Repeated planting failure analysis method based on multi-modal data fusion model

Abstract

The invention discloses a repeated implantation failure analysis method based on a multi-modal data fusion model, and the method comprises the steps: collecting basic information of a patient, medical history, laboratory and iconography examination, embryo transplantation and other multi-modal data, integrating dispersed information, providing a comprehensive data basis for analysis, breaking an information island, and enabling the analysis to be based on a complete image; meanwhile, a neural network model is trained according to the preprocessed data to obtain a multi-modal data fusion model, potential correlation among factors can be mined, internal relations can be accurately reflected through fusion features, complex disease causes can be comprehensively revealed, and the limitation that only one aspect can be concerned in the prior art is overcome; besides, a prediction method based on the multi-modal data fusion model is more accurate and reliable than traditional experience judgment and single index analysis, targeted suggestions can be given according to prediction, personalized guidance is provided for clinicians, formulation of precise strategies is assisted, and the defects of precise prediction and personalized suggestions in the prior art are overcome.

Description

Technical Field

[0001] This invention relates to the field of planting analysis technology, and in particular to a method for analyzing repeated planting failures based on a multimodal data fusion model. Background Technology

[0002] Recurrent implantation failure usually refers to the inability to achieve a successful pregnancy after multiple high-quality embryo transfers. Its pathogenesis is complex and diverse, involving multiple levels such as embryonic factors, maternal factors, and the interaction between the two.

[0003] In current clinical practice, the analysis of recurrent implantation failure mainly relies on single examination methods and clinical experience. Doctors usually make a comprehensive judgment based on the patient's basic information, medical history, laboratory test results, and imaging examination results. However, this information is often scattered and lacks effective integration and analysis methods. At the same time, different examination methods may only reflect one aspect of the problem and are difficult to fully reveal the complex causes of recurrent implantation failure. Summary of the Invention

[0004] In view of this, the present invention proposes a method for analyzing recurrent planting failure based on a multimodal data fusion model, which can effectively solve the shortcomings of existing technologies, such as the lack of effective integration and analysis methods and the difficulty in fully revealing the complex causes of recurrent planting failure.

[0005] The technical solution of this invention is implemented as follows:

[0006] A method for analyzing repeated planting failures based on a multimodal data fusion model includes:

[0007] Collect multimodal data from patients, including basic patient information, medical history, laboratory test results, imaging results, and embryo transfer-related data.

[0008] The collected multimodal data is preprocessed to obtain preprocessed multimodal data;

[0009] The neural network model is trained based on the preprocessed multimodal data to obtain a multimodal data fusion model. The fusion features output by the multimodal data fusion model can accurately reflect the intrinsic relationship between the planting process and repeated planting failures.

[0010] Input the multimodal data of the current planting process into the multimodal data fusion model. After the multimodal data fusion model outputs the fused features, it uses a preset classifier to predict and analyze whether the current planting will result in repeated planting failures, and provides corresponding analysis results and suggested measures.

[0011] As a further optional solution to the repeated planting failure analysis method based on the multimodal data fusion model, the preprocessing of the collected multimodal data to obtain preprocessed multimodal data specifically includes:

[0012] Calculate the dynamic threshold of data characteristics based on the historical data distribution for the same period;

[0013] Data cleaning is performed on multimodal data based on dynamic thresholds of data characteristics to obtain cleaned multimodal data;

[0014] Calculate the information entropy corresponding to the cleaned multimodal data;

[0015] The normalization formula parameters are dynamically adjusted based on information entropy to obtain the normalization formula after parameter adjustment.

[0016] The cleaned multimodal data is normalized according to the normalization formula after parameter adjustment to obtain preprocessed multimodal data.

[0017] As a further optional scheme of the repeated planting failure analysis method based on the multimodal data fusion model, the step of training the neural network model based on the preprocessed multimodal data to obtain the multimodal data fusion model specifically includes:

[0018] A neural network model is constructed, wherein the neural network model has a multi-branch network, each branch network corresponding to a data input mode; each branch network uses a hybrid structure combining convolutional neural network and long short-term memory network for feature extraction; a feature fusion layer is set after each branch network, and an attention mechanism is used to perform weighted fusion of the features extracted by each branch network; after the feature fusion layer, a fully connected layer and an output layer are connected, the fully connected layer is used to further integrate and transform the fused features, and the output layer uses an activation function to output the final fused features;

[0019] The constructed neural network model is trained using preprocessed multimodal data to obtain a multimodal data fusion model.

[0020] As a further optional solution to the recurring planting failure analysis method based on the multimodal data fusion model, the step of predicting and analyzing whether recurring planting failures will occur in the current planting by using a preset classifier specifically includes:

[0021] Build a classifier based on the Stacking ensemble learning framework;

[0022] Based on the classifier and fusion features, the system predicts whether repeated planting failures will occur in the current planting process, and obtains the corresponding analysis results.

[0023] As a further optional solution to the repeated planting failure analysis method based on the multimodal data fusion model, the construction of a classifier based on the Stacking ensemble learning framework specifically includes:

[0024] Acquire historical fusion feature analysis data and divide the historical fusion feature analysis data into training set and validation set;

[0025] The base learners are trained independently based on the training set to obtain trained base learner models.

[0026] The trained base learner model is used to make predictions on the validation set, and the base learner model outputs the prediction results.

[0027] The prediction results of the base learner model on the validation set are used as new features, and together with the true class labels of the validation set, they form a meta-feature dataset.

[0028] The meta-learner is trained using the meta-feature dataset to obtain the meta-learner with optimized parameters.

[0029] The trained base learner model and the optimized meta learner are integrated to obtain a classifier based on the Stacking ensemble learning framework.

[0030] A system for analyzing repeated planting failures based on a multimodal data fusion model, comprising:

[0031] Collect multimodal data from patients, including basic patient information, medical history, laboratory test results, imaging results, and embryo transfer-related data.

[0032] The preprocessing module is used to preprocess the acquired multimodal data to obtain preprocessed multimodal data;

[0033] The training module is used to train the neural network model based on the preprocessed multimodal data to obtain a multimodal data fusion model. The fusion features output by the multimodal data fusion model can accurately reflect the intrinsic relationship between the planting process and repeated planting failures.

[0034] The predictive analysis module is used to input multimodal data of the current planting process into the multimodal data fusion model. After the multimodal data fusion model outputs fused features, it uses a preset classifier to predict whether the current planting will result in repeated planting failures, and provides corresponding analysis results and suggested measures.

[0035] A computing device includes 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 any of the steps of the aforementioned method for analyzing repeated planting failures based on a multimodal data fusion model.

[0036] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for analyzing repeated planting failures based on a multimodal data fusion model.

[0037] The beneficial effects of this invention are as follows: By collecting multimodal data from patients, covering basic information, medical history, laboratory test results, imaging results, and embryo transfer-related data, this comprehensive data collection integrates previously scattered and isolated information, providing a rich and comprehensive data foundation for subsequent analysis. For example, basic patient information such as age and body mass index may be related to endocrine status; adverse pregnancy and surgical history in medical history can reflect potential problems in the reproductive system; laboratory tests and imaging examinations provide information on bodily functions and organ structure from different perspectives; and embryo transfer-related data is directly related to the implantation process. This integration of multi-source data breaks down information silos, enabling analysis based on a more complete picture. Simultaneously, by training a neural network model based on the preprocessed multimodal data, a multimodal data fusion model is obtained, which can mine... This approach reveals potential connections between various factors, such as the subtle relationship between endocrine indicators and endometrial receptivity, and the interaction between immune factors and embryo quality. The fusion features output by the model can more accurately reflect these intrinsic relationships, thus comprehensively revealing the complex causes of recurrent implantation failure. Existing technologies often only focus on one aspect of the problem and cannot delve into the synergistic effects between multiple factors. Furthermore, prediction methods based on multimodal data fusion models have higher accuracy and reliability compared to traditional experience-based judgments and single-indicator analysis. Simultaneously, providing corresponding analysis results and recommendations based on the prediction results can offer clinicians more targeted and personalized guidance. This helps doctors develop more precise strategies and improve the pregnancy success rate for patients with recurrent implantation failure. Existing technologies, lacking effective integration and analysis methods, struggle to provide such accurate predictions and personalized recommendations. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart of a method for analyzing repeated planting failures based on a multimodal data fusion model according to the present invention;

[0040] Figure 2 This is a schematic diagram of the composition of a repetitive planting failure analysis system based on a multimodal data fusion model according to the present invention;

[0041] Figure 3 This is a schematic diagram of the composition of a computing device according to the present invention. Detailed Implementation

[0042] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] refer to Figures 1 to 3 A method for analyzing repeated planting failures based on a multimodal data fusion model, comprising:

[0044] Collect multimodal data from patients, including basic patient information, medical history, laboratory test results, imaging results, and embryo transfer-related data.

[0045] Specifically, the collection of multimodal data, including basic patient information, medical history, laboratory test results, imaging results, and embryo transfer-related data, comprehensively reflects the patient's condition from multiple dimensions. Basic patient information, such as age and body mass index, may be related to the physiological functions and endocrine status of the reproductive system. Medical history, including menstrual history, reproductive history, and adverse pregnancy history, provides important clues to understanding the patient's reproductive health background. Laboratory test results accurately present biochemical indicators such as hormone levels, immune status, and coagulation function. Imaging results visually demonstrate the anatomical structure of the reproductive system. Embryo transfer-related data is directly linked to the specific details of the implantation process. This comprehensive data collection ensures that no important information is missed when analyzing the causes of recurrent implantation failure.

[0046] Different types of data have different characteristics and advantages, and they can complement each other. For example, while laboratory test results can provide quantitative biochemical indicators, they may not be able to directly reflect the morphological structure of the reproductive system, while imaging test results can make up for this deficiency. Combining subjective descriptions in medical history information with objective data from laboratory tests can more accurately assess the patient's condition. By collecting multimodal data and making full use of the advantages of each data source, the analysis can be based on richer and more accurate information, thereby improving the accuracy of the analysis of the causes of recurrent implantation failure.

[0047] It should be noted that the multimodal data of the patients were collected from a database.

[0048] The collected multimodal data is preprocessed to obtain preprocessed multimodal data; in some embodiments, the preprocessing of the collected multimodal data to obtain preprocessed multimodal data specifically includes:

[0049] Based on the distribution of historical data within the same period, dynamic thresholds for data features are calculated. Specifically, the range of historical data within the same period is determined. For example, for planting data, data from the same growth stage, season, or climatic conditions in the past are selected as historical data within the same period. For each data feature, a dynamic threshold is calculated using a method based on the mean and standard deviation. Let the set of historical data within the same period be... Then, the dynamic lower limit threshold of this data feature at the current moment is... and dynamic upper limit threshold The calculation formula is as follows:

[0050] ;

[0051] ;

[0052] in, It is the average of historical data for the same period. It is the standard deviation of historical data for the same period. and It is a dynamic coefficient adjusted based on the stability of data characteristics and business needs, and >0, >0.

[0053] Based on the calculated dynamic threshold, outliers in multimodal data are identified. When the current data value exceeds […], […]. , When the value falls within the specified range, it is considered an outlier. For identified outliers, appropriate methods are used for processing. For example, for numerical data, the mean or median of adjacent normal data can be used for replacement. For outliers with many missing values, direct deletion can be considered. Simultaneously, the impact of data cleaning on data integrity is assessed. If deleting data points results in insufficient data volume, data interpolation or other methods can be used to supplement it. Finally, the cleaned multimodal data is obtained.

[0054] If multimodal data contains multiple features, calculate the information entropy of each feature separately. Let the data set for a certain feature be denoted as . Discretize the data into N intervals, where the frequency of data in the i-th interval is... Then the information entropy of this feature The calculation formula is:

[0055] ;

[0056] in, is the data frequency of the i-th interval. By calculating the information entropy of each feature, we obtain the information entropy corresponding to the cleaned multimodal data.

[0057] By dynamically adjusting the parameters of the normalization formula based on information entropy, a normalized formula with adjusted parameters is obtained. Based on the calculated information entropy, a functional relationship is established between the information entropy and the parameters of the normalization formula. For example, let the normalization formula be... ,in and These are the minimum and maximum values ​​in the data set D, respectively, when the information entropy When the information entropy is large, it indicates that the data distribution is relatively scattered. Appropriately increasing α can broaden the range of the normalized data. If the data is relatively concentrated in a certain range, we can appropriately reduce α to avoid the data being too concentrated in a certain range after normalization. At the same time, we can dynamically adjust the value of β according to the overall offset of the data to obtain the normalization formula after adjusting the parameters.

[0058] Based on the normalization formula after parameter adjustment, the cleaned multimodal data is normalized to map the data to a suitable numerical range, thus obtaining preprocessed multimodal data.

[0059] A neural network model is trained based on preprocessed multimodal data to obtain a multimodal data fusion model. The fusion features output by the multimodal data fusion model accurately reflect the intrinsic relationship between the planting process and repeated planting failures. In some embodiments, training the neural network model based on preprocessed multimodal data to obtain the multimodal data fusion model specifically includes:

[0060] A neural network model is constructed, wherein the neural network model has a multi-branch network, each branch network corresponding to a data input of a certain modality; each branch network uses a hybrid structure combining convolutional neural network and long short-term memory network for feature extraction; a feature fusion layer is set after each branch network, and an attention mechanism is used to weight and fuse the features extracted by each branch network, so that the features of key modality data are more fully reflected in the fusion process; after the feature fusion layer, a fully connected layer and an output layer are connected. The fully connected layer is used to further integrate and transform the fused features, and the output layer uses an appropriate activation function (such as the sigmoid function, which is suitable for binary classification problems to determine whether repeated planting failures have occurred) to output the final fused feature, which can accurately reflect the intrinsic relationship between the planting process and repeated planting failures;

[0061] The constructed neural network model is trained using preprocessed multimodal data to obtain a multimodal data fusion model.

[0062] Specifically, a neural network model with multiple branches is constructed, with each branch corresponding to a specific modality of data input. This structural design is well-suited to the characteristics of multimodal data, as different modalities have different features and distributions. The multiple branches can process each modality of data separately, avoiding interference and feature confusion that may occur when different modalities are mixed with input, thus more effectively extracting the intrinsic features of each modality. Each branch uses a hybrid structure combining convolutional neural networks (CNN) and long short-term memory networks (LSTM) for feature extraction. For data with spatial or temporal features, CNN can effectively extract local spatial features, while LSTM can capture long-term dependencies in the time series. This hybrid structure can fully leverage the advantages of both networks, more comprehensively and deeply mining useful information in multimodal data.

[0063] A feature fusion layer is set after each branch network, and an attention mechanism is used to perform weighted fusion of the features extracted by each branch network. The attention mechanism can automatically assign different weights according to the importance of different modal features to the final task (analyzing repeated planting failures). For example, in a specific planting scenario, the physiological data of the planted crop may have a greater impact on planting failures, and the attention mechanism will assign it a higher weight, so that the features of this modality will receive more attention during the fusion process. This approach can highlight key features, enhance the expressive power of the fused features, and more accurately reflect the intrinsic relationship between the planting process and repeated planting failures.

[0064] Following the feature fusion layer, a fully connected layer is connected. This fully connected layer further integrates and transforms the fused features, learning complex combinations between different features and mapping the fused features to a more suitable classification space, providing more effective information for the final classification decision. The output layer uses an activation function to output the final fused features. The activation function can introduce non-linear factors, enabling the model to fit more complex functional relationships, thus outputting more accurate and discriminative fused features. These fused features can be better used in subsequent predictions by a classifier to analyze whether repeated planting failures will occur, improving the accuracy and reliability of the predictions.

[0065] The multimodal data of the current planting process is input into a multimodal data fusion model. After the multimodal data fusion model outputs fused features, it uses a preset classifier to predict and analyze whether repeated planting failures will occur, and provides corresponding analysis results and suggested measures. In some embodiments, the step of using a preset classifier to predict and analyze whether repeated planting failures will occur specifically includes:

[0066] Build a classifier based on the Stacking ensemble learning framework;

[0067] Based on the classifier and fusion features, the system predicts whether repeated planting failures will occur in the current planting process, and obtains the corresponding analysis results.

[0068] Specifically, a classifier based on the Stacking ensemble learning framework is constructed. The Stacking method, by combining the prediction results of multiple base learners, can fully utilize the advantages of different learning algorithms and make up for the shortcomings of a single algorithm. Different base learners can capture diverse information patterns from multimodal fusion features. The ensembled classifier can integrate this information, thereby improving the prediction accuracy of whether repeated planting failures will occur in the current planting. Based on the classifier combined with fusion features, predictive analysis is performed. The fusion features have integrated key information from multimodal data. The classifier can effectively analyze and mine these rich features, further explore the potential relationship between features and planting failure, and more accurately judge the possibility of planting failure.

[0069] The Stacking ensemble learning framework offers a degree of flexibility, allowing users to select and combine different base learners and meta-learners based on specific circumstances, and adjust the classifier's composition to better adapt to various complex situations.

[0070] In some embodiments, constructing a classifier based on the Stacking ensemble learning framework specifically includes:

[0071] Acquire historical fusion feature analysis data and divide the historical fusion feature analysis data into training set and validation set;

[0072] The base learners are trained independently based on the training set to obtain trained base learner models.

[0073] The trained base learner model is used to make predictions on the validation set, and the base learner model outputs the prediction results.

[0074] The prediction results of the base learner model on the validation set are used as new features, and together with the true class labels of the validation set, they form a meta-feature dataset.

[0075] The meta-learner is trained using the meta-feature dataset to obtain the meta-learner with optimized parameters.

[0076] The trained base learner model and the optimized meta learner are integrated to obtain a classifier based on the Stacking ensemble learning framework.

[0077] Specifically, Stacking ensemble learning leverages the unique strengths of different models by combining multiple base learners. Independently trained base learners can extract diverse information patterns from historical fusion feature data. The meta-learner then performs secondary learning and integration of the prediction results from these base learners, effectively reducing the bias and variance of a single model and significantly improving prediction accuracy for situations such as repeated planting failures. Different base learners may have different sensitivities to noise and outliers in the data. By integrating the results of multiple models, the Stacking framework can reduce the error impact of individual models caused by data fluctuations, enabling the final classifier model to maintain relatively stable performance when facing different data distributions or noise levels.

[0078] Using the predictions of the base learner model on the validation set as new features, and combining them with the true class labels of the validation set to form a meta-feature dataset, this step further mines the potential information in the data. The new meta-features can reflect the inherent structure and patterns of the data from different perspectives, providing the meta-learner with richer and more discriminative inputs, which helps to improve the model's understanding and modeling ability of complex relationships. Historical fusion feature analysis data is obtained and reasonably divided into training and validation sets, making full use of the accumulated data resources. Through learning and modeling of historical data, the model can capture the long-term correlation and patterns between various factors and repeated planting failures in the planting process, enabling the model to have better generalization ability and adapt to planting conditions in different time periods and environments.

[0079] A system for analyzing repeated planting failures based on a multimodal data fusion model, comprising:

[0080] The acquisition module is used to collect multimodal data of patients, including basic patient information, medical history, laboratory test results, imaging test results, and embryo transfer-related data.

[0081] The preprocessing module is used to preprocess the acquired multimodal data to obtain preprocessed multimodal data;

[0082] The training module is used to train the neural network model based on the preprocessed multimodal data to obtain a multimodal data fusion model. The fusion features output by the multimodal data fusion model can accurately reflect the intrinsic relationship between the planting process and repeated planting failures.

[0083] The predictive analysis module is used to input multimodal data of the current planting process into the multimodal data fusion model. After the multimodal data fusion model outputs fused features, it uses a preset classifier to predict whether the current planting will result in repeated planting failures, and provides corresponding analysis results and suggested measures.

[0084] A computing device includes 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 any of the steps of the aforementioned method for analyzing repeated planting failures based on a multimodal data fusion model.

[0085] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for analyzing repeated planting failures based on a multimodal data fusion model.

[0086] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for analyzing repeated planting failures based on a multimodal data fusion model, characterized in that, include: Collect multimodal data from patients, including basic patient information, medical history, laboratory test results, imaging results, and embryo transfer-related data. The collected multimodal data is preprocessed to obtain preprocessed multimodal data; The neural network model is trained based on the preprocessed multimodal data to obtain a multimodal data fusion model. The fusion features output by the multimodal data fusion model can accurately reflect the intrinsic relationship between the planting process and repeated planting failures. Input the multimodal data of the current planting process into the multimodal data fusion model. After the multimodal data fusion model outputs the fused features, it uses a preset classifier to predict and analyze whether the current planting will result in repeated planting failures, and provides corresponding analysis results and suggested measures.

2. The method for analyzing repeated planting failures based on a multimodal data fusion model according to claim 1, characterized in that, The preprocessing of the acquired multimodal data to obtain preprocessed multimodal data specifically includes: Calculate the dynamic threshold of data characteristics based on the historical data distribution for the same period; Data cleaning is performed on multimodal data based on dynamic thresholds of data characteristics to obtain cleaned multimodal data; Calculate the information entropy corresponding to the cleaned multimodal data; The normalization formula parameters are dynamically adjusted based on information entropy to obtain the normalization formula after parameter adjustment. The cleaned multimodal data is normalized according to the normalization formula after parameter adjustment to obtain preprocessed multimodal data.

3. The method for analyzing repeated planting failures based on a multimodal data fusion model according to claim 2, characterized in that, The process of training a neural network model based on preprocessed multimodal data to obtain a multimodal data fusion model specifically includes: A neural network model is constructed, wherein the neural network model has a multi-branch network, each branch network corresponding to a data input mode; each branch network uses a hybrid structure combining convolutional neural network and long short-term memory network for feature extraction; a feature fusion layer is set after each branch network, and an attention mechanism is used to perform weighted fusion of the features extracted by each branch network; after the feature fusion layer, a fully connected layer and an output layer are connected, the fully connected layer is used to further integrate and transform the fused features, and the output layer uses an activation function to output the final fused features; The constructed neural network model is trained using preprocessed multimodal data to obtain a multimodal data fusion model.

4. The method for analyzing repeated planting failures based on a multimodal data fusion model according to claim 3, characterized in that, The step of predicting and analyzing whether repeated planting failures will occur in the current planting process using a preset classifier specifically includes: Build a classifier based on the Stacking ensemble learning framework; Based on the classifier and fusion features, the system predicts whether repeated planting failures will occur in the current planting process, and obtains the corresponding analysis results.

5. The method for analyzing repeated planting failures based on a multimodal data fusion model according to claim 4, characterized in that, The construction of the classifier based on the Stacking ensemble learning framework specifically includes: Acquire historical fusion feature analysis data and divide the historical fusion feature analysis data into training set and validation set; The base learners are trained independently based on the training set to obtain trained base learner models. The trained base learner model is used to make predictions on the validation set, and the base learner model outputs the prediction results. The prediction results of the base learner model on the validation set are used as new features, and together with the true class labels of the validation set, they form a meta-feature dataset. The meta-learner is trained using the meta-feature dataset to obtain the meta-learner with optimized parameters. The trained base learner model and the optimized meta learner are integrated to obtain a classifier based on the Stacking ensemble learning framework.

6. A system for analyzing repeated planting failures based on a multimodal data fusion model, characterized in that, include: The acquisition module is used to collect multimodal data of patients, including basic patient information, medical history, laboratory test results, imaging test results, and embryo transfer-related data. The preprocessing module is used to preprocess the acquired multimodal data to obtain preprocessed multimodal data; The training module is used to train the neural network model based on the preprocessed multimodal data to obtain a multimodal data fusion model. The fusion features output by the multimodal data fusion model can accurately reflect the intrinsic relationship between the planting process and repeated planting failures. The predictive analysis module is used to input multimodal data of the current planting process into the multimodal data fusion model. After the multimodal data fusion model outputs fused features, it uses a preset classifier to predict whether the current planting will result in repeated planting failures, and provides corresponding analysis results and suggested measures.

7. A computing device, characterized in that, The method includes 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 repeated planting failure analysis method based on a multimodal data fusion model as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the repeated planting failure analysis method based on a multimodal data fusion model as described in any one of claims 1-5.

Images