An Intelligent Recognition Method for Farmland Weeds Based on Evolutionary PU Learning

Abstract

The present invention discloses an intelligent recognition method for farmland weeds based on evolutionary PU learning, including: 1: Collecting labeled and unlabeled crop image data, extracting and normalizing the attribute features of plants, and constructing a data set; 2: Constructing a label prediction model for crop image data samples; 3: Constructing a main population and an auxiliary population for the training set based on a binary competition strategy and setting their states; 4: Iteratively optimizing and improving the model performance by an evolutionary multi-task optimization method; 5: Gradually updating the population state during the evolution process; 6: Finally, selecting the optimal individual from the population as the final classifier for screening out weeds from the target crop images. The present invention overcomes the limitations of unlabeled samples for weed recognition, thereby improving the accuracy and efficiency of weed recognition to achieve a more scientific and efficient intelligent spraying operation for farmland.

Description

Technical Field

[0001] The present invention relates to the technical field of statistical learning classification, and particularly to an intelligent recognition method for farmland weeds based on evolutionary PU learning. Background Art

[0002] In agricultural production, farmland spraying is an important link in preventing and controlling pests and diseases and increasing crop yields. However, traditional spraying methods have many deficiencies. The manual targeted spraying method relies on the subjective judgment and experience of agricultural producers, which is time-consuming and laborious, and is prone to the risk of pesticide poisoning. On the other hand, although the uniform large-area spraying by agricultural machinery and drones is time-saving and labor-saving, due to the uneven distribution of weeds in the farmland, quantitative spraying often leads to excessive waste of pesticides in sparse weed areas, polluting the environment, while in dense weed areas, it may be ineffective due to insufficient dosage, increasing the agricultural production cost and being unfavorable to the sustainable development of the agricultural ecology. At the same time, when traditional supervised learning methods process crop image data, they often rely on a large number of labeled samples, which brings difficulties in obtaining labels for weed recognition. Furthermore, the spraying technology cannot accurately distinguish weeds and crops, resulting in waste of pesticides and environmental pollution. At the same time, it may not be able to effectively control weeds due to insufficient dosage, increasing the agricultural production cost. This is both time-consuming and expensive in agricultural practice, limiting the accuracy and robustness of classification. Summary of the Invention

[0003] The present invention is to solve the above-mentioned deficiencies existing in the prior art, and proposes an intelligent recognition method for farmland weeds based on evolutionary PU learning, in order to improve the accuracy of farmland weed recognition by intelligently optimizing the PU learning process in combination with evolutionary algorithms, thereby laying a foundation for accurate spraying decision-making and realizing more scientific and efficient intelligent spraying operations in farmland.

[0004] To achieve the above invention object, the present invention adopts the following technical solutions:

[0005] The intelligent recognition method for farmland weeds based on evolutionary PU learning of the present invention is characterized in that it is carried out according to the following steps:

[0006] Step 1: Collect labeled and unlabeled crop image data, extract the plant attribute features in the crop image data, and then perform standard normalization processing to obtain a crop image sample set, denoted as , where represents the th crop image sample, is a -dimensional attribute feature vector, represents the th crop image sample in the th attribute feature; and The set of labeled crop image samples is denoted as , The label of , denotes the th labeled crop image sample, denotes The label of The set of unlabeled crop image samples in , denotes the th unlabeled crop image sample; and , , , = 1, indicating that is a crop category, = -1 indicates that is a weed category;

[0007] Step 2: Initialization;

[0008] Step 2.1: Define the current iteration number as , the maximum iteration number as , and initialize ;

[0009] Step 2.2: Use the Euclidean distance formula to calculate The labeled similarity measurement values between the attribute features of any two crop image samples in

[0010] are obtained, and a number of labeled similarity measurement values are input into a sub-classifier for training to obtain a trained labeled crop sub-classifier; The trained labeled crop sub-classifier is used as an original individual, and thus original individuals form the th generation of the original population where denotes the th generation of the original population and the

[0011] Step 2.3: Use the Euclidean distance formula to calculate The similarity measurement values between the attribute features of any one crop image sample in and the attribute features of any one crop image sample in

[0012] are obtained, and a number of similarity measurement values are input into another sub-classifier for training to obtain a trained crop sub-classifier; The trained crop sub-classifier is used as an auxiliary individual, and thus Auxiliary population of the ; where represents the auxiliary population of the th generation, and represents the

[0013]

[0014] Step 3: Solve the crop image prediction model using the evolutionary multi-task optimization algorithm; :

[0015] (1)

[0016] In Equation (1), TP is the number of image samples in which the class is correctly predicted, and FN is the number of image samples in which the class is incorrectly predicted;

[0017] Step 3.2: Establish the second objective function of the crop image prediction model using Equation (2) :

[0018] (2)

[0019] For Equation (2), and are the sets of image samples predicted as crop classes and weed classes in the unlabeled crop image sample set , respectively, and , represents the number of represents the number of and respectively represent the th weed-class predicted image sample and the th crop-class predicted image sample in represents the classification function, is the indicator function. If , then , otherwise ;

[0020] Step 3.3: Establish the objective function of the crop image prediction model using Equation (3) :

[0021] ​​

[0022] Step 3.4: Calculate the objective function values of each individual in the g - th generation of the original population and the auxiliary population ;

[0023] Step 3.5: Conduct mating pool selection for and respectively to obtain the g - th generation of the original parental population and the g - th generation of the auxiliary parental population ;

[0024] Step 3.6: Generate offspring:

[0025] Use the genetic operators in the NSGA - Ⅱ algorithm to process each individual in and respectively to generate the g - th generation of the offspring population and the g - th generation of the auxiliary offspring population ;

[0026] Step 3.7: Merge with , merge with , and then conduct environmental selection on the two merged populations respectively to obtain the g - th generation of the original population and the g - th generation of the auxiliary population ;

[0027] Step 3.8: Obtain population knowledge:

[0028] Step 3.8.1: If when, conduct non - dominated ranking on the g - th generation of the original population , select all individuals in the highest non - dominated rank, and then use the voting - based integration method to process all individuals in the highest non - dominated rank to obtain the g - th generation of the original knowledge guiding vector , otherwise, assign to , and return to Step 3.4; is the set iteration threshold;

[0029] Step 3.8.2: Process the g - th generation of the auxiliary population according to the process of Step 3.8.1 to obtain the g - th generation of the auxiliary knowledge guiding vector ;

[0030] Step 3.9: Bidirectional knowledge transfer mechanism between populations:

[0031] Step 3.9.1: When holds, transfer to all individuals in the highest non-dominated rank and all individuals in the lowest non-dominated rank of the g-th generation original population ; otherwise, assign +1 to , and return to Step 3.4; where represents another set iteration threshold;

[0032] Step 3.9.2: When holds, transfer to all individuals in the highest non-dominated rank of the -th generation auxiliary population , then generate several individuals using the hybrid update operator and replace some individuals in the remaining non-dominated higher ranks; otherwise, assign +1 to , and return to Step 3.4;

[0033] Step 4: Assign +1 to , and determine whether holds. If it holds, it means the final -th generation original population and -th generation auxiliary population are obtained, and execute Step 5; otherwise, assign +1 to , and return to Step 3.4 to execute sequentially;

[0034] Step 5: Select the individual corresponding to the optimal objective function value from as the crop image data prediction model, which is used to screen out weeds from the target plant image and spray pesticides on the weeds.

[0035] The characteristics of the intelligent farmland weed identification method based on evolutionary PU learning according to the present invention also lie in that the said Step 3.5 includes:

[0036] Step 3.5.1: According to the objective function values of each individual in , after non-dominated sorting and calculating the crowding distance for each individual in the -th generation original population using the mating pool selection method in the NSGA-II algorithm, then perform binary tournament selection until individuals are selected as the Replace the original parent ;

[0037] Step 3.5.2: Process the generation of the auxiliary population in accordance with the process of Step 3.5.1 to obtain the generation of the auxiliary parent .

[0038] The said Step 3.7 includes:

[0039] Step 3.7.1: Combine with to form the generation of the temporary original population , and calculate the objective function value of each individual in the generation of the temporary population according to Equations (1) and (2). Then, rank each individual in by non-dominated ranking based on the objective function value, and successively select all individuals in each non-dominated ranking level in descending order of the non-dominated ranking level and use them as the individuals in the generation of the original population until the number of individuals in the generation of the original population reaches N; if the number of individuals in the selected level is greater than the number of remaining individuals in the generation of the original population , calculate the crowding distance of all individuals in the selected level, and select the individual with the smallest crowding distance and add it to the generation of the original population ;

[0040] Step 3.7.2: Process the combined with to form the generation of the temporary auxiliary population in accordance with the process of Step 3.7.1 to obtain the generation of the auxiliary population .

[0041] The attribute feature values of the crop image samples include: color features, texture features, shape features, and spectral features.

[0042] Generate a number of individuals by using the hybrid update operator according to the following process:

[0043] According to , calculate the Euclidean distance formula for the crop image samples The similarity measurement values between the attribute features of all crop image samples, and then select several similarity measurement values according to the descending order of the similarity measurement values and input them into several individual sub-classifiers for training to obtain several trained crop sub-classifiers and use them as several generated individuals.

[0044] An electronic device of the present invention includes a memory and a processor, characterized in that the memory is used to store a program for supporting the processor to execute the intelligent recognition method for farmland weeds, and the processor is configured to execute the program stored in the memory.

[0045] A computer-readable storage medium of the present invention, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the intelligent recognition method for farmland weeds when run by a processor.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0047] 1. Different from the traditional supervised learning classification method, the present invention proposes an intelligent recognition method for farmland weeds based on evolutionary PU learning, breaking the excessive dependence of traditional supervised learning on a large number of labeled samples, and providing a more efficient solution for crop image classification by fully mining the information relationship between unlabeled samples and a small number of normally growing samples.

[0048] 2. In the process of training the parameters of the crop image data prediction model, the present invention directly optimizes the objective function and two objectives, which can well measure the overall performance of label uncertainty and class imbalance data processing, thereby overcoming the problem of dependence on data labels in traditional classification methods, making the trained crop prediction model better than the traditional prediction model, and thus improving the accuracy of farmland weed recognition and the accuracy of spraying decision-making.

[0049] 3. The present invention adopts a framework of evolutionary task optimization, which not only improves the speed of the population to find the optimal solution, saves computing resources, but also overcomes the limitations of traditional crop image technologies in dealing with a large number of unlabeled samples, successfully solves the deficiencies of traditional supervised learning methods in the utilization of crop image sample information, enables the crop image prediction model to be trained using a large number of unlabeled samples, improves the accuracy and robustness of image classification, and reduces the time-consuming and cost of data labels in agricultural practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0051] In this embodiment, an intelligent recognition method for farmland weeds based on evolutionary PU learning is proposed to overcome the over - reliance of traditional supervised learning methods on a large number of labeled samples. An evolutionary algorithm is introduced as a heuristic optimization algorithm to simulate the process of natural selection, thereby improving the accuracy of weed recognition in crop images. Specifically, as Figure 1 shown, the method includes the following steps:

[0052] Step 1: Training the prediction model for crop image data:

[0053] Step 1: Collect labeled and unlabeled crop image data, extract the plant attribute features in the crop image data, and then perform standard normalization processing to obtain a crop image sample set, denoted as , where represents the th crop image sample, is a -dimensional attribute feature vector, represents the th crop image sample in the th attribute feature; Denote the set of labeled crop image samples in as , the label of is denoted as , represents the th labeled crop image sample, represents the label of ; Denote the set of unlabeled crop image samples in as , represents the th unlabeled crop image sample; And , , = 1, indicating that the category of is a crop, = - 1 indicates that the category of is a weed;

[0054] Step 2: Initialization;

[0055] Step 2.1: Define the current iteration number as , the maximum iteration number as , and initialize ;

[0056] Step 2.2: Calculate The labeled similarity metric values between the attribute features of any two crop image samples are obtained, and a number of labeled similarity metric values are input into a sub-classifier for training to obtain a trained labeled crop sub-classifier;

[0057] The trained labeled crop sub-classifier is used as an original individual, and thus original individuals form the th generation of the original population , where represents the th generation of the original population and the th original individual.

[0058] Step 2.3: Use the Euclidean distance formula to calculate the similarity metric values between the attribute features of any one crop image sample in and the attribute features of any one crop image sample in to obtain a number of similarity metric values and input them into another sub-classifier for training to obtain a trained crop sub-classifier;

[0059] The trained crop sub-classifier is used as an auxiliary individual, and thus auxiliary individuals form the th generation of the auxiliary population ; where represents the th generation of the auxiliary population and the th auxiliary individual.

[0060] Step 3: Use the evolutionary multi-task optimization algorithm to solve the crop image prediction model;

[0061] Step 3.1: Use Equation (1) to establish the first objective function true positive rate of the crop image prediction model :

[0062] (1)

[0063] In Equation (1), TP is the number of image samples in that are correctly predicted for the category, and FN is the number of image samples in that are incorrectly predicted for the category;

[0064] Step 3.2: Use Equation (2) to establish the second objective function of the crop image prediction model :

[0065] (2)

[0066] For Equation (2), and is an image sample set of unlabeled crops in the image sample sets predicted as crop categories and the image sample sets predicted as weed categories, and , represents the number of represents the number of and respectively represent the th image sample predicted as a weed category and the th image sample predicted as a crop category in represents the classification function is the indicator function. If , then let . Otherwise, let

[0067] Step 3.3: Establish the objective function of the crop image prediction model using Equation (3) :[[]]END]]

[0068] (3)

[0069] Step 3.4: Calculate the objective function values of each individual in the th generation of the original population and the auxiliary population ;

[0070] Step 3.5: Mating pool selection:

[0071] Step 3.5.1: According to the objective function values of each individual in the th generation of the original population perform non-dominated sorting and calculate the crowding distance for each individual, and then perform binary tournament selection until individuals are selected as the th generation of the original parental generation

[0072] Step 3.5.2: Process the th generation of the auxiliary population in the same process as Step 3.5.1 to obtain the th generation of the auxiliary parental generation .

[0073] Step 3.6: Generate offspring:

[0074] Use the genetic operators in the NSGA-II algorithm to separately and each individual in is processed to correspondingly generate a g-th generation offspring population of size and an auxiliary offspring population of the -th generation ; ;

[0075] Step 3.7: Merge populations and environmental selection:

[0076] Step 3.7.1: Merge with to form a temporary original population of the -th generation, and calculate the objective function value of each individual in the temporary population of the -th generation according to Equations (1) and (2). Thus, non-dominated ranking is performed on each individual in based on the objective function value, and all individuals in each non-dominated ranking level are sequentially selected in descending order of the non-dominated ranking level and used as the individuals in the original population of the -th generation until the number of individuals in the original population of the -th generation reaches N; if the number of individuals in the selected level is greater than the number of remaining individuals in the original population of the -th generation , then calculate the crowding distance of all individuals in the selected level, and select the individual with the smallest crowding distance to add to the original population of the -th generation ; -th generation original population -th generation ;

[0077] Step 3.7.2: According to the process of Step 3.7.1, process the merged temporary auxiliary population of the -th generation formed by merging to obtain the auxiliary population of the -th generation ; .

[0078] Step 3.8: Obtain population knowledge:

[0079] Step 3.8.1: If when, perform non-dominated ranking on the original population of the -th generation , and select all individuals in the highest non-dominated level. Then, process all individuals in the highest non-dominated level using a voting-based integration method to obtain the original knowledge guiding vector of the -th generation , otherwise, assign to , and return to Step 3.4; is the set iteration threshold;

[0080] Step 3.8.2: Process the generation auxiliary population to obtain the generation auxiliary knowledge-guided vector .

[0081] Step 3.9: Two-way knowledge transfer mechanism between populations:

[0082] Step 3.9.1: When , transfer to all individuals in the highest non-dominated rank and all individuals in the lowest non-dominated rank of the g-th generation original population ; otherwise, assign + 1 to , and return to Step 3.4; where represents another set iteration threshold;

[0083] Step 3.9.2: When , transfer to all individuals in the highest non-dominated rank of the generation auxiliary population , and then generate several individuals using the hybrid update operator and replace some individuals in the remaining non-dominated higher ranks; otherwise, assign + 1 to , and return to Step 3.4.

[0084] Step 4: Assign + 1 to , and judge whether holds. If it holds, it means the final generation original population and generation auxiliary population are obtained, and execute Step 5; otherwise, assign + 1 to , and return to execute Step 3.4 in sequence;

[0085] Step 5: Select the individual corresponding to the optimal objective function value from as the crop image data prediction model, which is used to screen out weeds from the target plant image and spray pesticides on the weeds.

[0086] Step 2: Crop image data test sample prediction step, using the crop image data prediction model obtained in Step 1 to screen out weeds from the target plant image and spray pesticides on the weeds:

[0087] Step 2.1: Input the crop image data captured in real time by the camera as the sample to be detected, and extract the attribute features of the crop image data, denoted as , where represents the attribute features of the th plant in the sample to be detected with respect to the crop image data;

[0088] Step 2.2: After performing standard normalization processing on the attribute features of the sample to be detected, input them into the crop image data prediction model obtained in Step 1, and output the predicted labels to screen out weeds, and spray pesticides on the weeds.

[0089] In this embodiment, an electronic device includes a memory and a processor. The memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0090] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium. When the computer program is run by a processor, it executes the steps of the above method.

[0091] An example of a simulation dataset is used to illustrate the specific implementation method of the present invention and verify the effect of the method of the present invention.

[0092] 1). Prepare a standard dataset:

[0093] The present invention uses the Fertility dataset as the standard dataset to verify the effectiveness of the crop image data prediction model. The Fertility dataset is a supervised classification benchmark dataset widely used in imbalanced classification. In the Fertility dataset, there are 100 pieces of data, that is, 100 pieces of crop image data collected correspondingly. All samples are used to verify the performance of the present invention for crop image data classification. For these 100 training samples, they are divided into training samples and test sample data.

[0094] 2). Evaluation metrics:

[0095] The classification accuracy is used as the evaluation metric in this example to evaluate the classification performance of the present invention on different crop image data. The higher the accuracy value, the better the classification effect.

[0096] 3). Conduct experiments on the standard dataset:

[0097] To verify the effectiveness of the method proposed by the present invention, the classification algorithm (EMT-PU) of the present invention is modeled and predicted on the Fertility dataset together with the previous classical uPU algorithm and MOEA-PU algorithm, and the prediction results of the three are compared. The experimental results are shown in Table 1.

[0098]

[0099] As can be seen from Table 1 in the standard dataset Fertility, the EMT-PU method of the present invention, by comparing the performance of different algorithms in crop weed detection, found that it has the best performance in terms of classification accuracy, with a classification accuracy of 0.8604, which is more accurate than other methods. This indicates that the present invention has significant advantages in accurately identifying crop weeds, and is expected to bring innovative breakthroughs to the field of intelligent agricultural machinery, improving the accuracy of farmland weed identification and the accuracy of spraying decision-making.

Claims

1. An intelligent method for identifying farmland weeds based on evolutionary PU learning, characterized in that: The steps are as follows: Step 1: Collect labeled and unlabeled crop image data, extract plant attribute features from the crop image data, and perform standard normalization to obtain a crop image sample set, denoted as ,in, Indicates Sample crop images, is a dimensional attribute feature vector, Indicates Sample crop images Middle attribute characteristics; The labeled crop image sample set is denoted as , The label is recorded as , Indicates labeled crop image samples, express ; The unlabeled crop image sample set is denoted as , Indicates unlabeled crop image samples; and , , , =1, indicating The category is crops, =-1 means The category is weeds; Step 2: Initialization; Step 2.1: Define the current number of iterations as , the maximum number of iterations is , and initialize ; Step 2.2: Calculate using the Euclidean distance formula The labeled similarity measure between the attribute features of any two crop image samples in , thereby obtaining several labeled similarity measures and inputting them into a sub-classifier for training, thereby obtaining a trained labeled crop sub-classifier; The trained labeled crop sub-classifier is taken as an original individual, thus The original individuals form the Original population ,in, Indicates Original population Middle original individual; Step 2.3: Calculate using the Euclidean distance formula The attribute characteristics of any crop image sample in The similarity measure value between the attribute features of any crop image sample in the image is obtained, thereby obtaining several similarity measure values ​​and inputting them into another sub-classifier for training to obtain a trained crop sub-classifier; The trained crop sub-classifier is used as an auxiliary individual, so that The auxiliary individuals constitute Generation of assisted population ;in, Indicates Generation of assisted population Middle Auxiliary individuals; Step 3: Use the evolutionary multi-task optimization algorithm to solve the crop image prediction model; Step 3.1: Use formula (1) to establish the first objective function of the crop image prediction model: True positive rate : (1) In formula (1), TP is The number of image samples whose categories are correctly predicted in , FN is The number of image samples whose categories are incorrectly predicted in ; Step 3.2: Use equation (2) to establish the second objective function of the crop image prediction model : (2) Formula (2): and is a sample set of unlabeled crop images The image sample set predicted as the crop category and the image sample set predicted as the weed category in , express The number of Indicated quantity, and Respectively expressed in Middle The image samples predicted to be weeds and image samples predicted to be crop classes, represents the classification function, is an indicator function, if , then let , otherwise, let ; Step 3.3: Use equation (3) to establish the objective function of the crop image prediction model : (3) Step 3.4: Calculate the Original population and auxiliary population The objective function value of each individual in ; Step 3.5: and Perform mating pool selection respectively to obtain Original parent and Auxiliary parent ; Step 3.6: Generate offspring: The genetic operators in the NSGA-Ⅱ algorithm are used to and Each individual in is processed to produce a corresponding size The g-th generation offspring population and Auxiliary progeny population ; Step 3.7: and Merge and Merge, and then conduct environmental selection on the two merged populations respectively, and get the Original population and Generation of assisted population ; Step 3.8: Obtain population knowledge: Step 3.8.1: If When, yes Original population Sort the non-dominated levels and select all individuals in the highest non-dominated level. Then use the voting-based ensemble method to process all individuals in the highest non-dominated level to obtain the first Original knowledge guide vector Otherwise, Assign to , and return to step 3.4; is the set iteration threshold; Step 3.8.2: Follow the process in step 3.8.1 to Generation of assisted population Process it and get Generation of auxiliary knowledge guidance vector ; Step 3.9: Bidirectional knowledge transfer mechanism between populations: Step 3.9.1: When When Transfer to the original population of generation g Among all individuals in the highest non-dominated rank and all individuals in the lowest non-dominated rank; otherwise, +1 assigned to , and return to step 3.4; where, represents another set iteration threshold; Step 3.9.2: When When Transfer to Generation of assisted population All individuals in the highest non-dominated level of , and then use the mixed update operator to generate several individuals and replace some individuals in the remaining non-dominated levels; otherwise, +1 assigned to , and return to step 3.4; Step 4: +1 assigned to ,judge Is it true? If true, it means that the final Original population and Generation of assisted population , execute step 5; otherwise, +1 assigned to , and return to step 3.4 to execute sequentially; Step 5: From The individual corresponding to the optimal objective function value is selected as the crop image data prediction model, which is used to screen out weeds from the target plant image and spray pesticides on the weeds.

2. The method for intelligent identification of farmland weeds based on evolutionary PU learning according to claim 1 is characterized in that: The step 3.5 comprises: Step 3.5.1: According to The objective function value of each individual in the NSGA-Ⅱ algorithm is selected by the mating pool selection method. The original population of the first generation After each individual in the non-dominated sort is performed and the crowding distance is calculated, a binary tournament selection is performed until the Individual as the first Original parent ; Step 3.5.2: Follow the process in step 3.5.1 to Generation of assisted population Process it and get Auxiliary parent .

3. The method for intelligent identification of farmland weeds based on evolutionary PU learning according to claim 2 is characterized in that: The step 3.7 comprises: Step 3.7.1: and Merged to form Temporary original population , and calculate the first Temporary original population The objective function value of each individual in Each individual in the non-dominated ranking is sorted, and all individuals in each non-dominated ranking are selected in descending order of non-dominated ranking and used as the first Original population individuals in the Original population Until the number of individuals in the selected level reaches N; if the number of individuals in the selected level is greater than Original population The number of individuals remaining in the selected level is calculated, and the individual with the smallest crowding distance is selected to join the first level. Original population middle; Step 3.7.2: Follow the process of step 3.7.1, To The merger formed Temporary auxiliary population Process it and get Generation of assisted population .

4. The method for intelligent identification of farmland weeds based on evolutionary PU learning according to claim 1, characterized in that: The attribute feature values ​​of the crop image samples include: color features, texture features, shape features, and spectral features.

5. The method for intelligent identification of farmland weeds based on evolutionary PU learning according to claim 1, characterized in that: Use the mixed update operator to generate several individuals in the following process: according to , using the Euclidean distance formula to calculate the crop image samples The similarity measure values ​​between the attribute features of all crop image samples in the dataset are obtained, and then several similarity measures are selected according to the descending order of the similarity measure values ​​and input into several individual sub-classifiers for training, and several crop sub-classifiers after training are obtained as several generated individuals.

6. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the method for intelligent identification of farmland weeds as described in any one of claims 1 to 5, and the processor is configured to execute the program stored in the memory.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for intelligent identification of farmland weeds in any one of claims 1 to 5 are executed.

Images