Precision advertisement putting method and system for agricultural terminal

Abstract

The present application relates to the technical field of data processing, and more particularly to a precision advertisement putting method and system for agricultural terminal, the system comprising: a data processing module, collecting agricultural data of the agricultural terminal for preprocessing to output standard agricultural data, and then performing feature extraction and fusion to output fused agricultural features; an analysis module, training an initial model according to the fused agricultural features to obtain a deep learning model and an analysis result; an advertisement putting module, generating an advertisement putting strategy according to the analysis result and putting the advertisement; a generation judgment module, judging the effective quality of the advertisement putting strategy generated according to the generation accuracy of the advertisement putting strategy; a filtering adjustment module, determining the weight sparsification filtering selection domain of the deep learning model according to the redundancy proportion of the fused agricultural features; and an inhibition adjustment module, determining the cross-correlation redundancy dynamic inhibition benchmark of the fused agricultural features according to the noise reduction efficiency of the standard agricultural data. The present application improves the accurate generation capability of the advertisement content.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for precise advertising delivery to agricultural input terminals. Background Technology

[0002] In existing technologies, advertising for agricultural inputs often relies on fixed channels, regional positioning, and manually set strategies, resulting in weak targeting and low matching accuracy. Some conventional advertising systems rely solely on delivery area and time for simple scheduling, failing to collect, denoise, and fuse features from multi-source, heterogeneous agricultural input data, making it difficult to uncover users' actual planting needs and potential consumption preferences. Furthermore, existing solutions often employ a pre-made ad content remote replacement model, lacking the ability to intelligently generate ad delivery strategies based on user needs, and lacking quantitative judgment mechanisms for generation quality and timeliness. In addition, traditional delivery methods lack deep learning model support and adaptive parameter adjustment mechanisms, failing to dynamically optimize for issues such as feature redundancy, data noise, and insufficient model generalization ability. This easily leads to large discrepancies between ad content and actual planting needs, as well as response delays, resulting in insufficient accuracy in generating ad delivery strategies.

[0003] Chinese Patent Publication No. CN107909387A discloses an advertising placement system and method. The system includes: an in-vehicle electronic device, an advertising terminal, and a server. The advertising terminal includes: an advertising generation module for generating an advertising placement request and sending the request to the server; the request carries advertising information, an advertising area, and an advertising time. A payment operation module for receiving a payment request from the server and receiving a payment instruction from a user for the request; the payment request includes advertising fees and a receiving account. A payment module for paying the advertising fees to the receiving account according to the payment instruction. The server includes: a fee generation module, used to receive the advertising placement request, generate the fee payment request based on the advertising placement area and advertising placement time, and send the fee payment request to the advertising terminal; a fee query module, used to query whether the receiving account has received the advertising fee; a positioning module, used to receive the positioning information reported by the vehicle-mounted electronic device, and after the receiving account receives the advertising fee, determine whether the vehicle-mounted electronic device is located within the advertising placement area based on the positioning information during the advertising placement time; and an advertising placement module, used to send the advertising information to the vehicle-mounted electronic devices located within the advertising placement area, and control the vehicle-mounted electronic devices to play the advertising information. Therefore, the aforementioned advertising placement system and method suffer from problems such as insufficient accuracy in generating advertising placement strategies because they only use advertising information, placement area, and placement time as criteria, lack multi-source data fusion and high-dimensional feature processing, can only remotely change advertising content in real time instead of intelligently generating it, and lack an adaptive optimization mechanism. Summary of the Invention

[0004] To address these issues, the present invention provides a method and system for precise advertising delivery to agricultural input terminals, which overcomes the problems of insufficient accuracy in generating advertising delivery strategies in existing technologies. These technologies rely solely on advertising information, delivery area, and delivery time as criteria, lack multi-source data fusion and high-dimensional feature processing, can only remotely change advertising content in real time instead of intelligently generating it, and lack an adaptive optimization mechanism.

[0005] To achieve the above objectives, the present invention provides a precision advertising delivery system for agricultural input terminals, comprising: The data processing module includes a collection unit for collecting agricultural input data from agricultural input terminals, a preprocessing unit connected to the collection unit for preprocessing the agricultural input data to output standard agricultural input data, and a fusion unit connected to the preprocessing unit for sequentially extracting and fusing features from the standard agricultural input data to output fused agricultural input features. The analysis module, which is connected to the data processing module, includes a training unit for training an initial model based on the integrated agricultural input characteristics to obtain a deep learning model, and an analysis unit connected to the training unit for analyzing agricultural input data based on the deep learning model to obtain analysis results. An advertising delivery module, which is connected to the analysis module, includes a strategy generation unit for generating advertising delivery strategies based on the analysis results, and a delivery unit connected to the strategy generation unit for delivering corresponding advertising content to different users based on the advertising delivery strategies. A generation and judgment module, which is connected to the advertising delivery module, is used to determine whether the effective quality of the generated advertising delivery strategy meets the requirements based on the generation accuracy of the advertising delivery strategy. A filtering adjustment module, which is connected to the generation and determination module, is used to determine the weight sparsification filtering domain of the deep learning model based on the redundancy ratio of the integrated agricultural input features when the effective quality generated by the advertising delivery strategy does not meet the requirements. The suppression adjustment module, which is connected to the filtering adjustment module, is used to determine the cross-correlation redundancy dynamic suppression benchmark of the fused agricultural input characteristics based on the noise reduction efficiency of standard agricultural input data.

[0006] Furthermore, the generation and determination module responds to the generation accuracy of the advertising delivery strategy to determine whether the effective quality of the generated advertising delivery strategy meets the requirements.

[0007] Furthermore, the generation and determination module determines that the effective quality of the generated advertising strategy meets the requirements if the generation accuracy rate of the advertising strategy is greater than or equal to the preset generation accuracy rate. The generation and determination module determines that the effective quality of the generated advertising strategy does not meet the requirements if the generation accuracy rate of the advertising strategy is less than the preset generation accuracy rate.

[0008] Furthermore, in response to the condition that the effective quality generated by the advertising delivery strategy does not meet the requirements, the filtering adjustment module determines whether the generalization ability of the deep learning model meets the requirements based on the redundancy ratio of the integrated agricultural input features.

[0009] Furthermore, the filtering adjustment module determines that the generalization ability of the deep learning model meets the requirements when the redundancy ratio of the integrated agricultural input features is less than or equal to a preset first ratio. The filtering adjustment module determines that the generalization ability of the deep learning model does not meet the requirements when the redundancy ratio of the integrated agricultural input features is greater than the preset first ratio.

[0010] Furthermore, in response to the fact that the redundancy ratio of the integrated agricultural input features is greater than the preset first ratio and less than the preset second ratio, the filtering adjustment module increases the weight sparsity filtering domain of the deep learning model. The filtering adjustment module responds to the fact that the redundancy ratio of the integrated agricultural input features is greater than or equal to the preset second ratio, and initially determines that the reasonable dimension capability of the integrated agricultural input features does not meet the requirements. It then determines whether the reasonable dimension capability of the integrated agricultural input features meets the requirements based on the noise reduction efficiency of the standard agricultural input data.

[0011] Furthermore, the increase in the weight sparsity filtering domain of the deep learning model is determined by the difference between the redundancy ratio of the fused agricultural input features and the preset first ratio.

[0012] Furthermore, the noise reduction efficiency of the suppression and adjustment module in response to standard agricultural input data is greater than or equal to the preset efficiency, and the ability to determine the reasonable dimension of the fused agricultural input features meets the requirements. The suppression adjustment module responds to the fact that the noise reduction efficiency of the standard agricultural input data is less than the preset efficiency, determines that the ability to integrate reasonable dimensions of agricultural input features does not meet the requirements, and reduces the cross-correlation redundancy dynamic suppression benchmark of the integrated agricultural input features.

[0013] Furthermore, the reduction range of the cross-correlation redundancy dynamic suppression benchmark that integrates agricultural input characteristics is determined by the difference between the preset effectiveness and the noise reduction effectiveness of standard agricultural input data.

[0014] This invention also provides a method for precise advertising delivery to agricultural input terminals, comprising: The collected agricultural input data from agricultural input terminals is preprocessed to output standard agricultural input data, and the standard agricultural input data is then sequentially subjected to feature extraction and fusion to output fused agricultural input features. The initial model is trained based on the integrated agricultural input features to obtain a deep learning model. The deep learning model is used to analyze the agricultural input data to obtain analysis results. Based on the analysis results, an advertising delivery strategy is generated to deliver corresponding advertising content to different users. Obtain the generation accuracy rate of the advertising placement strategy, and determine whether the effective quality of the generated advertising placement strategy meets the requirements based on the generation accuracy rate; If the quality of the effective products generated based on the advertising strategy does not meet the requirements, then the generalization ability of the deep learning model is determined based on the redundancy ratio of the integrated agricultural input features. If the generalization ability of the deep learning model does not meet the requirements, determine whether to increase the weight sparsity filtering domain of the deep learning model. If it is not necessary to increase the weight sparsity filtering domain of the deep learning model, then the cross-correlation redundancy dynamic suppression benchmark for integrating agricultural input features is determined based on the denoising efficiency of standard agricultural input data.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: The system of this invention, by setting up a data processing module, an analysis module, an advertising delivery module, a generation and judgment module, a filtering and adjustment module, and a suppression and adjustment module, judges the effective quality of the generated advertising delivery strategy based on the accuracy of the generated advertising delivery strategy. Due to insufficient model generalization ability, it cannot fully learn the actual agricultural input needs of users and the inherent laws of planting in different planting scenarios, resulting in deviations between the generated advertising delivery strategy and the actual planting problems of end users, thereby reducing the adaptability and delivery accuracy of the advertisement. By judging the effective quality of the generated advertising delivery strategy, the generalization defects of the model can be located in reverse, and the advertising delivery strategy generation strategy and model output weights can be dynamically corrected to improve the matching accuracy between the advertising delivery strategy and the user's planting needs. The weight sparsity filtering domain of the deep learning model is adjusted according to the redundancy ratio of the fused agricultural input features. Because the agricultural input features generated after feature fusion have too high dimensionality, the model cannot effectively filter and compress redundant features, introducing a large number of ineffective cross combinations during training, leading to... This exacerbates overfitting and insufficient generalization ability in the model. By increasing the sparsity filtering domain of the deep learning model's weights, the model's ability to eliminate invalid weights can be strengthened, reducing the proportion of parameters corresponding to redundant agricultural input features, decreasing the interference of invalid feature cross-combinations on model training, alleviating overfitting, and improving the model's generalization ability when collecting data from different agricultural input terminals. The cross-correlation redundancy dynamic suppression benchmark of the fused agricultural input features is adjusted according to the denoising effectiveness of standard agricultural input data. Due to the large number of original features and the high proportion of noise and redundant features in the collected agricultural input terminal data, effective feature selection and denoising cannot be achieved in the preprocessing stage, resulting in high dimensionality of the fused agricultural input features. By reducing the cross-correlation redundancy dynamic suppression benchmark of the fused agricultural input features, the strictness of redundancy judgment can be reduced, allowing more highly correlated redundant feature pairs to be effectively identified. Redundant features are eliminated or merged during feature fusion, thereby reducing the repeated superposition of highly similar features in the fused feature set, reducing the redundancy proportion of the fused agricultural input features, and improving the accuracy of advertising content generation.

[0016] Furthermore, the system described in this invention determines the effective quality of the generated advertising strategy by setting a preset generation accuracy rate. Due to insufficient model generalization ability, it cannot fully learn the actual agricultural input needs of users and the inherent laws of planting in different planting scenarios, resulting in deviations between the generated advertising strategy and the actual planting problems of end users, thereby reducing the adaptability and accuracy of advertising. By determining the effective quality of the generated advertising strategy, the generalization defects of the model can be located in reverse, and the weights of the advertising strategy generation strategy and model output can be dynamically corrected to improve the matching accuracy between the advertising strategy and the user's planting needs, thereby further improving the accuracy of advertising content generation.

[0017] Furthermore, the system of the present invention adjusts the weight sparsity filtering domain of the deep learning model by setting a preset first proportion and a preset second proportion. Because the agricultural input features generated after feature fusion have too high dimensionality, the model cannot effectively filter and compress redundant features, and a large number of invalid cross combinations are introduced during training, which aggravates the overfitting phenomenon and reduces the generalization ability. By increasing the weight sparsity filtering domain of the deep learning model, the model can strengthen the elimination of invalid weights, reduce the proportion of parameters corresponding to redundant agricultural input features, reduce the interference of invalid feature cross combinations on model training, alleviate overfitting, improve the generalization ability of the model when collecting data from different agricultural input terminals, and further improve the accuracy of advertising content generation.

[0018] Furthermore, the system of the present invention adjusts the dynamic suppression benchmark for cross-correlation redundancy of fused agricultural input features by setting a preset effectiveness rate. Since the collected agricultural input terminal data has many original features and a high proportion of noise and redundant features, it is impossible to achieve effective feature screening and noise reduction in the preprocessing stage, resulting in high dimensionality of fused agricultural input features. By reducing the dynamic suppression benchmark for cross-correlation redundancy of fused agricultural input features, the strictness of redundancy judgment can be reduced, allowing more highly correlated redundant feature pairs to be effectively identified. Redundant features are eliminated or merged during feature fusion, thereby reducing the repeated superposition of highly similar features in the fused feature set, reducing the redundancy ratio of fused agricultural input features, and further improving the accuracy of advertising content generation. Attached Figure Description

[0019] Figure 1 This is an overall structural block diagram of the precision advertising delivery system for agricultural input terminals according to an embodiment of the present invention; Figure 2 This is a logical flowchart illustrating the process of determining the effective quality of the generated advertising strategy in the precision advertising delivery system for agricultural input terminals according to an embodiment of the present invention. Figure 3This is a flowchart illustrating the process of determining the weight sparsity filtering domain of a deep learning model in a precision advertising delivery system for agricultural input terminals, as described in an embodiment of the present invention. Figure 4 This is an overall flowchart of the precise advertising delivery method for agricultural input terminals according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0021] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0022] Please see Figure 1 As shown, it is an overall structural block diagram of the precision advertising delivery system for agricultural input terminals according to an embodiment of the present invention.

[0023] An embodiment of the present invention provides a precision advertising delivery system for agricultural input terminals, comprising: The data processing module includes a collection unit for collecting agricultural input data from agricultural input terminals, a preprocessing unit connected to the collection unit for preprocessing the agricultural input data to output standard agricultural input data, and a fusion unit connected to the preprocessing unit for sequentially extracting and fusing features from the standard agricultural input data to output fused agricultural input features. The analysis module, which is connected to the data processing module, includes a training unit for training an initial model based on the integrated agricultural input characteristics to obtain a deep learning model, and an analysis unit connected to the training unit for analyzing agricultural input data based on the deep learning model to obtain analysis results. An advertising delivery module, which is connected to the analysis module, includes a strategy generation unit for generating advertising delivery strategies based on the analysis results, and a delivery unit connected to the strategy generation unit for delivering corresponding advertising content to different users based on the advertising delivery strategies. A generation and judgment module, which is connected to the advertising delivery module, is used to determine whether the effective quality of the generated advertising delivery strategy meets the requirements based on the generation accuracy of the advertising delivery strategy. A filtering adjustment module, which is connected to the generation and determination module, is used to determine the weight sparsification filtering domain of the deep learning model based on the redundancy ratio of the integrated agricultural input features when the effective quality generated by the advertising delivery strategy does not meet the requirements. The suppression adjustment module, which is connected to the filtering adjustment module, is used to determine the cross-correlation redundancy dynamic suppression benchmark of the fused agricultural input characteristics based on the noise reduction efficiency of standard agricultural input data.

[0024] Specifically, agricultural input terminals include agricultural input store management platforms, township agricultural input service platforms, and agricultural cooperative platforms.

[0025] Specifically, agricultural input data includes user purchase records, types of crops planted, and fertilization records.

[0026] Specifically, preprocessing includes data cleaning, noise reduction, and data standardization.

[0027] Specifically, standard agricultural input data includes user purchase records after data cleaning, crop types after noise reduction, and fertilizer records after data standardization.

[0028] Specifically, the characteristics of integrated agricultural inputs include crop-fertilizer interaction characteristics, procurement-fertilizer correlation, and fertilization-seasonal matching.

[0029] Specifically, the process of training an initial model based on the integrated agricultural input characteristics to obtain a deep learning model involves dividing the integrated agricultural input characteristics into a training set, a validation set, and a test set. The initial model is trained using the training set, the parameters are optimized and adjusted using the validation set, and finally the quality of the generated model is evaluated using the test set to obtain the deep learning model.

[0030] Specifically, the initial model is an initial framework that has basic capabilities for extracting agricultural input data features and perceiving the state of agricultural input scenarios, and is applicable to the collection of agricultural input terminal data, multi-source feature fusion, and the pre-generation of advertising placement strategies.

[0031] Specifically, the process of analyzing agricultural input data using a deep learning model to obtain analysis results involves inputting preprocessed standard agricultural input data into the deep learning model, which then performs comprehensive reasoning and classification based on the integrated agricultural input features, and finally outputs the analysis results.

[0032] Specifically, the deep learning model can be a recurrent neural network model, a Transformer model, or a convolutional neural network model, with the preferred embodiment being the Transformer model.

[0033] Specifically, the analysis results include the product categories in demand, the probability of end-user purchases, and recommended product specifications.

[0034] Specifically, advertising strategies include channel selection, time slot planning, and advertising frequency.

[0035] Specifically, the advertising content includes applicable crop types, purchasing guidance information, and explanations of product advantages.

[0036] In implementation, the system of this invention sets up a data processing module, an analysis module, an advertising delivery module, a generation and judgment module, a filtering and adjustment module, and a suppression and adjustment module. It judges the effective quality of the generated advertising delivery strategy based on the accuracy of the generated strategy. Due to insufficient model generalization ability, it cannot fully learn the actual agricultural input needs and inherent laws of planting in different planting scenarios, leading to deviations between the generated advertising delivery strategy and the actual planting problems of end users, thus reducing the adaptability and accuracy of the advertising. By judging the effective quality of the generated advertising delivery strategy, the generalization defects of the model can be located in reverse, and the advertising delivery strategy generation strategy and model output weights can be dynamically corrected to improve the matching accuracy between the advertising delivery strategy and the user's planting needs. The system adjusts the weight sparsity filtering domain of the deep learning model based on the redundancy ratio of the fused agricultural input features. Because the dimensionality of the agricultural input features generated after feature fusion is too high, the model cannot effectively filter and compress redundant features, introducing a large number of ineffective cross combinations during training, leading to model overfitting. The model suffers from overfitting and insufficient generalization ability. By increasing the weight sparsity filtering domain of the deep learning model, the model's ability to remove invalid weights can be strengthened, reducing the proportion of parameters corresponding to redundant agricultural input features, reducing the interference of invalid feature cross-combinations on model training, alleviating overfitting, and improving the model's generalization ability when collecting data from different agricultural input terminals. The cross-correlation redundancy dynamic suppression benchmark of the fused agricultural input features is adjusted according to the denoising effectiveness of standard agricultural input data. Due to the large number of original features and the high proportion of noise and redundant features in the collected agricultural input terminal data, effective feature screening and denoising cannot be achieved in the preprocessing stage, resulting in high dimensionality of the fused agricultural input features. By reducing the cross-correlation redundancy dynamic suppression benchmark of the fused agricultural input features, the redundancy judgment strictness can be reduced, allowing more highly correlated redundant feature pairs to be effectively identified. Redundant features are removed or merged during the feature fusion process, thereby reducing the repeated superposition of highly similar features in the fused feature set, reducing the redundancy proportion of the fused agricultural input features, and improving the accuracy of advertising content generation.

[0037] Please see Figure 2 As shown, this is a logical flowchart of the process for determining the effective quality of the generated advertising placement strategy according to an embodiment of the present invention. In this embodiment, the effective quality of the generated advertising placement strategy is quantified based on the accuracy of the strategy generation. The accuracy of ad placement strategy generation is determined by the ratio of the number of accurately generated ad placement strategies to the total number of generated ad placement strategies. Accurately generate advertising strategies that align with users' actual planting scenarios and agricultural input needs.

[0038] Specifically, the generation and determination module responds to the accuracy of the generated advertising strategy to determine whether the effective quality of the generated advertising strategy meets the requirements.

[0039] Specifically, the generation and determination module determines that the effective quality of the generated advertising strategy meets the requirements when the generation accuracy rate of the advertising strategy is greater than or equal to the preset generation accuracy rate. The generation and determination module determines that the effective quality of the generated advertising strategy does not meet the requirements if the generation accuracy rate of the advertising strategy is less than the preset generation accuracy rate.

[0040] It is understandable that the preset generation accuracy rate is based on the statistical distribution of historical statistical samples by those skilled in the art. Preferably, the preset generation accuracy rate is positively correlated with the effective quality of the generated ad placement strategy, wherein... The preset generation accuracy rate is based on the number of historical statistical samples. When the generation accuracy rate of the advertising strategy is greater than or equal to this value, it indicates that the currently generated advertising strategy has a high degree of consistency with the actual planting scenario and agricultural input needs of end users. The strategy generation can accurately match user needs based on the integration of agricultural input characteristics and related environmental data, and make effective decisions on the end user's placement priority, target channels, content type, and placement timing. Therefore, the effective quality of the generated advertising strategy is judged to meet the requirements. When the generation accuracy rate of the advertising strategy is less than this value, it indicates that there is a large deviation between the currently generated advertising strategy and the actual planting scenario and agricultural input needs of end users. The strategy generation has insufficient learning ability on the correlation between agricultural input characteristics and environmental factors, resulting in reduced reliability of the advertising strategy. This, in turn, affects the accuracy of subsequent advertising content generation and placement precision. Therefore, the effective quality of the generated advertising strategy is judged to not meet the requirements.

[0041] Regarding the preset accuracy rate, those skilled in the art can set the base number based on the number of historical statistical samples, and can set it according to the actual situation, which will not be elaborated here.

[0042] In practice, the system described in this invention determines the effective quality of the generated advertising strategies by setting a preset generation accuracy rate. Due to insufficient model generalization ability, it cannot fully learn the actual agricultural input needs of users and the inherent laws of planting in different planting scenarios, resulting in deviations between the generated advertising strategies and the actual planting problems of end users. This reduces the adaptability and accuracy of advertising. By determining the effective quality of the generated advertising strategies, the generalization defects of the model can be located in reverse, and the weights of the advertising strategy generation strategy and model output can be dynamically corrected to improve the matching accuracy between the advertising strategy and the user's planting needs, thereby further improving the accuracy of advertising content generation.

[0043] Please see Figure 3 The diagram shown is a logical flowchart illustrating the process of determining the weight sparsity filtering domain of a deep learning model according to an embodiment of the present invention. In this embodiment, the generalization ability of the deep learning model is quantified based on the redundancy ratio of fused agricultural input features. The redundancy ratio of integrated agricultural input features is determined by the ratio of the number of redundant features in the integrated agricultural input features to the total number of integrated agricultural input features; Redundancy refers to repetitive, highly correlated, or feature information that does not actually contribute to the generation of advertising strategies in the integration of agricultural input features.

[0044] Specifically, in response to the condition that the effective quality generated by the advertising delivery strategy does not meet the requirements, the filtering adjustment module determines whether the generalization ability of the deep learning model meets the requirements based on the redundancy ratio of the integrated agricultural input features.

[0045] Specifically, the filtering adjustment module determines that the generalization ability of the deep learning model meets the requirements when the redundancy ratio of the integrated agricultural input features is less than or equal to a preset first ratio. The filtering adjustment module determines that the generalization ability of the deep learning model does not meet the requirements when the redundancy ratio of the integrated agricultural input features is greater than the preset first ratio.

[0046] Specifically, the filtering adjustment module responds to the fact that the redundancy ratio of the integrated agricultural input features is greater than the preset first ratio and less than the preset second ratio by increasing the weight sparsity filtering domain of the deep learning model. The filtering adjustment module responds to the fact that the redundancy ratio of the integrated agricultural input features is greater than or equal to the preset second ratio, and initially determines that the reasonable dimension capability of the integrated agricultural input features does not meet the requirements. It then determines whether the reasonable dimension capability of the integrated agricultural input features meets the requirements based on the noise reduction efficiency of the standard agricultural input data.

[0047] In this embodiment, the preset first proportion is less than the preset second proportion, wherein... If the redundancy ratio of the integrated agricultural input features is less than or equal to the preset first ratio, the corresponding situation is that the generalization ability of the deep learning model meets the requirements. If the redundancy ratio of the fused agricultural input features is greater than the first preset ratio but less than the second preset ratio, the corresponding situation is that the agricultural input features generated after feature fusion have too high dimensionality, the model cannot effectively filter and compress redundant features, and a large number of invalid cross combinations are introduced during the training process, which aggravates the overfitting phenomenon of the model and results in insufficient generalization ability. If the redundancy ratio of the integrated agricultural input features is greater than or equal to the preset second ratio, the corresponding situation is that due to the large number of original features and the high proportion of noise and redundant features in the collected agricultural input terminal data, effective feature screening and noise reduction cannot be achieved in the preprocessing stage, resulting in high dimensionality of the integrated agricultural input features.

[0048] It is understood that the preset first proportion and preset second proportion are values ​​determined by those skilled in the art based on the statistical distribution of the redundancy proportion of integrated agricultural input features in the quality and timeliness of advertising content generation, and in combination with the generalization ability of deep learning models corresponding to different redundancy proportions of integrated agricultural input features. Preferably, the redundancy proportion of integrated agricultural input features is negatively correlated with the generalization ability of the deep learning model, wherein, The first preset percentage is the redundancy percentage corresponding to the inflection point where the accuracy of advertising content generation begins to decline as the redundancy percentage of integrated agricultural input features increases, based on historical statistical distribution. When the redundancy percentage of integrated agricultural input features is less than or equal to this value, the generalization ability of the deep learning model remains at a high level, indicating that the information redundancy of integrated agricultural input features is low, the proportion of effective independent features in the feature set is high, the model can learn robustly based on diverse features, the fusion strategy of the feature fusion unit is reasonable, and the model has good predictive stability in unseen new planting scenarios. The inflection point is determined based on the paired data of the accuracy of advertising content generation and the proportion of redundancy of integrated agricultural input features recorded in history. It is determined by statistically analyzing the downward trend of the accuracy of advertising content generation as the proportion of redundancy of integrated agricultural input features changes, and identifying the node where the rate of decline of the accuracy of advertising content generation first increases significantly. The baseline for setting the accuracy of advertising content generation is a standard reference value pre-set by those skilled in the art based on the application needs of agricultural input advertising placement tasks, the maximum allowable deviation range of the advertising content generation model, and the average level of advertising content generation accuracy under historical stable operating conditions.

[0049] The second presupposed percentage is the redundancy percentage of the integrated agricultural input features corresponding to the point where the accuracy of advertising content generation drops significantly in the historical statistical distribution. When the redundancy percentage of the integrated agricultural input features is greater than or equal to this value, it indicates that there is a large amount of redundant information in the integrated agricultural input features, the effective independent features are seriously insufficient, the model relies too much on repetitive information during training, which leads to the model overfitting to noise and random fluctuations in the training data, and the generalization ability drops significantly. The critical point of decline is determined based on the paired data of the accuracy of advertising content generation and the proportion of redundancy of integrated agricultural input features recorded in history. It is determined by statistically analyzing the downward trend of the accuracy of advertising content generation as the proportion of redundancy of integrated agricultural input features changes, and identifying the critical node where the accuracy of advertising content generation declines rapidly. Specifically, the process of determining the severe decline point based on the historical synchronous record of advertising content generation accuracy and the redundancy ratio of integrated agricultural input features is as follows: the historical data is sorted from small to large according to the redundancy ratio of integrated agricultural input features, and the decline rate of advertising content generation accuracy is counted segment by segment. When the decline rate of advertising content generation accuracy significantly increases after a certain redundancy ratio node and exceeds the preset accuracy decline rate benchmark, then the node is determined as the severe decline point.

[0050] The size of the preset first proportion and the preset second proportion depends on the number of historical statistical samples and the baseline set by those skilled in the art for the generalization ability of deep learning models. They can be set according to the actual situation and will not be elaborated here.

[0051] Specifically, the increase in the weight sparsity filtering domain of the deep learning model is determined by the difference between the redundancy ratio of the fused agricultural input features and the preset first ratio.

[0052] Specifically, the increase in the weight sparsity filtering domain of the deep learning model is achieved by obtaining the redundancy ratio of the integrated agricultural input features and the preset first ratio, calculating the difference, and using the difference as the benchmark for the increase in the weight sparsity filtering domain. If the difference is greater than 0, it indicates that the proportion of redundant features in the current integrated agricultural input features has deviated from the preset standard, and there is a problem of excessive feature redundancy and increased risk of model overfitting. The weight sparsity filtering domain of the deep learning model is increased according to the magnitude corresponding to the difference. The larger the difference value, the higher the increase in the weight sparsity filtering domain. If the difference is equal to or less than 0, it indicates that the redundancy ratio of the current integrated agricultural input features is within the preset standard range, and the generalization ability of the deep learning model meets the requirements. The weight sparsity filtering domain is not increased and the original weight sparsity filtering domain remains unchanged.

[0053] Specifically, the step of increasing the weight sparsity filtering domain of the deep learning model according to the magnitude corresponding to the difference is to use the difference between the preset first proportion and the redundancy proportion of the integrated agricultural input features as the basic magnitude benchmark, pre-establish a mapping rule between the difference value and the increase magnitude of the weight sparsity filtering domain, substitute the obtained difference into the mapping relationship, and match the unique corresponding increase magnitude of the weight sparsity filtering domain.

[0054] In practice, the system of this invention adjusts the weight sparsity filtering domain of the deep learning model by setting a preset first proportion and a preset second proportion. Because the agricultural input features generated after feature fusion have too high dimensionality, the model cannot effectively filter and compress redundant features, and a large number of invalid cross combinations are introduced during training, which aggravates the overfitting phenomenon and reduces the generalization ability. By increasing the weight sparsity filtering domain of the deep learning model, the model can strengthen the removal of invalid weights, reduce the proportion of parameters corresponding to redundant agricultural input features, reduce the interference of invalid feature cross combinations on model training, alleviate overfitting, improve the generalization ability of the model when collecting data from different agricultural input terminals, and further improve the accuracy of advertising content generation.

[0055] In this embodiment, the noise reduction efficiency based on standard agricultural input data is used to quantify the reasonable dimensionality of agricultural input characteristics. The noise reduction efficiency of standard agricultural input data is determined based on the ratio of the amount of noise data that has been effectively reduced to the total amount of noise data in the agricultural input data after noise reduction processing. Effective data refers to real noise data that has been successfully identified and removed during the preprocessing process.

[0056] Specifically, the noise reduction efficiency of the suppression and adjustment module in response to standard agricultural input data is greater than or equal to the preset efficiency, and its ability to determine the reasonable dimensions of the integrated agricultural input features meets the requirements. The suppression adjustment module responds to the fact that the noise reduction efficiency of the standard agricultural input data is less than the preset efficiency, determines that the ability to integrate reasonable dimensions of agricultural input features does not meet the requirements, and reduces the cross-correlation redundancy dynamic suppression benchmark of the integrated agricultural input features.

[0057] It is understandable that the preset effectiveness rate is a value determined by those skilled in the art based on the statistical distribution of the denoising effectiveness rate of standard agricultural input data in the generalization of deep learning models, and in combination with the reasonable dimensionality capability of fusion of agricultural input features corresponding to different effectiveness rates. Preferably, the denoising effectiveness rate of standard agricultural input data is positively correlated with the reasonable dimensionality capability of fusion of agricultural input features. The preset effectiveness rate is the denoising effectiveness rate corresponding to the inflection point where the redundancy ratio of the fused agricultural input features in the historical statistical distribution begins to decrease as the denoising effectiveness rate of the standard agricultural input data increases. When the denoising effectiveness rate of the standard agricultural input data is greater than or equal to this value, it indicates that the preprocessing unit can effectively identify and remove noise features in the agricultural input data, the fused agricultural input feature set contains less noise information, the feature purity is high, the proportion of redundant features in the fused feature set is maintained at a low level, the dimensions of the fused agricultural input features are reasonable, and it can provide high-quality feature input for the deep learning model, thereby ensuring that the model can learn stably based on effective feature information during the training process. The inflection point is determined based on the historical synchronously recorded ratio of redundancy of integrated agricultural input features and the noise reduction efficiency of standard agricultural input data. It is determined by statistically analyzing the downward trend of the ratio of redundancy of integrated agricultural input features with the noise reduction efficiency and identifying the node where the rate of decrease of the ratio of redundancy of integrated agricultural input features first increases significantly. The baseline for setting the proportion of redundancy in integrated agricultural input features is a standard reference value pre-set by those skilled in the art based on the requirements of deep learning models for feature purity, the maximum redundancy tolerance range allowed by the model, and the average level of the proportion of redundancy in integrated agricultural input features under historical stable operating conditions.

[0058] The preset effectiveness depends on the number of historically collected samples and the baseline for the generalization ability of deep learning models set by those skilled in the art. It can be set according to the actual situation and will not be elaborated here.

[0059] Specifically, the reduction range of the cross-correlation redundancy dynamic suppression benchmark that integrates agricultural input characteristics is determined by the difference between the preset effectiveness and the noise reduction effectiveness of standard agricultural input data.

[0060] Specifically, the reduction range of the cross-correlation redundancy dynamic suppression benchmark for the fused agricultural input features is determined by obtaining the preset effectiveness rate and the denoising effectiveness rate of the standard agricultural input data, calculating the difference between the preset effectiveness rate and the denoising effectiveness rate of the standard agricultural input data, and using this difference as the benchmark for the reduction range of the cross-correlation redundancy dynamic suppression benchmark. If the difference is greater than 0, it indicates that the denoising effectiveness rate of the current standard agricultural input data has not reached the preset standard, and there is a risk of insufficient denoising capability in the preprocessing stage, excessive noise features and redundant features remaining, leading to an excessively high dimension of the fused agricultural input features. Therefore, it is necessary to effectively lower the redundancy judgment threshold in the fusion process to expand the redundancy features. The range of feature identification and elimination is determined by reducing the cross-correlation redundancy dynamic suppression benchmark of the integrated agricultural input features according to the magnitude corresponding to the difference. The larger the difference value, the greater the gap between the noise reduction efficiency and the preset standard, the more redundant information remains in the integrated agricultural input features, and the stronger the need to lower the redundancy suppression benchmark. Therefore, the reduction of the cross-correlation redundancy dynamic suppression benchmark is greater. If the difference is equal to or less than 0, it indicates that the noise reduction efficiency of the current standard agricultural input data is within the preset standard range, the dimensional rationality of the integrated agricultural input features meets the requirements, the cross-correlation redundancy dynamic suppression benchmark is not reduced, and the original cross-correlation redundancy dynamic suppression benchmark remains unchanged.

[0061] Specifically, the reduction of the cross-correlation redundancy dynamic suppression benchmark of the fused agricultural input features according to the amplitude corresponding to the difference is based on the difference between the preset efficiency and the noise reduction efficiency of the standard agricultural input data as the basic adjustment amplitude benchmark. A mapping rule between the noise reduction efficiency difference and the reduction amplitude of the cross-correlation redundancy dynamic suppression benchmark is established in advance. The obtained difference is substituted into the mapping relationship to match a unique corresponding cross-correlation redundancy dynamic suppression benchmark adjustment value. Based on the adjustment value, the cross-correlation redundancy dynamic suppression benchmark of the fused agricultural input features is reduced. This allows the feature fusion unit to lower the correlation coefficient threshold for judging the redundancy relationship between features during the processing of standard agricultural input data, expand the identification range of highly correlated redundant feature pairs, improve the elimination rate of redundant features in the fusion stage, reduce the redundancy ratio of the fused agricultural input features, suppress the interference of high-dimensional features on model training, and improve the generalization ability of the deep learning model in new planting scenarios.

[0062] In implementation, the system of the present invention adjusts the dynamic suppression benchmark of cross-correlation redundancy of integrated agricultural input features by setting a preset effectiveness rate. Since the original features of the collected agricultural input terminal data are numerous and the proportion of noise and redundant features is high, effective feature screening and noise reduction cannot be achieved in the preprocessing stage, resulting in high dimensionality of integrated agricultural input features. By reducing the dynamic suppression benchmark of cross-correlation redundancy of integrated agricultural input features, the strictness of redundancy judgment can be reduced, allowing more highly correlated redundant feature pairs to be effectively identified. Redundant features are eliminated or merged during feature fusion, thereby reducing the repeated superposition of highly similar features in the fused feature set, reducing the redundancy ratio of integrated agricultural input features, and further improving the accuracy of advertising content generation.

[0063] Please see Figure 4 As shown, it is an overall flowchart of the precise advertising delivery method for agricultural input terminals according to an embodiment of the present invention.

[0064] This invention also provides a method for precise advertising delivery to agricultural input terminals, including: Step S1: Preprocess the collected agricultural input data from the agricultural input terminal to output standard agricultural input data, and sequentially extract and fuse features from the standard agricultural input data to output fused agricultural input features. Step S2: Train the initial model based on the integrated agricultural input features to obtain a deep learning model, use the deep learning model to analyze the agricultural input data to obtain analysis results, and generate an advertising delivery strategy based on the analysis results to deliver corresponding advertising content to different users. Step S3: Obtain the generation accuracy of the advertising placement strategy, and determine whether the effective quality of the generated advertising placement strategy meets the requirements based on the generation accuracy of the advertising placement strategy. Step S4: If the effective quality generated based on the advertising delivery strategy does not meet the requirements, then the generalization ability of the deep learning model is determined based on the redundancy ratio of the integrated agricultural input features. Step S5: If the generalization ability of the deep learning model does not meet the requirements, determine whether to increase the weight sparsity filtering domain of the deep learning model. Step S6: If it is not necessary to increase the weight sparsity filtering domain of the deep learning model, then determine the cross-correlation redundancy dynamic suppression benchmark based on the denoising efficiency of standard agricultural input data.

[0065] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A precision advertising delivery system for agricultural input terminals, characterized in that, include: The data processing module includes a collection unit for collecting agricultural input data from agricultural input terminals, a preprocessing unit connected to the collection unit for preprocessing the agricultural input data to output standard agricultural input data, and a fusion unit connected to the preprocessing unit for sequentially extracting and fusing features from the standard agricultural input data to output fused agricultural input features. The analysis module, which is connected to the data processing module, includes a training unit for training an initial model based on the integrated agricultural input characteristics to obtain a deep learning model, and an analysis unit connected to the training unit for analyzing agricultural input data based on the deep learning model to obtain analysis results. An advertising delivery module, which is connected to the analysis module, includes a strategy generation unit for generating advertising delivery strategies based on the analysis results, and a delivery unit connected to the strategy generation unit for delivering corresponding advertising content to different users based on the advertising delivery strategies. A generation and judgment module, which is connected to the advertising delivery module, is used to determine whether the effective quality of the generated advertising delivery strategy meets the requirements based on the generation accuracy of the advertising delivery strategy. A filtering adjustment module, which is connected to the generation and determination module, is used to determine the weight sparsification filtering domain of the deep learning model based on the redundancy ratio of the integrated agricultural input features when the effective quality generated by the advertising delivery strategy does not meet the requirements. The suppression adjustment module, which is connected to the filtering adjustment module, is used to determine the cross-correlation redundancy dynamic suppression benchmark of the fused agricultural input characteristics based on the noise reduction efficiency of standard agricultural input data.

2. The precision advertising delivery system for agricultural input terminals according to claim 1, characterized in that, The generation and determination module responds to the accuracy of the generated advertising strategy to determine whether the effective quality of the generated advertising strategy meets the requirements.

3. The precision advertising delivery system for agricultural input terminals according to claim 2, characterized in that, The generation and determination module determines that the effective quality of the generated advertising strategy meets the requirements if the generation accuracy rate of the advertising strategy is greater than or equal to the preset generation accuracy rate. The generation and determination module determines that the effective quality of the generated advertising strategy does not meet the requirements if the generation accuracy rate of the advertising strategy is less than the preset generation accuracy rate.

4. The precision advertising delivery system for agricultural input terminals according to claim 3, characterized in that, In response to the condition that the effective quality generated by the advertising delivery strategy does not meet the requirements, the filtering adjustment module determines whether the generalization ability of the deep learning model meets the requirements based on the redundancy ratio of the integrated agricultural input features.

5. The precision advertising delivery system for agricultural input terminals according to claim 4, characterized in that, The filtering adjustment module determines that the generalization ability of the deep learning model meets the requirements when the redundancy ratio of the integrated agricultural input features is less than or equal to a preset first ratio. The filtering adjustment module determines that the generalization ability of the deep learning model does not meet the requirements when the redundancy ratio of the integrated agricultural input features is greater than the preset first ratio.

6. The precision advertising delivery system for agricultural input terminals according to claim 5, characterized in that, The filtering adjustment module responds to the fact that the redundancy ratio of the integrated agricultural input features is greater than the preset first ratio and less than the preset second ratio by increasing the weight sparsity filtering domain of the deep learning model. The filtering adjustment module responds to the fact that the redundancy ratio of the integrated agricultural input features is greater than or equal to the preset second ratio, and initially determines that the reasonable dimension capability of the integrated agricultural input features does not meet the requirements. It then determines whether the reasonable dimension capability of the integrated agricultural input features meets the requirements based on the noise reduction efficiency of the standard agricultural input data.

7. The precision advertising delivery system for agricultural input terminals according to claim 6, characterized in that, The increase in the weight sparsity filtering domain of the deep learning model is determined by the difference between the redundancy ratio of the fused agricultural input features and the preset first ratio.

8. The precision advertising delivery system for agricultural input terminals according to claim 6, characterized in that, The suppression adjustment module responds to the noise reduction efficiency of standard agricultural input data being greater than or equal to the preset efficiency, and determines that the ability to integrate reasonable dimensions of agricultural input features meets the requirements. The suppression adjustment module responds to the fact that the noise reduction efficiency of the standard agricultural input data is less than the preset efficiency, determines that the ability to integrate reasonable dimensions of agricultural input features does not meet the requirements, and reduces the cross-correlation redundancy dynamic suppression benchmark of the integrated agricultural input features.

9. The precision advertising delivery system for agricultural input terminals according to claim 8, characterized in that, The reduction range of the cross-correlation redundancy dynamic suppression benchmark that integrates agricultural input characteristics is determined by the difference between the preset effectiveness rate and the noise reduction effectiveness rate of standard agricultural input data.

10. A method for precise advertising delivery applied to the precise advertising delivery system for agricultural input terminals as described in any one of claims 1-9, characterized in that, include: The collected agricultural input data from agricultural input terminals is preprocessed to output standard agricultural input data, and the standard agricultural input data is then sequentially subjected to feature extraction and fusion to output fused agricultural input features. The initial model is trained based on the integrated agricultural input features to obtain a deep learning model. The deep learning model is used to analyze the agricultural input data to obtain analysis results. Based on the analysis results, an advertising delivery strategy is generated to deliver corresponding advertising content to different users. Obtain the generation accuracy rate of the advertising placement strategy, and determine whether the effective quality of the generated advertising placement strategy meets the requirements based on the generation accuracy rate; If the quality of the effective products generated based on the advertising strategy does not meet the requirements, then the generalization ability of the deep learning model is determined based on the redundancy ratio of the integrated agricultural input features. If the generalization ability of the deep learning model does not meet the requirements, determine whether to increase the weight sparsity filtering domain of the deep learning model. If it is not necessary to increase the weight sparsity filtering domain of the deep learning model, then the cross-correlation redundancy dynamic suppression benchmark for integrating agricultural input features is determined based on the denoising efficiency of standard agricultural input data.

Images