True coral image data determination method and device, equipment, medium and product

Abstract

The invention discloses a true coral image data determination method and device, equipment, a medium and a product, and relates to the technical field of image data processing, and the method comprises the steps: obtaining initial coral image data from an image search engine based on a preset keyword; inputting the initial coral image data into a pre-trained coral recognition model to obtain a coral recognition result output by the coral recognition model; if the coral identification result shows that the initial coral image data is true coral image data, performing data enhancement on the true coral image data to obtain true coral image enhancement data; normalizing the true coral image enhanced data to obtain true coral image normalized data; and inputting the normalized data of the real coral image into a pre-trained coral classification model to obtain a coral classification result of the real coral image data output by the coral classification model, thereby improving the accuracy of identification and classification of the real coral image data.

Description

Technical Field

[0001] The present application relates to the technical field of image data processing, and in particular to a method, device, equipment, medium and product for determining real coral image data. Background Art

[0002] Coral reef ecosystems play a vital role in maintaining the biodiversity and balance of marine ecosystems and achieving sustainable development of marine resources. However, in recent years, the global marine environment has continued to deteriorate, mainly manifested in the rise in ocean temperatures and the intensification of seawater acidification. Overfishing has had an unignorable impact on the balance of coral reef ecosystems. At the same time, marine pollutants such as wastewater discharge, oil pollution, and plastic pollution have also caused serious damage to coral reefs. These factors have posed severe challenges to the survival of coral reefs and threatened the stability and living space of their ecosystems.

[0003] Therefore, the protection of corals is urgent. It can be seen that the protection of corals is inseparable from a large number of studies on corals, and these studies require sufficient data support. In this context, it is particularly important to establish and improve the coral image dataset.

[0004] The construction of current coral datasets usually requires manual identification of real coral images from a large number of images, and the identification of real coral images needs to be manually classified. However, in practice, it is found that manual identification of real coral images from a large number of images and classification of coral images are prone to identification or classification errors, and the efficiency of manual identification and classification of massive images is low. Summary of the invention

[0005] The purpose of this application is to provide a method, device, equipment, medium and product for determining real coral image data, which can improve the accuracy of identifying and classifying real coral image data.

[0006] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides a method for determining real coral image data, comprising: Obtaining initial coral image data from an image search engine based on preset keywords; Inputting the initial coral image data into a pre-trained coral recognition model to obtain a coral recognition result output by the coral recognition model; If the coral recognition result indicates that the initial coral image data is real coral image data, data enhancement is performed on the real coral image data to obtain real coral image enhanced data; Normalizing the real coral image enhancement data to obtain real coral image normalized data; Input the normalized data of the true coral image into a pre-trained coral classification model to obtain the coral classification result of the true coral image data output by the coral classification model.

[0007] Optionally, the data augmentation of the true coral image data to obtain true coral image enhanced data specifically includes: Crop the true coral image data to obtain central region image data; wherein, the size of the central region image data is a first preset size; Perform random probability rotation on the central region image data to obtain rotated image data; Perform random probability brightness enhancement on the rotated image data to obtain first enhanced image data; Perform random probability color enhancement on the first enhanced image data to obtain second enhanced image data; Perform random probability contrast enhancement on the second enhanced image data to obtain third enhanced image data; Perform random probability sharpening enhancement on the third enhanced image data to obtain fourth enhanced image data; Crop the fourth enhanced image data to obtain true coral image enhanced data; wherein, the size of the true coral image enhanced data is a second preset size, and the second preset size is smaller than the first preset size.

[0008] Optionally, the training method of the coral recognition model is specifically: Obtain a non-coral image dataset and a classified true coral image dataset; wherein, the non-coral image dataset includes multiple non-coral image data, and the classified true coral image dataset includes multiple classified true coral image data; Based on the non-coral image dataset and the classified true coral image dataset, construct a first training dataset and a first validation dataset; Use the first training dataset to train a pre-constructed coral recognition model to obtain a trained coral recognition model; Use the first validation dataset to validate the trained coral recognition model to obtain a first validation result; If the first validation result indicates that the validation of the trained coral recognition model fails, then use the binary cross-entropy loss function to adjust the trained coral recognition model to obtain an adjusted coral recognition model; and use the first training dataset to train the adjusted coral recognition model to obtain a trained coral recognition model; and perform the step of using the first validation dataset to validate the trained coral recognition model to obtain a first validation result; If the first verification result indicates that the trained coral recognition model passes the verification, a trained coral recognition model is obtained.

[0009] Optionally, the training method of the coral classification model is specifically as follows: Construct a non-repetitive true coral image dataset from the classified true coral image dataset; wherein, the non-repetitive true coral image dataset contains at least one non-repetitive true coral image data, the non-repetitive true coral image data is obtained from the classified true coral image dataset, and any two non-repetitive true coral image data in the non-repetitive true coral image dataset are not the same; Based on the non-repetitive true coral image dataset, construct a second training dataset and a second verification dataset; Perform data augmentation and normalization on the second training dataset to obtain a second augmented training dataset; Perform data augmentation and normalization on the second verification dataset to obtain a second augmented verification dataset; Use the second augmented training dataset to train a pre-constructed coral classification model to obtain a trained coral classification model; Use the second augmented verification dataset to verify the trained coral classification model to obtain a second verification result; If the second verification result indicates that the trained coral classification model fails the verification, use the object detection loss function to adjust the coral classification model to obtain an adjusted coral classification model; and use the second augmented training dataset to train the adjusted coral classification model to obtain a trained coral classification model; and perform the step of using the second augmented verification dataset to verify the trained coral classification model to obtain a second verification result; If the second verification result indicates that the trained coral classification model passes the verification, a trained coral classification model is obtained.

[0010] Optionally, the construction of the non-repetitive true coral image dataset from the classified true coral image dataset specifically includes: Construct an initial non-repetitive true coral image dataset; wherein, the initial non-repetitive true coral image dataset is empty; Sequentially obtain a target classified true coral image data from the classified true coral image dataset and perform the following operations: Determine the similarity data set of the target classified true coral image data and the non-repetitive true coral image data in the non-repetitive true coral image dataset; wherein, the similarity data set includes global structural similarity and Hamming distance; If there is no global structural similarity of a similarity data set less than a first preset threshold and Hamming distance greater than a second preset threshold, then delete the target classified true coral image data from the classified true coral image data set; If there is a global structural similarity of a similarity data set less than the first preset threshold and Hamming distance greater than the second preset threshold, then add the target classified true coral image data as non-duplicate true coral image data to the non-duplicate true coral image data set, and delete the target classified true coral image data from the classified true coral image data set.

[0011] Optionally, the determining the similarity data set of the target classified true coral image data and the non-duplicate true coral image data in the non-duplicate true coral image data set specifically includes: Determine multiple local structural similarities between the target classified true coral image data and the non-duplicate true coral image data in the non-duplicate true coral image data set; wherein, the target classified true coral image data includes multiple channels, the non-duplicate true coral image data includes multiple channels, and the channels in the target classified true coral image data correspond one-to-one with the channels in the non-duplicate true coral image data, and the local structural similarity is the structural similarity of a corresponding channel between the target classified true coral image data and the non-duplicate true coral image data; Determine the average value of multiple local structural similarities as the global structural similarity; Perform DCT transformation on the target classified true coral image data to obtain a first frequency domain matrix; Perform DCT transformation on the non-duplicate true coral image data to obtain a second frequency domain matrix; Calculate a first hash sequence of the first frequency domain matrix and a second hash sequence of the second frequency domain matrix; Use the first hash sequence and the second hash sequence to calculate the Hamming distance; Use the global structural similarity and the Hamming distance to construct a similarity data set.

[0012] In a second aspect, the present application provides a true coral image data determination device, including: An acquisition unit, configured to acquire initial coral image data from an image search engine based on a preset keyword; A first input unit, configured to input the initial coral image data into a pre-trained coral recognition model to obtain a coral recognition result output by the coral recognition model; A data enhancement unit, configured to perform data enhancement on the true coral image data if the coral recognition result indicates that the initial coral image data is true coral image data, so as to obtain enhanced true coral image data; A normalization unit, configured to normalize the enhanced true coral image data to obtain normalized true coral image data; A second input unit, configured to input the normalized true coral image data into a pre-trained coral classification model, so as to obtain a coral classification result of the true coral image data output by the coral classification model.

[0013] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the steps of the true coral image data determination method described in any one of the above.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the true coral image data determination method described in any one of the above are implemented.

[0015] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the true coral image data determination method described in any one of the above are implemented.

[0016] In a sixth aspect, the present application provides a chip, the chip includes a processor and a communication interface, the communication interface is coupled to the processor, the processor is configured to run a program or an instruction, and when the processor executes the program or the instruction, the steps of the true coral image data determination method described in any one of the above are implemented.

[0017] According to the specific embodiments provided by the present application, the following technical effects are disclosed in the present application: The present application provides a method, apparatus, device, medium and product for determining true coral image data. Through preset keywords, initial coral image data can be quickly obtained from an image shrinking engine. Through a pre-trained coral recognition model, the authenticity of the initial coral image data can be accurately identified. And when it is determined that the initial coral image data is true coral image data, data enhancement can be performed on the true coral image data to obtain enhanced true coral image data. The enhanced true coral image data can enhance the key information in the true coral image data and improve the accuracy of coral classification. Then, the normalized true coral image normalized data can be input into a pre-trained coral classification model, so that the coral classification model can accurately classify the normalized true coral image data, thereby obtaining a coral classification result. It can be seen that in this way, the coral image data obtained from the image search engine can be identified and the coral species can be determined through a pre-trained model, thereby improving the accuracy of identifying and classifying true coral image data. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is a schematic flowchart of a method for determining true coral image data in an embodiment of the present application; Figure 2 It is a schematic structural diagram of a coral recognition model provided in an embodiment of the present application; Figure 3 It is a schematic structural diagram of a coral classification model provided in an embodiment of the present application; Figure 4 It is a schematic structural diagram of an SK-CBAM module provided in an embodiment of the present application; Figure 5 It is a schematic structural diagram of an SK module provided in an embodiment of the present application; Figure 6 It is a schematic structural diagram of a spatial self-attention module provided in an embodiment of the present application; Figure 7 It is a loss visualization diagram obtained by training a coral recognition model using a first training dataset in an embodiment of the present application; Figure 8 It is a schematic diagram of the recognition accuracy obtained by training a coral recognition model using a first training dataset in an embodiment of the present application; Figure 9A loss visualization graph obtained by training a coral recognition model using a first validation data set provided by an embodiment of the present application; Figure 10 A schematic diagram of the recognition accuracy obtained by training a coral recognition model using a first validation data set provided by an embodiment of the present application; Figure 11 A loss visualization graph obtained by training a coral recognition model using a second training data set provided by an embodiment of the present application; Figure 12 A schematic diagram of the recognition accuracy obtained by training a coral recognition model using a second training data set provided by an embodiment of the present application; Figure 13 A loss visualization graph obtained by training a coral recognition model using a second validation data set provided by an embodiment of the present application; Figure 14 A schematic diagram of the recognition accuracy obtained by training a coral recognition model using a second validation data set provided by an embodiment of the present application; Figure 15 A fine-grained classification result graph of a true network coral image data set provided by an embodiment of the present application; Figure 16 A schematic diagram of the functional modules of a true coral image data determination device provided by an embodiment of the present application.

[0020] Figure 17 A schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0021] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0022] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0023] In an exemplary embodiment, as Figure 1 shown, a method for determining true coral image data is provided. This method is executed by a computer device, and specifically can be executed alone by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiments of the present application, the following steps 101 to 105 are included. Among them: Step 101, obtaining initial coral image data from an image search engine based on a preset keyword.

[0024] In the embodiments of the present application, initial coral image data can be automatically retrieved and downloaded from an image search engine according to preset keywords through web crawler technology.

[0025] In the embodiments of the present application, the preset keywords can be multiple categories (such as 22 categories) of coral family names (such as Acroporidae, Agariciidae, Euphyllidae, etc.).

[0026] Optionally, the method for obtaining the initial coral image data from the image search engine based on the preset keywords may specifically include: Determine the URL of the image search engine as the initial URL; Obtain the keyword URL corresponding to the preset keyword; Use the initial URL and the keyword URL to form an image dataset URL; Send a crawler instruction to the image search engine using the image dataset URL, so that the image search engine returns HTML data; Extract an image URL list from the HTML data using a regular expression; Construct a non-repeating image URL list; Take out the image URL from the image URL list and compare it with the non-repeating image URL list to determine whether there is a repetition; If the image URL is repeated with the content in the non-repeating image URL list, discard the image URL; If the image URL is not repeated with the content in the non-repeating image URL list, add the image URL to the non-repeating image URL list; Until the image URL list is empty, send download requests to all the image URLs in the non-repeating image URL list in parallel using multiple threads, and save the downloaded images locally.

[0027] It can be seen that implementing this implementation method can efficiently and accurately automatically collect coral-related image data. By constructing a non-repeating image URL list and using multi-thread technology for downloading, data redundancy is avoided, and the speed of image collection and download is significantly improved. The entire process realizes automated processing, ensuring the diversity and effectiveness of the image data, and providing a rich and high-quality material library for subsequent coral ecological research, coral recognition models, and training of coral classification models.

[0028] Step 102: Input the initial coral image data into a pre-trained coral recognition model to obtain the coral recognition result output by the coral recognition model.

[0029] In the embodiment of the present application, the coral recognition model is used to recognize the input initial coral image data to determine whether the initial coral image data is image data containing corals; if so, the output coral recognition result is that the initial coral image data is true coral image data containing coral images; if not, the output coral recognition result is that the initial coral image data is false coral image data not containing coral images.

[0030] If the coral recognition model recognizes that the initial coral image data is false coral image data not containing coral images, the initial coral image data can be deleted. This avoids the contamination of the coral database by false data.

[0031] In the embodiment of the present application, the training method of the coral recognition model is specifically as follows: Obtain a non-coral image data set and a classified true coral image data set; wherein, the non-coral image data set includes multiple non-coral image data, and the classified true coral image data set includes multiple classified true coral image data; Based on the non-coral image data set and the classified true coral image data set, construct a first training data set and a first validation data set; Use the first training data set to train a pre-constructed coral recognition model to obtain a trained coral recognition model; Use the first validation data set to validate the trained coral recognition model to obtain a first validation result; If the first validation result indicates that the trained coral recognition model fails the validation, use the binary cross-entropy loss function to adjust the trained coral recognition model to obtain an adjusted coral recognition model; and use the first training data set to train the adjusted coral recognition model to obtain a trained coral recognition model; and perform the step of using the first validation data set to validate the trained coral recognition model to obtain a first validation result; If the first validation result indicates that the trained coral recognition model passes the validation, obtain a trained coral recognition model.

[0032] Among them, by comprehensively applying the non-coral image dataset and the classified true coral image dataset in this implementation method, the first training dataset and the first validation dataset are constructed, thus realizing the accurate training and effective verification of the coral recognition model. This training method not only ensures the accuracy and reliability of the model in recognizing true coral images, but also dynamically adjusts the model through the binary cross-entropy loss function, effectively improving the recognition performance of the model. In the model verification stage, if the verification fails, the model parameters are retrained and adjusted, and the model is continuously optimized until it meets the verification requirements, thus ensuring that the trained coral recognition model has high accuracy and stability.

[0033] In the embodiment of this application, the non-coral image dataset and the network true coral image dataset can be automatically retrieved and downloaded from the image search engine according to keywords through web crawler technology. The preset keywords can be multiple categories (such as 22 categories) of coral family names (such as Acroporidae, Agariciidae, Euphyllidae, etc.) and multiple categories (such as 995 categories) of non-coral keywords (such as Apple, Balloon, Banana, etc.).

[0034] Furthermore, the obtained network true coral image dataset can be judged by experts. After fine-grained classification, the classified true coral image dataset is obtained; for example, the classified true coral image dataset can include 22 coral family categories and a total of 5348 coral images.

[0035] In the embodiment of this application, based on the non-coral image dataset and the classified true coral image dataset, the quantity ratio of the constructed first training dataset and the first validation dataset can be a preset ratio (such as 9:1). The image data in the first training dataset and the first validation dataset can be randomly divided from the non-coral image dataset and the classified true coral image dataset. Therefore, both the first training dataset and the first validation dataset can contain non-coral image data and classified true coral image data.

[0036] In the embodiment of this application, the coral recognition model can be ResNet18. ResNet18 is a lightweight deep learning model based on the deep residual network (Residual Network), with 18 layers of depth. Its architecture effectively alleviates the gradient vanishing problem and improves the effect of training deep neural networks by introducing residual connections and cross-layer feature reuse mechanisms, and is widely used in computer vision tasks such as image classification and object recognition.

[0037] The specific network structure is as Figure 2As shown in the figure. Among them, the overall network structure includes a total of 5 Stages and a fully connected layer. Each Stage consists of a series of convolutional layers, batch normalization layers, ReLU activation function layers, and residual blocks. The CBR module first represents the input convolutional layer Conv, which is responsible for feature extraction and transformation of the input data, and then inputs the normalization layer BatchNorm to normalize the output of the convolutional layer, accelerate the network training process, and improve the stability of the model. Finally, it is processed by the activation function ReLU to introduce non-linearity and enhance the expression ability of the model. Since it is a binary classification, after passing through the last fully connected layer Fc, the Sigmoid activation function is used for normalization processing.

[0038] In the embodiment of the present application, the binary cross-entropy loss function can be used to adjust the coral recognition model, and its calculation formula is as follows: Among them, represents the number of samples, represents the true label of the i-th sample (0 represents non-coral image data, 1 represents true coral image data), represents the predicted probability of the i-th sample.

[0039] Step 103, if the coral recognition result indicates that the initial coral image data is true coral image data, then perform data augmentation on the true coral image data to obtain enhanced true coral image data.

[0040] As an optional implementation manner, the method for performing data augmentation on the true coral image data in step 103 to obtain enhanced true coral image data may specifically include: Crop the true coral image data to obtain central region image data; among them, the size of the central region image data is the first preset size; Perform random probability rotation on the central region image data to obtain rotated image data; Perform random probability brightness enhancement on the rotated image data to obtain the first enhanced image data; Perform random probability color enhancement on the first enhanced image data to obtain the second enhanced image data; Perform random probability contrast enhancement on the second enhanced image data to obtain the third enhanced image data; Perform random probability sharpening enhancement on the third enhanced image data to obtain the fourth enhanced image data; Crop the fourth enhanced image data to obtain enhanced true coral image data; among them, the size of the enhanced true coral image data is the second preset size, and the second preset size is smaller than the first preset size.

[0041] Among them, when implementing this implementation method, by performing a series of data augmentation processes with random probabilities on the true coral image data, including cropping the central region to focus on key information, random rotation, brightness enhancement, color enhancement, contrast enhancement, and sharpening enhancement, the diversity and complexity of the image data can be significantly increased, the ability of the model to extract features from coral images and the generalization ability can be improved, and at the same time, the robustness of the model to environmental factors such as illumination, color, and contrast can be enhanced. In addition, by flexibly adjusting the cropping size, it is ensured that the enhanced true coral image data meets the requirements of specific application scenarios or model inputs, and overall, the practicality of data processing and the model performance are improved.

[0042] In the embodiment of the present application, since the enhanced data of the true coral image is cropped based on the true coral image data, the size of the enhanced data of the true coral image is smaller than that of the true coral image data.

[0043] For example, the first preset size can be 384×384, the second preset size can be 224×224, and the random rotation angle range for randomly rotating the image data in the central region can be [-40°, 40°].

[0044] In the embodiment of the present application, the specific method for performing random probability color enhancement on the first enhanced image data can be: with a probability of 50%, select whether to perform brightness enhancement on the rotated image data, and randomly select an enhancement coefficient from [-0.5, 0.5].

[0045] In the embodiment of the present application, the specific method for performing random probability color enhancement on the first enhanced image data can be: with a probability of 50%, select whether to perform color enhancement on the first enhanced image data, and randomly select an enhancement coefficient from [-0.5, 0.5].

[0046] In the embodiment of the present application, the specific method for performing random probability contrast enhancement on the second enhanced image data can be: with a probability of 50%, select whether to perform contrast enhancement on the second enhanced image data, and randomly select an enhancement coefficient from [-0.5, 0.5].

[0047] In the embodiment of the present application, the specific method for performing random probability sharpening enhancement on the third enhanced image data can be: with a probability of 50%, select whether to perform sharpening enhancement on the third enhanced image data, and randomly select an enhancement coefficient from [-0.5, 0.5].

[0048] Step 104, normalize the enhanced data of the true coral image to obtain the normalized data of the true coral image.

[0049] In the embodiments of the present application, the method for normalizing the enhanced data of the true coral image may be to divide each piece of enhanced data of the true coral image by 255, so as to obtain the normalized data of the true coral image.

[0050] Step 105: Input the normalized data of the true coral image into a pre-trained coral classification model, and obtain the coral classification result of the true coral image data output by the coral classification model.

[0051] In the embodiments of the present application, the coral classification model is used to perform fine-grained classification on the input normalized data of the true coral image, so as to obtain the name of the coral family corresponding to the coral included in the true coral image data, that is, the coral classification result.

[0052] In the embodiments of the present application, the coral classification model is ResNet50, and the ResNet50 network structure is as Figure 3 shown. It also includes five Stages. In the present invention, the internal structures of the 5 Stages in the original ResNet50 network structure are kept unchanged, and SK-CBAM modules are added in the middle of Stage1 and Stage2, Stage2 and Stage3, and Stage3 and Stage4 respectively. Since it is a multi-classification, finally, the Softmax activation function is used for processing to obtain the prediction probability of each type of coral.

[0053] Furthermore, the structure of the SK-CBAM module is as Figure 4 shown. Among them, the number of channels of the input feature map is C, the height is H, and the width is W. This module keeps the spatial attention module in the CBAM module unchanged, discards the original channel attention module, and replaces the weighted feature map obtained by processing the input feature map by the original channel attention module with the feature map obtained by processing by the SK module.

[0054] Furthermore, the SK module (Selective Kernel Module) is a module used in deep neural networks, and the model performance can be improved by dynamically selecting feature information of different scales. In this example, the structure of the SK module is as Figure 5As shown in the figure. The module consists of four branches, where the number of channels of the input feature map is C, the height is H, and the width is W. Each branch is processed through the corresponding CBR_1 (1×1, C), CBR_2 (3×3, C), CBR_3 (5×5, C), and CBR_4 (7×7, C) modules to obtain the feature maps extracted by the four branches respectively. Then, the feature maps extracted by the four branches are summed up, and the height and width dimensions of the feature map are reduced to 1 through the average pooling layer (Avgpool), and the number of channel dimensions is converted from C to D through the fully connected layer Fc_1 (D). Subsequently, after being processed by the four fully connected layers Fc_2 (C), Fc_3 (C), Fc_4 (C), and Fc_5 (C) respectively, and then processed by the Softmax activation function, the channel attention weights of the four branches are obtained. The final output feature map is the result of weighted summation of the feature maps extracted by the four branches.

[0055] Furthermore, the Convolutional Block Attention Module (CBAM) is a deep learning module based on channel attention and spatial attention mechanisms, which is used to improve the performance of neural networks in computer vision tasks. The structure of the spatial attention module in CBAM is as Figure 6 shown in the figure. In this module, the input feature map (C, H, W) first undergoes max pooling (Maxpool) and average pooling (Avgpool) respectively in the channel dimension, and then generates a two-channel feature map (2, H, W) through the concatenation operation (Concat). Next, this feature map is processed by a convolutional layer Conv (k×k, 1) and the Sigmoid activation function, and finally generates the spatial attention weight (1, H, W). The convolutional layer scales of the spatial attention modules adopted by different SK-CBAM modules are different. In this example, the convolutional layer scales of the spatial attention modules used by SK-CBAM_1, SK-CBAM_2, and SK-CBAM_3 are 31×31, 15×15, and 7×7 respectively.

[0056] In the embodiment of the present application, the training method of the coral classification model is specifically as follows: Construct a non-repetitive true coral image dataset from the classified true coral image dataset; wherein, at least one non-repetitive true coral image data is included in the non-repetitive true coral image dataset, the non-repetitive true coral image data is obtained from the classified true coral image dataset, and any two non-repetitive true coral image data in the non-repetitive true coral image dataset are not the same; Based on the non-repetitive true coral image dataset, construct a second training dataset and a second validation dataset; Perform data augmentation and normalization on the second training dataset to obtain a second augmented training dataset; Perform data augmentation and normalization on the second validation dataset to obtain a second augmented validation dataset; Use the second augmented training dataset to train a pre - constructed coral classification model to obtain a trained coral classification model; Use the second augmented validation dataset to validate the trained coral classification model to obtain a second validation result; If the second validation result indicates that the validation of the trained coral classification model fails, use the object detection loss function to adjust the coral classification model to obtain an adjusted coral classification model; and use the second augmented training dataset to train the adjusted coral classification model to obtain a trained coral classification model; and perform the step of using the second augmented validation dataset to validate the trained coral classification model to obtain a second validation result; If the second validation result indicates that the validation of the trained coral classification model passes, obtain a trained - completed coral classification model.

[0057] Among them, implementing this implementation method effectively improves the quality and diversity of training data through strict pre - processing steps, including deduplication, data augmentation, and normalization, laying a solid foundation for the accurate training of the coral classification model. Using the non - repetitive true coral image dataset to construct the training and validation sets ensures that the training process of the model is not interfered by redundant data and improves the generalization ability of the model. The data augmentation strategy further enriches the training samples and enhances the model's ability to recognize the characteristics of coral images under different conditions. The normalization process helps to eliminate the differences between data and improves the stability and efficiency of model training. In the model validation and adjustment stage, using the object detection loss function to finely tune the model ensures that the trained - completed coral classification model reaches the optimal state in terms of accuracy and robustness, thereby improving the accuracy of coral classification.

[0058] In the embodiments of the present application, to solve the model performance problem caused by data imbalance, the object detection loss function (Focal loss) is used, and its calculation formula is as follows: Among them, is the predicted probability of the model for the correct class of the sample, is an adjustable factor, which is set to 2 in this embodiment.

[0059] Optionally, constructing a non - repetitive true coral image dataset from the classified true coral image dataset specifically includes: Construct an initial non-repetitive true coral image dataset; wherein, the initial non-repetitive true coral image dataset is empty; Successively obtain a target classified true coral image data from the classified true coral image dataset and perform the following operations: Determine a similarity data set between the target classified true coral image data and the non-repetitive true coral image data in the non-repetitive true coral image dataset; wherein, the similarity data set includes global structural similarity and Hamming distance; If there is no global structural similarity in a similarity data set that is less than a first preset threshold and the Hamming distance is greater than a second preset threshold, then delete the target classified true coral image data from the classified true coral image dataset; If there is a global structural similarity in a similarity data set that is less than the first preset threshold and the Hamming distance is greater than the second preset threshold, then add the target classified true coral image data as non-repetitive true coral image data to the non-repetitive true coral image dataset, and delete the target classified true coral image data from the classified true coral image dataset.

[0060] Among them, by implementing this implementation method, through calculating the global structural similarity and Hamming distance, efficient and accurate deduplication of the classified true coral image dataset is achieved. This method not only effectively avoids the interference of redundant data on the training process of the coral classification model, improves the generalization ability and recognition accuracy of the model, but also ensures the high quality of the finally obtained non-repetitive true coral image dataset. By setting the preset thresholds of the global structural similarity and Hamming distance, the strictness of deduplication can be flexibly controlled, avoiding both insufficient sample size caused by excessive deduplication and ensuring the diversity and representativeness of the dataset. This deduplication strategy provides reliable data support for the subsequent training of the coral classification model, helping to improve the accuracy and efficiency of coral classification.

[0061] As an alternative implementation method, the method for determining the similarity data set between the target classified true coral image data and the non-repetitive true coral image data in the non-repetitive true coral image dataset may specifically include: Determine multiple local structural similarities between the target classified true coral image data and the non-repetitive true coral image data in the non-repetitive true coral image dataset; wherein, the target classified true coral image data contains multiple channels, the non-repetitive true coral image data contains multiple channels, and the channels in the target classified true coral image data correspond one-to-one with the channels in the non-repetitive true coral image data, and the local structural similarity is the structural similarity between a corresponding channel of the target classified true coral image data and the non-repetitive true coral image data; Determine the average value of multiple local structure similarities as the global structure similarity; Perform DCT transform on the target classified true coral image data to obtain a first frequency domain matrix; Perform DCT transform on the non-repetitive true coral image data to obtain a second frequency domain matrix; Calculate a first hash sequence of the first frequency domain matrix and a second hash sequence of the second frequency domain matrix; Use the first hash sequence and the second hash sequence to calculate the Hamming distance; Construct a similarity data set using the global structure similarity and the Hamming distance.

[0062] Among them, implementing this implementation mode, by carefully analyzing the local structure similarities of the target classified true coral image data and the non-repetitive true coral image data on each corresponding channel and taking their average value as the global structure similarity, this approach can comprehensively and accurately evaluate the structural similarity degree between two images, effectively avoiding misjudgment caused by differences in a single channel or local features. At the same time, using DCT transform to convert the image to the frequency domain, calculating the hash sequence of the frequency domain matrix, and then obtaining the Hamming distance, this step not only simplifies the process of image feature extraction, but also significantly improves the calculation efficiency, making the deduplication operation faster and more accurate. This method not only ensures the accuracy and efficiency of the deduplication process, but also provides a more pure and high-quality data set for the training of subsequent coral classification models, helping to improve the recognition performance and stability of the model.

[0063] In the embodiments of the present application, the Structural Similarity Index (SSIM) is an index for measuring the similarity between two images. It is mainly used to evaluate the quality of image processing, such as the algorithm performance in fields such as image compression, denoising, and super-resolution. The SSIM index compares the similarity of images from three aspects: brightness, contrast, and structure.

[0064] In practical applications, the value range of SSIM is usually between -1 and 1. When two images are exactly the same, the value of SSIM is 1; when two images are completely different (or one image is the complete negative of the other), the value of SSIM is -1; but in practical applications, due to the characteristics of image data, the value of SSIM rarely falls below 0. Usually, the higher the value of SSIM, the more similar the two images are.

[0065] Therefore, a first preset threshold can be set, and only when the global structure similarity is less than the first preset threshold, can it be considered that the target classified true coral image data corresponding to the global structure similarity and the non-repetitive true coral image data are different.

[0066] In the embodiments of the present application, the calculation formula of the local structural similarity SSIM is as follows: Wherein, represents the channel pixel mean value of the target classified true coral image data, represents the channel pixel mean value of the non-repetitive true coral image data, represents the channel pixel standard deviation of the target classified true coral image data, represents the channel pixel standard deviation of the non-repetitive true coral image data, represents the channel pixel covariance between the target classified true coral image data and the non-repetitive true coral image data, and are stability constants, which are taken as 6.5025 and 58.5225 respectively in this embodiment.

[0067] In the embodiments of the present application, before performing the DCT transformation on the target classified true coral image data and the non-repetitive true coral image data, the two image data can be grayscaled respectively and the size of the image data can be adjusted to 28×28.

[0068] In the embodiments of the present application, the DCT transformation converts the spatial domain signal to the frequency domain to obtain a frequency domain matrix. The calculation formula of the DCT transformation is as follows: Wherein, represents the DCT transformation coefficient of the frequency domain matrix. Since the transformation is performed on a 28×28 image, the values of M and N are both 28, and are both orthogonalization factors, represents the value at the corresponding position of the image matrix after grayscaling and size adjustment.

[0069] In the embodiments of the present application, the method for calculating the first hash sequence of the first frequency domain matrix can be implemented by the perceptual hash algorithm. Specifically, it can be: Only keep the low-frequency part of the upper left 8×8 in the first frequency domain matrix; Calculate the mean value of the low-frequency part of the first frequency domain matrix; At the same time, traverse each value in the low-frequency part of the first frequency domain matrix. If the value is greater than or equal to the mean value, set the value to 1, otherwise set it to 0; Finally, combine the comparison results together to form the first hash sequence of a 64-bit binary integer of the first frequency domain matrix.

[0070] The calculation method of the second hash sequence of the second frequency domain matrix is the same as that of the first hash sequence of the first frequency domain matrix, which will not be elaborated here.

[0071] In the embodiments of the present application, the Hamming distance between the first hash sequence and the second hash sequence is the number of different binary integers at all corresponding positions in the first hash sequence and the second hash sequence.

[0072] Obviously, the larger the Hamming distance, the greater the difference between the target classified true coral image data and the non-repeated true coral image data; therefore, a second preset threshold can be set, and only when the Hamming distance is greater than the second preset threshold, can it be considered that the target classified true coral image data corresponding to the Hamming distance and the non-repeated true coral image data are different.

[0073] Therefore, if the global structural similarity of a similarity data set is less than the first preset threshold and the Hamming distance is greater than the second preset threshold, it means that the similarity between the target classified true coral image data corresponding to the similarity data set and the non-repeated true coral image data is small, and they are two different image data. Therefore, the target classified true coral image data can be added to the non-repeated true coral image data set as non-repeated true coral image data, and the target classified true coral image data can be deleted from the classified true coral image data set; If the global structural similarity of a similarity data set is greater than or equal to the first preset threshold or the Hamming distance is less than or equal to the second preset threshold, it means that the similarity between the target classified true coral image data corresponding to the similarity data set and the non-repeated true coral image data is large, and they are two relatively similar image data. Therefore, the target classified true coral image data cannot be added to the non-repeated true coral image data set, and at this time, the target classified true coral image data needs to be deleted from the classified true coral image data set.

[0074] To further illustrate the application effect of the present invention, the experimental results of the embodiments of the present invention are shown below: The experimental environment uses PyTorch 1.10.0 as the deep learning framework, runs on Python 3.8 and Ubuntu20.04 operating systems, and uses CUDA 11.3 for GPU acceleration. The hardware configuration includes a GPU of NVIDIA RTX 4090 (24GB video memory), equipped with an AMD EPYC 7T83 64-core processor (22 vCPUs) and 90GB of memory.

[0075] In the embodiments of the present application, a total of 7323 network coral images and 3648 network non-coral images are crawled using web crawler technology.

[0076] Please refer to Figures 7 - 10 ,Figure 7 The loss visualization graph obtained by training the coral recognition model using the first training dataset; Figure 8 The schematic diagram of the recognition accuracy rate obtained by training the coral recognition model using the first training dataset; Figure 9 The loss visualization graph obtained by training the coral recognition model using the first validation dataset; Figure 10 The schematic diagram of the recognition accuracy rate obtained by training the coral recognition model using the first validation dataset; In the embodiment of the present application, when training the ResNet18 network, the Epoch is set to 80, the Batch size is set to 64, the initial learning rate is set to 0.001, the optimizer adopted is Adam, and the learning rate decay strategy adopted in the training is: if there is no higher accuracy rate calculated in the validation set for 5 consecutive Epochs, the learning rate is updated with a multiplication factor of 0.1. The results are as Figures 7 - 10 shown, where the optimal accuracy rate of the validation set reaches 95.43%. Finally, the model with the best performance in the validation set is used to classify the network coral images, and 4219 real network coral images and 3104 fake network coral images are classified.

[0077] In the embodiment of the present invention, a total of 231 duplicate images in 4219 real network coral images are removed by combining the SSIM algorithm and the perceptual hashing algorithm and setting thresholds respectively.

[0078] Please refer to Figures 11 - 14 , Figure 11 The loss visualization graph obtained by training the coral recognition model using the second training dataset; Figure 12 The schematic diagram of the recognition accuracy rate obtained by training the coral recognition model using the second training dataset; Figure 13 The loss visualization graph obtained by training the coral recognition model using the second validation dataset; Figure 14 The schematic diagram of the recognition accuracy rate obtained by training the coral recognition model using the second validation dataset; In the embodiment of the present invention, when training the improved ResNet50, the Epoch is set to 100, the Batch size is set to 32, the initial learning rate is set to 0.001, the optimizer adopted is Adam, and the learning rate decay strategy adopted in the training is: if there is no higher accuracy rate calculated in the validation set for 5 consecutive Epochs, the learning rate is updated with a multiplication factor of 0.1. The results are as Figures 11 - 14 shown, where the optimal accuracy rate of the validation set reaches 67.85%.

[0079] Finally, the specific results of the fine-grained classification of 3988 real network coral image datasets using the model with the best performance in the validation set are as Figure 15 shown.

[0080] By implementing the above steps 101 to 105, the coral image data obtained from the image search engine can be recognized and the coral species can be determined through a pre-trained model, thereby improving the accuracy of recognizing and classifying real coral image data. In addition, this application can also ensure that the trained coral recognition model has high accuracy and stability. In addition, this application can also improve the practicality of data processing and the model performance. In addition, this application can also ensure that the trained coral classification model reaches the optimal state in terms of accuracy and robustness, thereby improving the accuracy of coral classification. In addition, this application can flexibly control the strictness of duplicate removal, avoiding insufficient sample size caused by excessive duplicate removal and ensuring the diversity and representativeness of the data set. In addition, this application can also provide a more pure and high-quality data set for the subsequent training of the coral classification model, which helps to improve the recognition performance and stability of the model.

[0081] Based on the same inventive concept, the embodiment of this application also provides a real coral image data determination device for implementing the real coral image data determination method involved above. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the real coral image data determination device provided below can refer to the limitations on the real coral image data determination method in the above text and will not be repeated here.

[0082] In an exemplary embodiment, as Figure 16 shown, a real coral image data determination device is provided, including: An acquisition unit 1601, configured to obtain initial coral image data from an image search engine based on a preset keyword; A first input unit 1602, configured to input the initial coral image data into a pre-trained coral recognition model to obtain a coral recognition result output by the coral recognition model; A data enhancement unit 1603, configured to perform data enhancement on the real coral image data if the coral recognition result indicates that the initial coral image data is real coral image data, to obtain real coral image enhanced data; A normalization unit 1604, configured to normalize the real coral image enhanced data to obtain real coral image normalized data; A second input unit 1605, configured to input the real coral image normalized data into a pre-trained coral classification model to obtain a coral classification result of the real coral image data output by the coral classification model.

[0083] By implementing the above-described embodiments, the coral image data obtained from the image search engine can be recognized by a pre-trained model and the coral species can be determined, thereby improving the accuracy of recognizing and classifying real coral image data.

[0084] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structural diagram can be as Figure 17 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the data for determining real coral image data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for determining real coral image data.

[0085] Those skilled in the art can understand that Figure 17 the structure shown in

[0086] is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0087] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, it implements the steps in the above method embodiments.

[0088] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, it implements the steps in the above method embodiments.

[0089] In an exemplary embodiment, a chip is provided. The chip includes a processor and a communication interface, and the communication interface is coupled to the processor. The processor is configured to run programs or instructions to implement the steps in the above method embodiments, and can achieve the same technical effects. To avoid repetition, details are not described herein again.

[0090] It should be understood that the chip mentioned in the embodiments of the present application may also be referred to as a system-on-chip, system chip, chip system, or system-on-a-chip, etc.

[0091] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0092] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it may include the processes of the above method embodiments. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application may include at least one of non-volatile and volatile memories. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0093] In each of the embodiments provided in this application, the database involved may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., without limitation. In each of the embodiments provided in this application, the processor may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without limitation.

[0094] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0095] Specific examples are used in this article to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for determining real coral image data, characterized in that: The real coral image data determination method comprises: Obtaining initial coral image data from an image search engine based on preset keywords; Inputting the initial coral image data into a pre-trained coral recognition model to obtain a coral recognition result output by the coral recognition model; If the coral recognition result indicates that the initial coral image data is real coral image data, data enhancement is performed on the real coral image data to obtain real coral image enhanced data; Normalizing the real coral image enhancement data to obtain real coral image normalized data; The normalized data of the real coral image is input into a pre-trained coral classification model to obtain a coral classification result of the real coral image data output by the coral classification model.

2. The method for determining real coral image data according to claim 1, characterized in that: The performing data enhancement on the real coral image data to obtain real coral image enhanced data specifically includes: The real coral image data is cropped to obtain central area image data; wherein the size of the central area image data is a first preset size; Performing random probability rotation on the central area image data to obtain rotated image data; Performing random probability brightness enhancement on the rotated image data to obtain first enhanced image data; Performing random probability color enhancement on the first enhanced image data to obtain second enhanced image data; performing random probability contrast enhancement on the second enhanced image data to obtain third enhanced image data; Performing random probability sharpening enhancement on the third enhanced image data to obtain fourth enhanced image data; The fourth enhanced image data is cropped to obtain real coral image enhanced data; wherein the size of the real coral image enhanced data is a second preset size, and the second preset size is smaller than the first preset size.

3. The method for determining real coral image data according to claim 1, characterized in that: The training method of the coral recognition model is specifically as follows: Acquire a non-coral image dataset and a classified real coral image dataset; wherein the non-coral image dataset includes a plurality of non-coral image data, and the classified real coral image dataset includes a plurality of classified real coral image data; Based on the non-coral image dataset and the classified real coral image dataset, construct a first training dataset and a first validation dataset; Using the first training data set to train a pre-built coral recognition model to obtain a trained coral recognition model; Using the first verification data set to verify the trained coral recognition model, to obtain a first verification result; If the first verification result indicates that the trained coral identification model verification fails, the trained coral identification model is adjusted using a binary cross entropy loss function to obtain an adjusted coral identification model; the adjusted coral identification model is trained using the first training data set to obtain a trained coral identification model; and the trained coral identification model is verified using the first verification data set to obtain a first verification result. If the first verification result indicates that the trained coral identification model passes the verification, a pre-trained coral identification model is obtained.

4. The method for determining real coral image data according to claim 3, characterized in that: The training method of the coral classification model is specifically as follows: Constructing a non-repeating real coral image dataset from the classified real coral image dataset; wherein the non-repeating real coral image dataset contains at least one non-repeating real coral image data, the non-repeating real coral image data is obtained from the classified real coral image dataset, and any two non-repeating real coral image data in the non-repeating real coral image dataset are different; Based on the non-repetitive real coral image dataset, construct a second training dataset and a second verification dataset; Performing data augmentation and normalization on the second training data set to obtain a second enhanced training data set; Performing data augmentation and normalization on the second verification data set to obtain a second enhanced verification data set; Using the second enhanced training data set to train the pre-built coral classification model to obtain a trained coral classification model; Using the second enhanced verification data set to verify the trained coral classification model, to obtain a second verification result; If the second verification result indicates that the trained coral classification model has failed verification, the coral classification model is adjusted using the target detection loss function to obtain an adjusted coral classification model; the adjusted coral classification model is trained using the second enhanced training data set to obtain a trained coral classification model; and the trained coral classification model is verified using the second enhanced verification data set to obtain a second verification result. If the second verification result indicates that the trained coral classification model has passed the verification, a trained coral classification model is obtained.

5. The method for determining real coral image data according to claim 4, characterized in that: The step of constructing a non-repetitive real coral image dataset from the classified real coral image dataset specifically includes: Constructing an initial non-repetitive real coral image dataset; wherein the initial non-repetitive real coral image dataset is empty; Obtain a target classified real coral image data from the classified real coral image data set in turn and perform the following operations: Determine a similarity data set between the target classified real coral image data and the non-repeated real coral image data in the non-repeated real coral image data set; wherein the similarity data set includes global structural similarity and Hamming distance; If there is no similarity data set whose global structural similarity is less than the first preset threshold and whose Hamming distance is greater than the second preset threshold, deleting the target classified real coral image data from the classified real coral image data set; If there exists a similarity data set whose global structural similarity is less than the first preset threshold and whose Hamming distance is greater than the second preset threshold, the target classified real coral image data is added to the non-duplicate real coral image data set as non-duplicate real coral image data, and the target classified real coral image data is deleted from the classified real coral image data set.

6. The method for determining real coral image data according to claim 5, characterized in that: The determining of a similarity data set between the target classified real coral image data and the non-repeating real coral image data in the non-repeating real coral image data set specifically includes: Determine multiple local structural similarities between the target classified real coral image data and the non-repeating real coral image data in the non-repeating real coral image data set; wherein the target classified real coral image data includes multiple channels, the non-repeating real coral image data includes multiple channels, and the channels in the target classified real coral image data correspond to the channels in the non-repeating real coral image data one by one, and the local structural similarity is the structural similarity of a corresponding channel between the target classified real coral image data and the non-repeating real coral image data; The average value of multiple local structural similarities is determined as the global structural similarity; Performing DCT transformation on the target classified real coral image data to obtain a first frequency domain matrix; Performing DCT transformation on the non-repetitive real coral image data to obtain a second frequency domain matrix; Calculate and obtain a first hash sequence of the first frequency domain matrix and a second hash sequence of the second frequency domain matrix; Using the first hash sequence and the second hash sequence, calculate a Hamming distance; A similarity data set is constructed using the global structural similarity and the Hamming distance.

7. A device for determining real coral image data, characterized in that: The real coral image data determining device comprises: An acquisition unit, used for acquiring initial coral image data from an image search engine based on preset keywords; A first input unit, used to input the initial coral image data into a pre-trained coral recognition model to obtain a coral recognition result output by the coral recognition model; a data enhancement unit, configured to perform data enhancement on the real coral image data to obtain real coral image enhanced data if the coral recognition result indicates that the initial coral image data is real coral image data; A normalization unit, used for normalizing the real coral image enhancement data to obtain real coral image normalized data; The second input unit is used to input the normalized data of the real coral image into a pre-trained coral classification model to obtain a coral classification result of the real coral image data output by the coral classification model.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for determining real coral image data according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for determining real coral image data according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for determining real coral image data according to any one of claims 1 to 6 are implemented.

Images