Construction method and equipment of oncomelania subspecies fine-grained recognition model and storage medium

Abstract

The invention discloses an oncomelania subspecies fine-grained recognition model construction method and device and a storage medium, and relates to the technical field of artificial intelligence, and the oncomelania subspecies fine-grained recognition model construction method comprises the steps: building a classification image data set of oncomelania subspecies and simulated oncomelania through sample collection; preprocessing the classified image data set to obtain a target classified image data set; pre-training the initial model based on the standard data set to obtain a target pre-training weight; and importing the target pre-training weight into the pre-trained initial model, and training by using the target classification image data set to obtain an oncomelania subspecies fine-grained recognition model. The technical defect that the oncomelania identification effect is poor is overcome, and the oncomelania type identification capability of the model is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device and storage medium for constructing a fine-grained recognition model for Oncomelania snail subspecies. Background Art

[0002] Yunnan Province is an endemic area for schistosomiasis in my country. The snails distributed here are mainly the Dianchuan subspecies of Oncomelania hupensis, whose external morphology is very similar to that of Tricula, another intermediate host of the parasite in the endemic area of ​​Yunnan Province. In the past, identification was mainly carried out through morphological characteristics to identify snails. The accuracy of this identification method is highly dependent on the experience and professional knowledge of the staff, which affects the rapid and accurate assessment of the snail situation.

[0003] In recent years, convolutional neural network models have been successfully used for intelligent identification of snails. Among the related technologies, the Inception ResNet V2 model was mainly constructed to identify Oncomelania hupensis and its similar snails, and to determine the optimal learning strategy. The snail identification performance of the constructed deep learning model is similar to that of snail inspectors.

[0004] However, the data set used in the training of the models in the related technologies is not comprehensive enough, resulting in poor recognition of snails. Summary of the invention

[0005] The main purpose of the present application is to provide a method, device and storage medium for constructing a fine-grained recognition model of Oncomelania snail subspecies, aiming to solve the technical problem of poor recognition effect of Oncomelania snails.

[0006] To achieve the above objectives, the present application proposes a method for constructing a fine-grained recognition model for Oncomelania snail subspecies, the method comprising:

[0007] Through sample collection, a classification image dataset of Oncomelania snail subspecies and Oncomelania pseudo-oncomelania was established;

[0008] Preprocessing the classified image data set to obtain a target classified image data set;

[0009] Pre-train the initial model based on the standard data set to obtain the target pre-training weights;

[0010] The target pre-training weights are imported into the pre-trained initial model, and the target classification image data set is used for training to obtain a fine-grained recognition model for snail subspecies.

[0011] In one embodiment, after the step of importing the target pre-trained weight into the pre-trained initial model and using the target classification image data set for training to obtain a fine-grained recognition model for Oncomelania subspecies, the method further includes:

[0012] Using a preset training strategy to train and compare the snail subspecies fine-grained recognition model to obtain a target strategy;

[0013] The target strategy is introduced into the Oncomelania snail subspecies fine-grained recognition model to obtain a target model.

[0014] In one embodiment, the training strategy includes data enhancement conditions and hyperparameters, the data enhancement conditions include basis, mixing and Gaussian blur, the hyperparameters include optimizer, loss function and learning rate decay strategy, and the step of training and comparing the Oncomelania subspecies fine-grained recognition model using a preset training strategy to obtain a target strategy includes:

[0015] Keeping the hyperparameters in the Oncomelania snail subspecies fine-grained identification model unchanged, a standard model is obtained;

[0016] Importing the classified image data sets under different data enhancement conditions into the standard model for training comparison, and screening out target data enhancement conditions;

[0017] Keep the data enhancement conditions of the classification image dataset unchanged to obtain a standard classification image dataset

[0018] Importing the standard classification image data into the Oncomelania snail subspecies fine-grained recognition model under different hyperparameters for training and comparison, and screening out target hyperparameters;

[0019] The target data enhancement condition and the target hyperparameter are used as the target strategy.

[0020] In one embodiment, the step of establishing a classification image dataset of Oncomelania hupensis subspecies and Oncomelania pseudo-Helium through sample collection comprises:

[0021] For several samples of Oncomelania hupensis and Oncomelania pseudo-oncomelania, sample images of the samples are collected in four directions: shell opening upward, shell opening downward, shell opening to the left, and shell opening to the right;

[0022] Classify according to the shell opening direction in the sample image to obtain a basic image data set;

[0023] The classified image dataset is obtained by classifying the snails according to the types of snails in the basic image dataset.

[0024] In one embodiment, before the step of pre-training the initial model based on the standard data set to obtain the target pre-training weight, the step further includes:

[0025] Taking the target object in the classification image dataset as the center, cutting out the background area without information in the image, and obtaining a standard classification image dataset;

[0026] Performing pixel scaling on the standard classification image data set to obtain an initially processed classification image data set;

[0027] A randomization operation is performed on the initially processed classified image data set to obtain the target classified image data set.

[0028] In one embodiment, the step of performing a randomization operation on the initially processed classified image data set to obtain the target classified image data set includes:

[0029] Randomly rotating the images in the initially processed classified image data set to obtain a first classified image data set;

[0030] Randomly adjusting the sharpness, brightness, and contrast of the images in the first classified image data set to obtain a second classified image data set;

[0031] The images in the second classified image data set are randomly gray-scaled to obtain the target classified image data set.

[0032] In one embodiment, the step of importing the target pre-trained weight into the pre-trained initial model and using the classified image data set for training to obtain the snail subspecies fine-grained recognition model further includes:

[0033] Importing the target pre-trained weights into the pre-trained initial model to obtain a basic model;

[0034] The data of the basic model is updated, and the target classification image data set is used for training to obtain a fine-grained recognition model for snail subspecies.

[0035] In one embodiment, the step of pre-training the initial model based on the standard data set to obtain the target pre-training weights includes:

[0036] By importing the target classification image dataset into the initial model pre-training, the basic pre-training weights are obtained;

[0037] The basic pre-training weights are updated and adjusted to obtain the target pre-training weights.

[0038] In addition, to achieve the above-mentioned objectives, the present application also proposes a device for a fine-grained identification model of Oncomelania subspecies, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of a method for constructing a fine-grained identification model of Oncomelania subspecies as described in any of the above items.

[0039] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the method for constructing a fine-grained identification model of snail subspecies as described in any of the above items are implemented.

[0040] The present application proposes a method, device and storage medium for constructing a fine-grained recognition model for Oncomelania subspecies. The present application establishes a classification image dataset of Oncomelania subspecies and Oncomelania pseudo-Oncomelania through sample collection; preprocesses the classification image dataset to obtain a target classification image dataset; pretrains the initial model based on a standard dataset to obtain a target pretraining weight; imports the target pretraining weight into the pretrained initial model, and uses the target classification image dataset for training to obtain a fine-grained recognition model for Oncomelania subspecies. Therefore, the present application further increases the training data of the fine-grained recognition model for Oncomelania subspecies by training the model multiple times with the assistance of a large number of Oncomelania subspecies and snail databases, thereby improving the model's ability to recognize Oncomelania subspecies.

[0041] In summary, the present application combines a large number of databases of Oncomelania snail subspecies and snails to train a model that can identify Oncomelania snail subspecies and make its identification ability meet the expected results. Multiple training is performed using a large number of processed databases, which overcomes the technical defects of the current model in that the recognition effect of Oncomelania snails is poor and improves the model's ability to recognize Oncomelania snail species. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 A schematic diagram of a process for constructing a fine-grained identification model of Oncomelania subspecies provided in Example 1 of the present application;

[0045] Figure 2 A schematic diagram of a refinement process provided in Example 2 of the method for constructing a fine-grained identification model of Oncomelania subspecies in this application;

[0046] Figure 3 A schematic diagram of a process for constructing a fine-grained identification model of Oncomelania subspecies provided in Example 3 of the present application;

[0047] Figure 4ROC curves of different training strategies provided in Example 3 of the method for constructing a fine-grained recognition model of Oncomelania subspecies in this application;

[0048] Figure 5 A curve chart showing the variation of the accuracy of different training strategies provided in the third implementation of the method for constructing a fine-grained identification model for Oncomelania subspecies in this application;

[0049] Figure 6 A graph showing the change in loss values ​​of different training strategies provided in Example 3 of the method for constructing a fine-grained recognition model for Oncomelania subspecies of this application;

[0050] Figure 7 Evaluation index diagrams of different models provided in Example 3 of the method for constructing a fine-grained identification model of Oncomelania subspecies in this application;

[0051] Figure 8 A schematic diagram of a process for constructing a fine-grained identification model of Oncomelania subspecies provided in Example 4 of the present application;

[0052] Fig. 9 A schematic diagram of a process for constructing a fine-grained identification model of Oncomelania snail subspecies according to Embodiment 5 of the present application;

[0053] Fig.10 A schematic diagram of a process for constructing a fine-grained identification model of Oncomelania snail subspecies according to Example 6 of the present application;

[0054] Fig.11 A schematic diagram of a process for constructing a fine-grained identification model of Oncomelania snail subspecies according to Embodiment 8 of the present application;

[0055] Fig.12 This is a schematic diagram of the structure of the device for the fine-grained identification model of Oncomelania snail subspecies in this application.

[0056] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0057] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0058] In the related art, the Inception ResNet V2 model was mainly constructed to identify Oncomelania hupensis and its similar snails, and to determine the optimal learning strategy. The snail recognition performance of the constructed deep learning model is similar to that of snail inspectors. However, the data set used in the training of the model in the related art is not comprehensive enough, resulting in poor recognition of Oncomelania hupensis.

[0059] The present application provides a solution: establish a classification image dataset of Oncomelania hupensis subspecies and Oncomelania pseudo-Hails through sample collection; preprocess the classification image dataset to obtain a target classification image dataset; pretrain the initial model based on the standard dataset to obtain a target pretraining weight; import the target pretraining weight into the pretrained initial model, and use the target classification image dataset for training to obtain a fine-grained recognition model for Oncomelania hupensis subspecies. The present application combines a large number of databases of Oncomelania hupensis subspecies and snails to train a model that can identify Oncomelania hupensis subspecies and make its recognition ability meet the expected effect. It uses a large number of processed databases for multiple trainings, overcomes the technical defect of the current model's poor recognition effect on Oncomelania hupensis snails, and improves the model's ability to recognize Oncomelania hupensis species.

[0060] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a device of a fine-grained identification model of Oncomelania subspecies, etc. The following takes the device of a fine-grained identification model of Oncomelania subspecies as an example to illustrate this embodiment and the following embodiments.

[0061] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0062] The present application embodiment provides a method for constructing a fine-grained recognition model for Oncomelania subspecies, referring to Figure 1 The method for constructing a fine-grained recognition model for Oncomelania snail subspecies includes steps S10 to S40:

[0063] Step S10, establishing a classification image dataset of Oncomelania hupensis subspecies and Oncomelania pseudo-hupensis through sample collection;

[0064] In the present embodiment, Oncomelania hupensis is an important mollusk, which is the only intermediate host of Schistosoma japonicum. Oncomelania hupensis has male and female, is an amphibious organism, and is composed of two parts: a shell and a soft body. The soft body part includes a head, a neck, a foot and a mantle, and the shell can be divided into types such as a rib shell Oncomelania hupensis according to surface features. Oncomelania hupensis is a type of mollusk living in freshwater environments, belonging to the class Gastropoda, the order Mesogastropoda, and the family Ophiopodidae, and has many species, such as Oncomelania hupensis, Oncomelania fujianensis, and Oncomelania ninghaiensis.

[0065] As an optional implementation method, as the basis for data set construction, field collection of Oncomelania hupensis and Oncomelania pseudo-Helium snails is involved to ensure the representativeness and diversity of the samples.

[0066] Step S20, preprocessing the classification image data set to obtain a target classification image data set;

[0067] In this embodiment, classified image datasets are a broad category that refers to datasets containing classified or labeled images, which are usually used to train, validate, and test machine learning or deep learning models. Classified image datasets can cover a variety of topics and fields, such as natural scenes, object recognition, face recognition, etc., and their size, annotation accuracy, and image quality may vary.

[0068] As an optional implementation, a preprocessing operation is performed on the classification image dataset to generate a target classification image dataset that meets specific requirements or conditions.

[0069] Step S30, pre-training the initial model based on the standard data set to obtain the target pre-training weight;

[0070] In this embodiment, the pre-trained weights are model parameters that have been trained on large-scale datasets, including weight matrices and bias vectors. Standard datasets such as ImageNet and COCO cover millions to billions of images or texts. The pre-trained weights contain rich information and help the model achieve better performance on the target task. By fine-tuning the pre-trained weights, the model can converge to good performance faster and shorten the training time.

[0071] As an optional implementation, the image dataset needs to be preprocessed first, and then the initial model is pretrained using the processed dataset to obtain the initial weights of the model. This process is an important step in model training, which helps the model better learn data features and provides a basis for subsequent training and tuning.

[0072] Step S40, importing the target pre-trained weights into the pre-trained initial model, and using the target classification image data set for training to obtain a fine-grained recognition model for Oncomelania subspecies.

[0073] In this embodiment, the pre-trained weights are adjusted appropriately and then imported into the pre-trained initial model for training, so that the model can be trained faster and more accurately. The model is trained in large quantities through the classification image set, so that it can accurately identify Oncomelania snails and Oncomelania snail subspecies.

[0074] As an optional implementation, the initial model is pre-trained based on a standard data set to obtain target pre-training weights, which is regarded as the first pre-training to obtain a pre-trained initial model; then the target pre-training weights are imported into the pre-trained initial model to obtain a pre-trained model; and then the pre-trained model is trained using a target classification image data set to obtain a fine-grained recognition model for snail subspecies.

[0075] As an optional implementation, before establishing a classification image dataset, a large number of various snail samples should be collected to ensure the diversity of image data and avoid randomness in the identification of the trained model. The classification image dataset is then preprocessed to remove unclear and defective images, and the target classification image dataset is imported into the initial model for pre-training to obtain target pre-training weights. The pre-training weights are fine-tuned and imported into the pre-trained initial model, and the target classification image dataset is used for training to obtain a fine-grained recognition model for snail subspecies.

[0076] Exemplarily, several Oncomelania subspecies and Pseudomonas snails are collected, photographed or video recorded, and a classification data set is established after obtaining the images and videos. The classification data set is preprocessed and then imported into the model for training to obtain target pre-training weights. Finally, the pre-training weights and the classification data set are imported into the initial model after pre-training for training to obtain a fine-grained recognition model for Oncomelania subspecies.

[0077] By using a large number of databases for model training and retraining, the ability of the snail subspecies fine-grained recognition model to identify snail species has been improved.

[0078] Based on any of the above embodiments, in Embodiment 2 of the present application, refer to Figure 2 , after step S30, further comprising steps S50 to S60:

[0079] Step S50, using a preset training strategy to train and compare the snail subspecies fine-grained recognition model to obtain a target strategy;

[0080] In this embodiment, the training strategy includes different data enhancement conditions and different hyperparameter conditions. The fine-grained recognition model of snail subspecies is trained through different training strategies, and the training results are compared to find the target strategy with the best recognition effect.

[0081] As an optional implementation, in the process of training the fine-grained recognition model of Oncomelania subspecies, researchers will use a variety of preset training strategies and compare the effects of these strategies. Through comparison, the optimal training strategy, namely the target strategy, can be selected to further improve the recognition performance and accuracy of the model. This process is an important part of model optimization and selection, which helps to improve the effect of the model in practical applications.

[0082] Step S60, importing the target strategy into the snail subspecies fine-grained recognition model to obtain a target model.

[0083] In this embodiment, the target strategy refers to the target data enhancement method and target hyperparameters obtained in the experiment. The model after importing the target data enhancement method and target hyperparameters is the target model, and the recognition of the target model is more accurate.

[0084] As an optional implementation method, during the training process, different training strategies are adopted, the principle of controlling one variable is followed, training comparison is performed, a target variable is selected, and then the next variable is replaced for training comparison, and so on. Through experimental comparison, the target strategy is found, and the target strategy is introduced into the fine-grained recognition model of Oncomelania subspecies to obtain the target model.

[0085] Exemplarily, the preset training strategy includes 7 groups of different training strategies according to different data enhancement conditions and different hyperparameters. Groups 1 to 4 compare the basic, basic + hybrid, basic + Gaussian blur and basic + hybrid + Gaussian blur data enhancement conditions, and keep the model hyperparameters as Adam optimizer + Focal Loss + CosineAnnealingLR; Groups 4 to 7 compare the optimizer, loss function, and learning rate decay strategy, respectively, and keep the image processing method as basic + hybrid + Gaussian blur. Groups 4-7 compare the optimizer, loss function, and learning rate decay strategy, respectively, and keep the image processing method unchanged. The data enhancement method in group 1 is based, and the model hyperparameters are Adam optimizer + FocalLoss + CosineAnnealingLR; the data enhancement method in group 2 is based + hybrid, and the model hyperparameters are Adam optimizer + FocalLoss + Cosine AnnealingLR; the data enhancement method in group 3 is based + Gaussian blur, and the model hyperparameters are Adam optimizer + FocalLoss + CosineAnnealingLR; the data enhancement method in group 4 is based + hybrid + Gaussian blur, and the model hyperparameters are Adam optimizer + FocalLoss + CosineAnnealingLR; the data enhancement method in group 5 is based + hybrid + Gaussian blur, and the model hyperparameters are SGD optimizer + FocalLoss + Cosine neAnnealingLR; the data enhancement method in group 6 is base+mixing+Gaussian blur, and the model hyperparameters are Adam optimizer+CrossEntropyLoss+CosineAnnealingLR; the data enhancement method in group 7 is base+mixing+Gaussian blur, and the model hyperparameters are Adam optimizer+FocalLoss+MultiStepLR.

[0086] By comparing the training of various training strategies, the target strategy was obtained, which improved the recognition accuracy and efficiency of the fine-grained recognition model of the snail subspecies.

[0087] Based on any of the above embodiments, in Embodiment 3 of the present application, the training strategy includes data enhancement conditions and hyperparameters, the data enhancement conditions include basis, mixing and Gaussian blur, and the hyperparameters include optimizer, loss function and learning rate decay strategy, refer to Figure 3 , after step S50, further comprising steps A11 to A14:

[0088] Step A11, keeping the hyperparameters in the Oncomelania snail subspecies fine-grained identification model unchanged to obtain a standard model;

[0089] In this embodiment, the hyperparameters are not learned by the learning algorithm itself, but are parameters that need to be manually adjusted. The hyperparameters include the optimizer, the loss function, and the learning rate decay strategy.

[0090] As an optional implementation, in the task of fine-grained identification of Oncomelania subspecies, in order to obtain a stable and reliable standard model, the hyperparameters in the model need to be kept unchanged. In this way, under the same hyperparameter settings, a standard model with consistent performance and recognition ability can be obtained through training. This process is an important part of model development and optimization, which helps to improve the effect of the model in practical applications.

[0091] Step A12, importing the classified image data sets under different data enhancement conditions into the standard model for training comparison, and screening out target data enhancement conditions;

[0092] In this embodiment, the data enhancement conditions include basic, mixed and Gaussian blur. The classified image data set is adjusted by combining or singly utilizing different data enhancement conditions and then imported into the model for training comparison. The most suitable data enhancement condition is screened out based on the comparison results.

[0093] As an optional implementation, researchers will import classified image datasets processed under different data augmentation conditions into the trained standard model and conduct training comparisons. By comparing the training effects under different data augmentation conditions, the optimal data augmentation conditions, i.e., the target conditions, can be screened out. This process helps optimize the model training process and improve the performance and accuracy of the model in practical applications.

[0094] For example, Figure 5 The comparison between different data augmentation conditions is shown in Groups 1 to 4. Each group achieves a high accuracy at the beginning of training, reaches a plateau at around 110 rounds, and then gradually converges. The accuracy curve of Group 4 is the highest. The curve shows that mixed data augmentation has poor stability and slow convergence in the early stage of training. The loss curve reaches a low level at the beginning and remains stable during the training process. There is little difference between different data augmentation conditions.

[0095] Figure 7The performance of the external test set is shown in Figure 1. Group 1, which uses only ordinary data enhancement, performs the worst. Using hybridization can improve model performance, but the effect is not as obvious as using Gaussian blur. The combined use of hybridization and Gaussian blurring for data enhancement has the best effect, and the F1 index and Youden index comprehensive indicators are also the highest. The confusion matrix shows that when only ordinary data enhancement and only hybridization are used, the model identifies most images as snails, resulting in a high recall rate.

[0096] It should be noted that Figure 5 The horizontal axis is the time segment, and the vertical axis is the accuracy. Figure 6 The horizontal axis is the time segment, and the vertical axis is the loss value. Figure 7 The horizontal axis is the group number, and the vertical axis is the percentage of the evaluation index. Figure 4 Both the horizontal and vertical axes represent specificity.

[0097] Step A13, keeping the data enhancement conditions of the classification image dataset unchanged, to obtain a standard classification image dataset;

[0098] In this embodiment, data enhancement, also called data amplification or augmentation, is a technique for expanding training samples by transforming data. It aims to increase the diversity and quantity of training samples so that the model can learn more generalized features. It includes homogeneous data enhancement and mixed-class data enhancement. Homogeneous data enhancement includes flipping, rotating, scaling, shifting, blurring, etc.; mixed-class data enhancement includes mixing, etc. In addition, there are methods such as random cropping, brightness adjustment, contrast adjustment, and adding noise.

[0099] As an optional implementation, by comparing the model performance under different data augmentation conditions, the data augmentation conditions that are most suitable for the current task and goal can be found. This screening process is of great significance for optimizing model training and improving the effectiveness of the model in practical applications. It helps ensure that the model can learn the most extensive and representative features, so as to better cope with various situations in practical applications.

[0100] Step A14, importing the standard classification image data into the Oncomelania snail subspecies fine-grained recognition model under different hyperparameters for training and comparison, and screening out target hyperparameters;

[0101] In this embodiment, by using the determined target data enhancement conditions to set the classification image data set, then inputting different hyperparameters into the model, and finally importing the classification image data set for training comparison, the target hyperparameters are obtained according to the comparison results. Since different data enhancement conditions can be superimposed and have different effects on the model, superimposing multiple data enhancement conditions can explore the optimal strategy in model training.

[0102] The target data enhancement condition and the target hyperparameter are used as the target strategy.

[0103] As an optional implementation method, a fine-grained recognition model and system for Oncomelania snail subspecies is obtained by training the model, and a target model is obtained by using a variety of training strategies, thereby improving the recognition stability of the fine-grained recognition model for Oncomelania snail subspecies. Through the recognition model and system, intelligent visual recognition of Oncomelania snail subspecies is realized, solving the problem of lack of relevant research on intelligent visual recognition of Oncomelania snail subspecies and reducing the difficulty of prevention and control of Oncomelania snail subspecies.

[0104] Exemplarily, groups 4 to 7 are models with different optimizers, loss functions, and learning rate decay strategies. In terms of optimizers, the SGD optimizer is not as effective as the Adam optimizer. The accuracy curve shows that the accuracy of the validation set reaches a maximum of 94.03% in 47 rounds, then gradually decreases and does not converge. The loss curve drops to the minimum around 40 rounds, and then increases, showing a similar trend. The accuracy is lowest in the external test set, indicating that the model is overfitting. In terms of loss functions, for groups 4 and 6, the accuracy curve using the CrossEntropyLoss function converges faster than FocalLoss, but is not as stable as FocalLoss. The loss value using the CrossEntropyLoss function is higher in the early stage of training, and higher than FocalLoss after stabilization, and the curve stability is not as good as FocalLoss. Groups 4 and 7 compared different learning rate decay strategies, Figure 5 The accuracy curve using the MultiStepLR decay strategy begins to converge around 60 rounds, with the fastest convergence speed and maintained at the highest level. There is no obvious difference between the two groups of loss curves. In the external test set, the accuracy, F1 index, AUC and Youden index of the MultiStepLR decay strategy are slightly improved compared to CosineAnnealingLR.

[0105] By comparing the training of various training strategies, the target strategy was obtained, which improved the recognition accuracy and efficiency of the fine-grained recognition model of the snail subspecies.

[0106] Based on any of the above embodiments, in Embodiment 4 of the present application, refer to Figure 8 , the step S10 further includes steps B11 to B13:

[0107] Step B11, for a plurality of samples of Oncomelania hupensis and Oncomelania pseudo-Helium, sample images of the samples are collected in four directions: shell opening upward, shell opening downward, shell opening to the left, and shell opening to the right;

[0108] In this embodiment, a large number of snail samples can achieve better model training effects and more accurate snail identification.

[0109] As an optional implementation, in order to conduct in-depth research on Oncomelania hupensis and Oncomelania pseudo-oncomelania, it is necessary to collect sample images of them from different directions. This includes but is not limited to the four directions of shell opening upward, shell opening downward, shell opening to the left, and shell opening to the right, to ensure that their morphological characteristics are fully captured. This collection method is helpful for subsequent classification, identification, and ecological research, and provides an important basis for the prevention and control of Oncomelania hupensis and Oncomelania pseudo-oncomelania.

[0110] Step B12, classifying according to the shell opening direction in the sample image to obtain a basic image data set;

[0111] In this embodiment, since the morphological characteristics of snails are diverse, hundreds of different samples are selected for each snail, photographed from multiple angles, and then classified and unqualified images are screened out, which can improve the recognition accuracy of the recognition model.

[0112] As an optional implementation, by classifying the sample images of Oncomelania hupensis and Oncomelania pseudo-oncomelaniae, an ordered and structured classified image dataset can be obtained. Such a dataset provides a basis for subsequent image analysis, feature extraction, model training, etc., and is helpful for in-depth research on the classification and identification of Oncomelania hupensis and Oncomelania pseudo-oncomelaniae.

[0113] Step B13, classifying the snails according to the types of snails in the basic image dataset to obtain the classified image dataset.

[0114] In this embodiment, snails are classified by type to facilitate model training and recognition. For snails with more details, diversified training is performed to improve the recognition accuracy of the model.

[0115] As an optional implementation, it includes a basic image dataset, snail types and a classified image dataset, which together constitute the core elements of the snail image classification task.

[0116] Exemplarily, snails are randomly collected, a large number of types of snails are collected, and they are photographed from multiple directions. Sample images are collected, and the sample images are classified and processed to obtain a classified image data set.

[0117] Because the model is trained through a large number of randomly processed images, the recognition accuracy and efficiency of the fine-grained recognition model of the snail subspecies are improved.

[0118] Based on any of the above embodiments, in Embodiment 5 of the present application, refer to Fig. 9 , before step S30, steps D11 to D13 are also included:

[0119] Step D11, taking the sample in the classification image dataset as the center, cutting out the background area without information in the image, and obtaining a standard classification image dataset;

[0120] In this embodiment, the snails are placed at the center of the captured image to make the snail features more obvious, the informationless background is cropped to make the storage memory smaller, and the pixels are scaled to a suitable size to facilitate model learning and training.

[0121] As an optional implementation, in order to improve the accuracy and efficiency of image classification, it is usually necessary to preprocess the samples in the original classification image dataset. One important step is to crop the background area without information in the image. Through this processing, noise interference can be reduced and the target features can be highlighted, thereby obtaining a more standardized and high-quality classification image dataset. Such a dataset helps to improve the performance and accuracy of subsequent image classification, feature extraction and other tasks.

[0122] Step D12, performing pixel scaling on the standard classification image data set to obtain an initially processed classification image data set;

[0123] In this embodiment, the images in the data set are pixel-scaled to achieve standard appropriate pixels, and an initially processed classification image data set is obtained to facilitate the training operation of the model.

[0124] As an optional implementation, in order to obtain an image dataset suitable for subsequent processing, the images in the standard classification image dataset need to be pixel scaled. This process involves adjusting the size of the image to ensure that all images have a uniform size to facilitate model training and evaluation. Through pixel scaling, the image can be made to meet specific input requirements while maintaining the clarity and details of the image as much as possible. The scaled images will constitute the initial processed classification image dataset, providing high-quality input data for subsequent image classification, feature extraction and other tasks. This step is an important part of image preprocessing and is of great significance for improving the performance and accuracy of the model.

[0125] Step D13, performing a randomization operation on the initially processed classified image data set to obtain the target classified image data set.

[0126] In this embodiment, randomization is performed on the images in the data set to obtain a target classification image data set, which is used for training the model to improve its recognition accuracy.

[0127] As an optional implementation, in image classification tasks, in order to improve the generalization ability of the model and avoid overfitting, it is usually necessary to further randomize the initially processed classification image dataset. This step may include randomly disrupting the order of images in the dataset, randomly selecting images for training and validation sets, etc. Through randomization, it can be ensured that the model is exposed to diverse data during training, thereby improving its performance on new data. The target classification image dataset will provide a solid foundation for subsequent model training and evaluation.

[0128] For example, random processing of images includes image smoothing: used to reduce noise in the image and make the image clearer; image enhancement: by strengthening the high-frequency or low-frequency components in the image, highlighting the interesting parts in the image and reducing the impact of noise; image transformation: including geometric transformation, such as translation, rotation, mirroring, transposition and scale transformation, such as scaling, interpolation algorithms, and transformation between spatial domain and frequency domain, such as Fourier transform; image segmentation: extracting meaningful feature parts from the image, such as edges, regions, etc., is the basis of image recognition, analysis and understanding; histogram equalization: used to improve the contrast of the image, which is achieved by adjusting the histogram distribution of the image. In addition, image processing also includes point operations, filtering, global optimization and other methods to increase the complexity of model training and improve the recognition ability of the model.

[0129] Because the model is trained with a large number of random snail images, the recognition ability of the fine-grained recognition model of the snail subspecies is improved.

[0130] Based on any of the above embodiments, in Embodiment 6 of the present application, refer to Fig.10 , the step D13 includes steps E11 to E13:

[0131] E11, randomly rotating the images in the initially processed classification image data set to obtain a first classification image data set;

[0132] In this embodiment, the complexity of recognition is increased by rotating the image, so that the trained model can recognize images at various angles.

[0133] As an optional implementation, in image classification tasks, in order to increase the diversity of the training data set and improve the generalization ability of the model, the images in the initially processed classification image data set are often randomly rotated. By randomly selecting the rotation angle and rotating the image, the changes of objects in different directions in real scenes can be simulated, so that the model can learn the characteristics of objects at different angles. The processed images will constitute the first classification image data set, providing richer and more diverse data for subsequent model training and evaluation. This method helps the model perform better during testing and reduces the risk of overfitting.

[0134] E12, randomly adjusting the sharpness, brightness, and contrast of the images in the first classified image data set to obtain a second classified image data set;

[0135] In this embodiment, random adjustments are made to the image, including adjustments to sharpness, brightness, and contrast, thereby avoiding randomness in the recognition of the trained model.

[0136] As an optional implementation, the second classification image dataset can be obtained by randomly adjusting the sharpness, brightness and contrast of the images in the first classification image dataset. This processing method helps to increase the diversity of the dataset and improve the adaptability of the model to different image conditions.

[0137] E13, randomly performing grayscale processing on the images in the second classified image data set to obtain the target classified image data set.

[0138] In this embodiment, image grayscale processing is the process of converting a color image into a grayscale image. Simplify the image matrix and improve computing efficiency. Color images have a large amount of information, but image recognition often only requires grayscale information, so grayscale processing can significantly improve computing speed. Random adjustment of the image makes the model training more complex and can adapt to a variety of different usage scenarios.

[0139] As an optional implementation, since the snail samples have many features and are small in shape, in order to clearly identify the features of the snail samples, sufficient light is selected and a certain distance is maintained for horizontal shooting, which can ensure that the features of the snail samples are all photographed and recorded. Since the scenes recognized in the actual visual recognition process of snails are different, in order to prevent the occurrence of contingency in actual visual recognition, random rotation and various adjustments of the image can avoid erroneous recognition caused by environmental interference.

[0140] Exemplarily, the image is randomly processed, including but not limited to image rendering and image transformation, to increase the complexity of model training and improve the recognition ability of the model.

[0141] Because the model is trained through a large number of randomly processed images, the recognition accuracy and efficiency of the fine-grained recognition model of the snail subspecies are improved.

[0142] Based on any of the above embodiments, in Embodiment 7 of the present application, the step S40 includes:

[0143] Importing the target pre-trained weights into the pre-trained initial model to obtain a basic model;

[0144] In this embodiment, model retraining refers to the process of retraining the model with new data after the model is initially trained and deployed due to changes in data distribution or degradation of model performance, in order to adapt to data changes and improve model performance.

[0145] As an optional implementation, the classified image dataset is imported into the model for training to obtain the pre-trained weights of the model, and then the pre-trained weights are imported into the model. Finally, the classified image dataset is imported into the model for retraining, so that the performance of the model is more efficient and the recognition is more accurate.

[0146] The data of the basic model is updated, and the classified image data set is used for training to obtain the snail subspecies fine-grained recognition model.

[0147] In this embodiment, the data of the basic model is fine-tuned, and some or all layers of the pre-trained model are fine-tuned according to the requirements of the new task. Usually, the bottom layer is retained and fine-tuned, and the top layer is replaced with an output layer adapted to the new task. When training the model, the parameters of the newly added output layer are updated, and some layer parameters of the pre-trained model are fine-tuned to better adapt to the data distribution and characteristics of the new task. Improve the performance and generalization ability of the model.

[0148] As an optional implementation, in order to improve the recognition accuracy and generalization ability of the model, the data of the basic model is updated, which may include adding new training samples, optimizing data preprocessing methods, etc. During the model training process, carefully classified image datasets were used. These image datasets contain samples of different subspecies of Oncomelania snails, and each sample is accurately labeled to ensure that the model can learn effective features. Through training, a fine-grained recognition model that can accurately identify different subspecies of Oncomelania snails was finally obtained. The model can capture the subtle differences between Oncomelania snail subspecies, thereby achieving accurate classification and providing strong support for the monitoring and prevention of Oncomelania snails.

[0149] Exemplarily, after training the model and optimizing and supplementing the data set, the model is trained again, and this operation is repeated, so that the model recognition capability becomes faster and more accurate.

[0150] Thanks to multiple retrainings, the model has mastered the recognition capabilities, further improving its recognition speed and accuracy.

[0151] Based on any of the above embodiments, in Embodiment 8 of the present application, refer to Fig.11 , the step S30 includes C11-C12:

[0152] C11, by importing the target classification image dataset into the initial model pre-training, the basic pre-training weights are obtained;

[0153] In this embodiment, a suitable pre-trained model is selected, such as ResNet, VGG, etc., and the model structure is adjusted as needed. The model parameters are initialized using initialization methods such as Xavier and Kaiming.

[0154] As an optional implementation, load pre-trained weights: If the model structure has not changed, load the pre-trained weights directly; if the structure has changed, use the conversion tool to adjust the pre-trained weights to match the new structure.

[0155] C12, updating and adjusting the basic pre-training weights to obtain the target pre-training weights.

[0156] In this embodiment, the updating and adjustment of the basic pre-training weights includes that if the model structure changes, the pre-training weights may no longer be fully applicable. At this time, the model can be retrained on a large data set, or the pre-training weights can be partially used and the new structure can be initialized for training. To further improve the performance of the model, over-parameterization methods such as MPO decomposition and gradient descent can be used to update and adjust the pre-training model.

[0157] As an optional implementation, based on the existing pre-trained weights, a series of update and adjustment operations are performed to obtain new pre-trained weights that better meet specific target requirements.

[0158] Exemplarily, the process of obtaining the pre-training weights includes using the ImageNet dataset for model pre-training; data enhancement techniques such as random cropping and horizontal flipping are used during training to increase data diversity. The training batch size is 256, and the weights are continuously updated through the SGD optimization algorithm. The optimizer parameters are: initial learning rate lr = 0.1, momentum momentum = 0.9, and decay weight weight_decay = 0.0001. The MultiStepLR learning rate decay strategy is used to update the learning rate, and it is updated once every 30, 60, and 90 rounds. Each time the learning rate becomes 0.1 times the previous one, and the number of training rounds is 100 rounds. The training weights are saved as .pth files for initializing the model.

[0159] After the pre-trained weights are imported into the model, they are fine-tuned on the snail classification dataset. Image enhancement techniques are used during the experiment: Image resizing: resize the image to a uniform size of scale = 380 to ensure consistent input size. Adjust the sharpness of the image, with an amplitude range of -0.8 to 0.8. Add a Gaussian blur effect, with a blur range of 0.1 to 2.0. The image is randomly rotated, with a rotation angle range of 0° to 180°, and the unfilled area is filled with zeros with a probability of 80%, and there is a 50% probability of reverse rotation. Randomly adjust the two pixels, contrast, saturation and hue, randomly adjust the grayscale, and perform Mixup enhancement. The training batch size is 256, and the weights are continuously updated through the Adam optimization algorithm. The optimizer parameters are: initial learning rate lr = 0.001, weight decay weight_decay = 0.05, numerical stability parameter eps = 1e-8, momentum coefficient betas = (0.9, 0.999). The MultiStepLR learning rate decay strategy is used to update the learning rate. The learning rate is updated every 30, 60, and 90 rounds. Each time the learning rate becomes 0.1 times of the previous one. The number of training rounds is 150. The training weights are saved as a .pth file for initializing the model.

[0160] The present application provides a device for a fine-grained identification model of Oncomelania subspecies, the device for a fine-grained identification model of Oncomelania subspecies comprises: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for constructing a fine-grained identification model of Oncomelania subspecies in the above-mentioned embodiment 1.

[0161] Reference below Fig.12 , which shows a schematic diagram of the structure of a device suitable for implementing the Oncomelania subspecies fine-grained identification model of the embodiment of the present application. The device of the Oncomelania subspecies fine-grained identification model in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.12 The device of the snail subspecies fine-grained identification model shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0162] like Fig.12As shown, the device of the fine-grained identification model of Oncomelania subspecies may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. Various programs and data required for the device operation of the fine-grained identification model of Oncomelania subspecies are also stored in RAM1004. The processing device 1001, ROM1002, and RAM1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems may be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 may allow the device of the Oncomelania subspecies fine-grained identification model to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a device of the Oncomelania subspecies fine-grained identification model with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0163] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0164] The device for the fine-grained identification model of Oncomelania subspecies provided by the present application adopts the method for constructing the fine-grained identification model of Oncomelania subspecies in the above-mentioned embodiment, which can solve the technical problem that the model in the related art can only identify Oncomelania subspecies of a certain type. Compared with the prior art, the beneficial effects of the device for the fine-grained identification model of Oncomelania subspecies provided by the present application are the same as the beneficial effects of the method for constructing the fine-grained identification model of Oncomelania subspecies provided by the above-mentioned embodiment, and the other technical features of the device for the fine-grained identification model of Oncomelania subspecies are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0165] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0166] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0167] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the method for constructing a fine-grained identification model of Oncomelania subspecies in the above-mentioned embodiment.

[0168] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0169] The computer-readable storage medium may be included in the device of the Oncomelania snail subspecies fine-grained identification model; or may exist independently without being assembled into the device of the Oncomelania snail subspecies fine-grained identification model.

[0170] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by a device for a fine-grained recognition model of Oncomelania subspecies, the device for the fine-grained recognition model of Oncomelania subspecies: establishes a classification image dataset of Oncomelania subspecies and Oncomelania pseudo-Oncomelania through sample collection; pre-trains the initial model based on a standard dataset to obtain a target pre-training weight; imports the target pre-training weight into the pre-trained initial model, and uses the target classification image dataset for training to obtain a fine-grained recognition model of Oncomelania subspecies.

[0171] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0172] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0173] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0174] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the method for constructing the above-mentioned Oncomelania subspecies fine-grained identification model, and can solve the technical problem that the model in the related art can only identify Oncomelania subspecies of a certain type. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the method for constructing the Oncomelania subspecies fine-grained identification model provided in the above-mentioned embodiment, and will not be repeated here.

[0175] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent processing scope of the present application.

Claims

1. A method for constructing a fine-grained recognition model for Oncomelania snail subspecies, characterized in that: The method comprises: Through sample collection, a classification image dataset of Oncomelania snail subspecies and Oncomelania pseudo-oncomelania was established; Preprocessing the classified image data set to obtain a target classified image data set; Pre-train the initial model based on the standard data set to obtain the target pre-training weights; The target pre-training weights are imported into the pre-trained initial model, and the target classification image data set is used for training to obtain a fine-grained recognition model for Oncomelania subspecies.

2. The method for constructing a fine-grained identification model of Oncomelania subspecies according to claim 1, characterized in that: After the step of importing the target pre-trained weight into the pre-trained initial model and using the target classification image data set for training to obtain a fine-grained recognition model for Oncomelania subspecies, the method further includes: Using a preset training strategy to train and compare the snail subspecies fine-grained recognition model to obtain a target strategy; The target strategy is introduced into the Oncomelania snail subspecies fine-grained recognition model to obtain a target model.

3. The method for constructing a fine-grained identification model of Oncomelania subspecies according to claim 2, characterized in that: The training strategy includes data enhancement conditions and hyperparameters, the data enhancement conditions include basis, mixing and Gaussian blur, the hyperparameters include optimizer, loss function and learning rate decay strategy, and the step of training and comparing the Oncomelania subspecies fine-grained recognition model using the preset training strategy to obtain the target strategy includes: Keeping the hyperparameters in the Oncomelania snail subspecies fine-grained identification model unchanged, a standard model is obtained; Importing the classified image data sets under different data enhancement conditions into the standard model for training comparison, and screening out target data enhancement conditions; Keeping the data enhancement conditions of the classification image dataset unchanged, a standard classification image dataset is obtained; Importing the standard classification image data into the Oncomelania snail subspecies fine-grained recognition model under different hyperparameters for training and comparison, and screening out target hyperparameters; The target data enhancement condition and the target hyperparameter are used as the target strategy.

4. The method for constructing a fine-grained identification model of Oncomelania subspecies according to claim 1, characterized in that: The step of establishing a classification image data set of Oncomelania hupensis subspecies and Oncomelania pseudo-hupensis through sample collection comprises: For several samples of Oncomelania hupensis and Oncomelania pseudo-oncomelania, sample images of the samples are collected in four directions: shell opening upward, shell opening downward, shell opening to the left, and shell opening to the right; Classify according to the shell opening direction in the sample image to obtain a basic image data set; The classified image dataset is obtained by classifying the snails according to the types of snails in the basic image dataset.

5. The method for constructing a fine-grained identification model of Oncomelania subspecies according to claim 1, characterized in that: Before the step of pre-training the initial model based on the standard data set to obtain the target pre-training weight, the method further includes: Taking the target object in the classification image dataset as the center, cutting out the background area without information in the image, and obtaining a standard classification image dataset; Performing pixel scaling on the standard classification image data set to obtain an initially processed classification image data set; A randomization operation is performed on the initially processed classified image data set to obtain the target classified image data set.

6. The method for constructing a fine-grained identification model of Oncomelania subspecies according to claim 5, characterized in that: The step of performing a randomization operation on the initially processed classified image data set to obtain the target classified image data set comprises: Randomly rotating the images in the initially processed classified image data set to obtain a first classified image data set; Randomly adjusting the sharpness, brightness, and contrast of the images in the first classified image data set to obtain a second classified image data set; The images in the second classified image data set are randomly gray-scaled to obtain the target classified image data set.

7. The method for constructing a fine-grained identification model of Oncomelania subspecies according to claim 1, characterized in that: The step of importing the target pre-trained weights into the pre-trained initial model and using the target classification image data set for training to obtain a fine-grained recognition model for Oncomelania subspecies also includes: Importing the target pre-trained weights into the pre-trained initial model to obtain a basic model; The data of the basic model is updated, and the classified image data set is used for training to obtain the snail subspecies fine-grained recognition model.

8. The method for constructing a fine-grained identification model of Oncomelania subspecies according to claim 1, characterized in that: The step of pre-training the initial model based on the standard data set to obtain the target pre-training weights includes: By importing the target classification image dataset into the initial model pre-training, the basic pre-training weights are obtained; The basic pre-training weights are updated and adjusted to obtain the target pre-training weights.

9. A device for fine-grained identification model of Oncomelania snail subspecies, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the method for constructing a fine-grained identification model of Oncomelania subspecies according to any one of claims 1 to 8.

10. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the method for constructing a fine-grained identification model of Oncomelania subspecies according to any one of claims 1 to 8 are implemented.

Images