Agricultural precision planting pre-training and fine-tuning training method based on AI large model

Abstract

The invention discloses an agricultural precision planting pre-training and fine-tuning training method based on an AI large model. Based on a deep learning technology, key information is obtained from large-scale agricultural data, an optimal planting mode is predicted, and the crop yield and quality are improved. Through analysis of historical data and real-time monitoring, peasants are predicted and reminded about possible plant diseases and insect pests, so that measures are taken in advance. For different regions and different environmental conditions, the model can adapt to specific planting requirements through fine adjustment, and the universality and accuracy of the model are improved. The prediction of soil, weather and plant growth modes provides intelligent decision support for agricultural production, helps farmers to optimize a planting scheme, improve the yield and reduce resource waste, thereby improving the scientization and production benefits of agriculture. The model can help farmers to scientifically plant, improve crop yield and quality, save resources, reduce environmental burden and promote agricultural sustainable development.

Description

Technical Field

[0001] The present invention relates to the field of agricultural science and technology, and in particular to a method for pre-training and fine-tuning agricultural precision planting based on an AI large model. Background Art

[0002] With the advancement of agricultural modernization, precision agriculture has become an important way to improve agricultural production efficiency and reduce resource waste. In this context, the application of artificial intelligence technology has brought new opportunities to the agricultural field. In particular, the use of AI big models for agricultural precision planting pre-training and fine-tuning training has become a technology with great potential. The article "Agricultural Artificial Intelligence Technology: The Wings of Modern Agricultural Science and Technology" published in the journal "Journal of South China Agricultural University" in 2024 reviewed several key technologies of agricultural artificial intelligence and the current status of artificial intelligence in the application research fields of planting, poultry and livestock, and agricultural product traceability and grading. The article "Application of Artificial Intelligence Big Models in Smart Agriculture" published in the journal "Agricultural Engineering Technology" in 2024 introduced the basic concepts of artificial intelligence big models, analyzed the current status and challenges of smart agriculture, and elaborated on the application of artificial intelligence big models in the field of smart agriculture.

[0003] Traditional agricultural planting methods are often subject to complex factors such as geography, meteorology, and soil, resulting in fluctuations in production efficiency and quality. The pre-training and fine-tuning training methods based on AI large models, through deep learning technology, can pre-train the model based on a large amount of agricultural data, so that the model has better intelligence and adaptability. Subsequently, through fine-tuning training in actual farmland environments, the model is optimized to adapt to different agricultural scenarios, and intelligent analysis and prediction of multiple factors such as crop growth, soil conditions, and meteorological changes can be achieved. This technical background highlights the potential advantages of using AI large models for precision agricultural planting, including improving agricultural production efficiency, reducing production costs, and reducing resource waste, thereby promoting the sustainable development of modern agriculture. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an agricultural precision planting pre-training and fine-tuning training method based on an AI large model.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A method for agricultural precision planting pre-training and fine-tuning training based on an AI large model, comprising the following steps:

[0007] Step 1: Data preprocessing: Data will be collected from open source datasets and field data collection, and soil, weather, crop growth, crop type, location, time, crop problems and solutions will be cleaned and organized into a dataset, which consists of the following specific steps:

[0008] Step 1-1: Identify and handle missing values: Identify missing values ​​in the data set and mark them as preset placeholders. If the number of missing data is less than the preset amount, delete the rows or columns containing missing values. If the number of missing data exceeds the preset amount, fill the missing values ​​with the mean of the corresponding data or the adjacent data.

[0009] Step 1-2: Data conversion: Use the minimum and maximum values ​​to normalize the values ​​of each feature in the data set; the normalization method is:

[0010] x_norm = (x-min) / (max-min)

[0011] Among them, x norm is the normalized value, min and max are the minimum and maximum values ​​respectively;

[0012] Step 1-3: Identify and handle outliers: Identify values ​​greater than the standard deviation as outliers; Use the normal values ​​before and after to replace the outliers;

[0013] Step 2: Pre-training: The model structure consists of one or more stacked Transformer decoders; each decoder includes two main components: a self-attention mechanism and a feedforward neural network; the self-attention mechanism in the model allows the model to capture longer-term dependencies between different features when processing each feature;

[0014] Use the training set to pre-train the model: the region, soil moisture, meteorological information, and crop type information in the data set are used as the independent variables X = {x1, x1, ···, x m}, where m is the number of independent variables, the solution strategies for the crop problems in the dataset and the corresponding expert advice information are used as the label information of the model Y = {y1,y1,···,y n}, where y n is an output answer type of the model, and n is the number of output answer types; the loss function in the model training process uses the cross entropy loss function:

[0015]

[0016] in Represents the distribution of results predicted by the model;

[0017] Step 3: Fine-tuning: Use real-time farmland data, crop growth conditions, problem-solving strategies, and expert advice to fine-tune the pre-trained large model;

[0018] Step 4: Real-time monitoring and adjustment: Deploy the fine-tuned model in the agricultural production environment, monitor farmland data in real time, and output precise planting recommendations in real time.

[0019] Furthermore, the nearest neighbor method is used in step 1 to determine the missing values ​​to be filled. First, the K samples closest to the sample with missing data are determined based on the Euclidean distance, and then these K values ​​are weighted averaged to fill the current missing value.

[0020] Furthermore, the method used to identify outliers in steps 1-3 is the density-based spatial clustering method DBSCAN.

[0021] Furthermore, the humidity index in step 1-3 is the ratio of precipitation to temperature, and the seasonal cycle position is encoded by sine and cosine functions.

[0022] Furthermore, in step 2, the model training parameters, random seed is set to 3407, and learning rate is set to 5e -4 , the selected optimizer is set to Adam.

[0023] The beneficial effects of the present invention are:

[0024] 1. Through pre-training of large models, the system can learn complex soil, weather and plant growth patterns from a large amount of agricultural data, enabling the model to have a deeper understanding of the agricultural environment. Through fine-tuning training, the large model can be personalized and optimized according to the actual farmland conditions, improving the model's adaptability to specific regions and crops. This comprehensive pre-training and fine-tuning training method has significantly improved the level of intelligent agricultural production and provided farmers with more scientific and accurate planting plans.

[0025] 2. The present invention predicts the edge wear of farmland crops, which effectively improves the accurate identification of crop status. Through large-scale model learning and fine-tuning, the system can accurately predict the degree of crop wear at different growth stages, and then provide farmers with reasonable agricultural management suggestions. This technology improves the accuracy of crop status monitoring, helps farmers better adjust planting strategies, and minimizes the quality degradation caused by plants being in an undesirable environment for a long time.

[0026] 3. Compared with traditional methods, the present invention improves the efficiency of agricultural data processing. The pre-training stage enables the large model to automatically learn the key features in large-scale agricultural data, while the fine-tuning stage performs personalized adjustments based on actual farm data. This process not only makes full use of a large amount of data, but also avoids the tedious process of manually adjusting parameters in traditional methods, improves the computational efficiency of the algorithm, and enables farmers to obtain personalized agricultural management solutions more quickly.

[0027] 4. The present invention integrates multiple feature information, including soil data, meteorological data, and plant growth data, thereby achieving comprehensive optimization of agricultural precision planting. By integrating different features, the system can more comprehensively evaluate the status of the farm and improve the accuracy of fragment similarity. This method of multi-feature information fusion enables the system to more comprehensively consider multiple factors of agricultural production, thereby improving the accuracy and operability of planting recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0029] The following examples serve to illustrate the invention.

[0030] Example 1

[0031] A method for agricultural precision planting pre-training and fine-tuning training based on an AI large model, comprising the following steps:

[0032] Step 1: Data collection: Obtain a large amount of agricultural data, including but not limited to soil information, meteorological data, crop growth data, time, location, crop type, solutions to crop problems, and expert advice. These data can come from various sensors, such as: satellite remote sensing and meteorological station agricultural data collection equipment, agricultural department documents, and real sample data collected during the actual planting process, as well as various agricultural planting data sets: China's crop growth and development and farmland soil moisture ten-day value data set, China's main crop growth and development data set (V2.0).

[0033] Step 2: Data preprocessing: Clean and organize the open source data sets and the actual agricultural data collected on soil, weather, crop growth, crop type, location, time, and solutions to crop problems. It consists of the following specific steps:

[0034] Step 2-1: Identify and handle missing values. First, use the isnull(), isna() functions, and double equal sign judgment symbols in the Python data analysis library Pandas to identify missing values ​​in the data set. In the data set, missing values ​​are marked as NaN, Null, empty strings, or other placeholders. We use the .isnull() method, .dropna(), and .replace() methods in the Pandas library to review all missing values ​​in the data. If the missing values ​​are less than 5% of the data volume, delete the rows or columns containing missing values. If the missing values ​​exceed 5% of the data volume, the mean is used to fill the missing values ​​for some information. For time series data, the previous time value or the next time value of the data can be used to fill it. In order to obtain more appropriate missing values, the nearest neighbor method is also used to predict missing values ​​based on other variables, that is, first determine the K samples closest to the sample with missing data based on the Euclidean distance, and then use the weighted average of these K values ​​to estimate the missing data of the current sample with missing values.

[0035] Step 2-2: Perform data conversion and use minimum-maximum normalization to convert all data features to the same scale. First, calculate the minimum value (min) and maximum value (max) of each feature. Then, use the formula x_norm = (x-min) / (max-min) to convert each value x to the normalized value x. norm , which scales all values ​​to between 0 and 1. The above formula is then applied to all numerical features in the dataset to ensure that they are all normalized.

[0036] Step 2-3: Identify and handle outliers. First, use the standard deviation to identify outliers, which may be caused by errors, faulty sensors, or atypical events. The method used to identify outliers is the density-based spatial clustering method DBSCAN. The DBSCAN algorithm has two main parameters: eps (ε, neighborhood radius) and minPts (minimum number of samples). EPS determines the neighborhood range of a point, and minPts specifies the minimum number of points contained in a cluster. The values ​​we use are minPts of 15 and eps of 14. Cluster the data in the dataset and treat points that do not belong to any cluster as noise, that is, outliers. Then judge the initially identified outliers one by one. If the outlier is wrong, it should be deleted. If the outlier is caused by a known error, it can be corrected according to the context. If the outlier is a real and important data point (for example, agricultural data under extreme climate conditions), it is retained.

[0037] Step 3: Pre-training: The structure of the model used is based on the Transformer structure. Our model is composed of multiple stacked Transformer decoders. Each decoder includes two main components: a self-attention mechanism and a feedforward neural network. The self-attention mechanism in the model allows the model to pay attention to other features in the data when processing each feature, and capture longer-term dependencies between different features. The dataset is divided into three categories: training set, validation set, and test set at a ratio of 80%, 10%, and 10%, and the model is pre-trained using the data in the training set. The region, soil moisture, meteorological information, and crop type information are used as the independent variables X={x1,x1,···,x m}, where x m is an independent variable, and m is the number of independent variables. The solution strategies for crop problems and expert advice are used as the label information of the model Y = {y1,y1,···,y n}, where y n is an answer output type of the model, and n is the number of output types. The cross entropy loss function is used during model training. The random seed is set to 3407 according to the article "Torch.manual_seed(3407)is all you need:On the influence of random seeds in deep learningarchitectures for computer vision". To avoid oscillation and overfitting of the model on the training set, the learning rate is 5e-4, and the optimizer is Adam, which can achieve fast convergence and is not very sensitive to the choice of initial learning rate, and has a strong fault tolerance. Through pre-training, the model can learn the complex relationship between soil, weather and plant growth, and can predict the problems encountered in farmland and give corresponding solutions based on given agricultural information.

[0038] Step 4: Fine-tuning: Apply the pre-trained large model to the actual farmland environment, and use the actual planting information to fine-tune the model. The data processing process is the same as step 3. The data for fine-tuning the model includes real-time farmland data, the growth of farmland crops, solutions to problems encountered in actual situations, and expert advice information. Through fine-tuning, the large model can better adapt to agricultural scenarios under different geographical, meteorological and soil conditions, making the model output more accurate and credible.

[0039] Step 5: Real-time monitoring and adjustment: Deploy the trained precision planting recommendation model in the agricultural production environment, monitor the farmland data in real time, and use the acquired data as the input X of the model.

[0040] Step 6: Output precise planting suggestions: Based on the fine-tuned large model, precise planting suggestions are output in real time. This includes but is not limited to the best crop selection, the best sowing time for the crop, fertilization schemes, irrigation strategies, and specific measures to deal with pests and diseases or expert advice during the planting process. Planting suggestions can be implemented through agricultural equipment and intelligent agricultural machinery, thereby efficiently improving the yield and quality of farmland crops.

Claims

1. A method for agricultural precision planting pre-training and fine-tuning training based on AI large model, characterized in that: The following steps are involved: Step 1: Data preprocessing: Data on soil, weather, crop growth, crop type, location, time, crop problems and solutions collected from open source datasets and actual data are organized into datasets, and data cleaning is performed. The steps are as follows: Step 1-1: Identify and handle missing values: Identify missing values ​​in the data set and mark them as preset placeholders. If the number of missing data is less than the preset amount, delete the rows or columns containing missing values. If the number of missing data exceeds the preset amount, fill the missing values ​​with the mean or adjacent data of the response data; Step 1-2: Data conversion: Use minimum-maximum normalization to convert all data to the same scale; first, calculate the minimum value (min) and maximum value (max) of each feature in the data set one by one; then, convert it to the normalized value x norm : x_norm = (x-min) / (max-min). Step 1-3: Identify and handle outliers: Use standard deviation to identify outliers; Step 2: Pre-training: The model structure is composed of more than one stacked Transformer decoder; each decoder includes two components: self-attention mechanism and feedforward neural network; the model is pre-trained with the training set: that is, the region, soil moisture, meteorological information, and crop type information in the data set are used as the independent variables X = {x1, x1, ···, x m ], where x m is an independent variable, m is the number of independent variables; the solution strategies for the crop problems in the data set and the expert advice information are used as the label information of the model Y = [y1,y1,···,y n }, where y n is an answer output type of the model, and n is the number of output types. The cross entropy loss function is used during model training: in Represents the distribution of results predicted by the model. Step 3: Fine-tuning: Use the real-time collected farmland data, the growth of farmland crops, the solutions to the problems encountered in actual situations, and the expert advice information to fine-tune the pre-trained large model; during fine-tuning, the real local values ​​collected by various sensors are pre-processed according to step 1, the data in the correct format is input into the model, and the model parameters are fine-tuned; Step 4: Real-time monitoring and adjustment: Deploy the fine-tuned model in the agricultural production environment, monitor farmland data in real time, and output precise planting suggestions in real time. If problems occur in actual production, return to step 3 and fine-tune the model with actual data.

2. The agricultural precision planting pre-training and fine-tuning training method based on AI large model according to claim 1 is characterized in that: In step 1, the nearest neighbor method is used to determine the missing values. First, the K samples closest to the sample with missing data are determined based on the Euclidean distance, and then these K values ​​are weighted averaged to fill the current missing values.

3. The agricultural precision planting pre-training and fine-tuning training method based on AI large model according to claim 1 is characterized in that: The method used to identify outliers in steps 1-3 is the density-based spatial clustering method DBSCAN.

4. The agricultural precision planting pre-training and fine-tuning training method based on the AI ​​large model according to claim 1 is characterized in that: The humidity index in step 1-3 is the ratio of precipitation to temperature, and the seasonal cycle position is encoded by sine and cosine functions.

5. The agricultural precision planting pre-training and fine-tuning training method based on AI large model according to claim 1 is characterized in that: In step 2, the model training parameters are as follows: the random seed is set to 3407 and the learning rate is set to 5e -4 , the selected optimizer is Adam.

Images