of the acquired microfossils from different shooting angles into a pre-constructed genus and species identification model to identify the genus and species of the microfossils. The genus and species identification model is obtained by training and optimizing the parameters of an initial convolutional neural network. The genus and species identification model is used to extract features from images of microfossils taken from different angles and to fuse the extracted features according to a preset fusion rule, and to classify the genus and species using the fused features. The fusion rule includes at least one of the fusion of a first convolutional depth feature extracted from images of microfossils taken from different angles and the fusion of a second convolutional depth feature extracted from images of microfossils taken from different angles, wherein the convolutional layer depth used to extract the second convolutional depth feature is greater than the convolutional layer depth used to extract the first convolutional depth feature.

17. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the intelligent identification method for microfossil genera and species as described in any one of claims 1-15.

18. A machine-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the intelligent identification method for microfossil genera and species as described in any one of claims 1-15.

Images