Paddy rice fertilizer deficiency real-time monitoring and precise fertilization system and method based on image recognition

Through image recognition technology and drone multi-spectral cameras, the nutritional element status of rice fields is monitored in real time, and the lag and subjective problems of rice fertilization management are solved, precise fertilization is achieved, and agricultural production efficiency and environmental protection are improved.

CN118941946APending Publication Date: 2024-11-12HEILONGJIANG AGRI INVESTMENT BIG DATA TECH CO LTD
View PDF 0 Cites 9 Cited by

Patent Information

Application Number
CN202410941157.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing rice fertilization management methods are subjective and lagging, and it is impossible to accurately determine whether rice lacks fertilizer in real time, resulting in blind fertilization or missed the best fertilization opportunity, affecting yield and increasing environmental pollution.

Method used

Image recognition technology is adopted to collect image data through a drone equipped with a multi-spectral camera, perform image preprocessing, feature extraction and model training, and to build an identification model that predicts the degree of fertilizer deficiency in rice to achieve real-time monitoring and precise fertilization.

Benefits of technology

It has achieved high-precision identification of rice nutrient deficiencies from macro to micro, reducing blind fertilization and resource waste, reducing agricultural costs, reducing negative environmental impacts, and improving the accuracy and foresight of agricultural management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941946A_ABST
    Figure CN118941946A_ABST
Patent Text Reader

Abstract

The invention discloses a rice fertilizer deficiency real-time monitoring and precise fertilization system and method based on image recognition, and particularly relates to the field of monitoring, and the system comprises a data acquisition module, an image data preprocessing module, a feature extraction module, a recognition model module for constructing and predicting the rice fertilizer deficiency degree, a model training and verification refining module, and a real-time monitoring module. According to the invention, image data are collected and transmitted to the image preprocessing module, and then radiation, geometric and atmospheric correction is carried out; then, a feature extraction module extracts vegetation indexes and spectral features from the processed data for use in constructing a rice fertilizer deficiency degree prediction model; the model construction module is responsible for sample collection, actual measurement and model design, and the training and verification refining module is used for data set division, image annotation, preprocessing and model training; and finally, the real-time monitoring module photographs the paddy field through a multispectral camera carried by the unmanned aerial vehicle, and transmits data to a control end in real time, thereby realizing real-time monitoring of the growth state of the paddy rice.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of monitoring technology, and more specifically, to a system and method for real-time monitoring of rice fertilizer deficiency and precise fertilization based on image recognition. Background Art

[0002] As one of the most important food crops in the world, the yield and quality of rice directly affect food security and social and economic stability. During the growth and development of rice, the rational application of fertilizers is crucial to its yield and quality. However, due to the influence of various factors such as differences in soil fertility, climatic conditions, and farming management, field rice may often have varying degrees of nutrient deficiency or excess. Existing rice fertilization management mainly relies on farmers' experience and conventional soil testing. This method is subjective and lagging, and cannot accurately judge whether rice is really lacking in fertilizer in real time. It is easy to lead to blind fertilization or miss the best time for fertilization, which in turn affects yield and increases environmental pollution.

[0003] Traditional rice fertilizer deficiency diagnosis technologies mainly include laboratory soil analysis methods and on-site diagnosis methods based on plant physiological indicators. The laboratory soil analysis method obtains the content of major nutrients such as nitrogen, phosphorus, and potassium through chemical testing of soil samples, and then guides fertilization in combination with crop nutrient absorption rules and soil fertility evaluation standards. However, this method is time-consuming and labor-intensive, and cannot reflect the actual nutrient needs of crops in real time. Another existing implementation scheme is visual assessment and image analysis technology based on plant appearance characteristics. For example, aerial remote sensing monitoring is carried out using drones equipped with multispectral cameras, and the changes in parameters such as leaf color index and leaf area index of rice plants under different nutrient levels are compared to indirectly judge whether they are deficient in fertilizer.

[0004] However, there are still some shortcomings in its actual use. For example, the above two traditional methods cannot realize the real-time, accurate and non-destructive diagnosis of fertilizer deficiency of rice at the individual or group level. Therefore, an image recognition based real-time monitoring and precise fertilization system and method for rice fertilizer deficiency came into being. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a system and method for real-time monitoring of rice fertilizer deficiency and precise fertilization based on image recognition to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] Data acquisition module: used to acquire image data using acquisition equipment and transmit the acquired image data to the image preprocessing module;

[0008] Image data preprocessing module: used to perform radiation correction, geometric correction and atmospheric correction preprocessing steps on the original image;

[0009] Feature extraction module: used to extract vegetation index and spectral features from the processed data;

[0010] Constructing a recognition model module for predicting the degree of rice fertilizer deficiency: It is used to construct a recognition model module for predicting the degree of rice fertilizer deficiency, including sample collection, field measurement and model design;

[0011] Model training and verification refinement module: used to divide the data set, annotate and preprocess the images, and train the model;

[0012] Real-time monitoring module: It is used to use the multi-spectral camera carried by the drone to continuously conduct periodic and all-round aerial photography of the rice fields, and transmit the data to the control end in real time.

[0013] Preferably, in the data acquisition module, drones equipped with multispectral or hyperspectral cameras are used to conduct periodic, all-round, multi-angle aerial photography of rice fields to obtain high-quality image data containing multiple time stages, multiple viewing angles and multiple spectral information; these cameras can capture electromagnetic waves of different bands, some of which are closely related to the chlorophyll content, water status and biochemical properties of important nutrients such as nitrogen, phosphorus and potassium in rice leaves.

[0014] Preferably, in the image data preprocessing module, the steps in data processing are to offset the interference of factors such as solar radiation, terrain undulation, atmospheric scattering, etc. on the spectral signal to ensure the consistency and accuracy of the data.

[0015] Preferably, in the feature extraction module, the vegetation index includes a normalized vegetation index and a vegetation restoration index;

[0016] The specific method for extracting spectral features is as follows:

[0017] Spectral feature extraction: By decomposing the hyperspectral data into multiple narrow band spectral curves, the spectral features of each band are extracted;

[0018] Feature selection: Spectral analysis is used to identify the areas most sensitive to changes in nitrogen, phosphorus, and potassium content by characteristic wavelengths and spectral intervals, and feature selection is performed using machine learning algorithms;

[0019] Model building: Based on the selected characteristic wavelengths and spectral ranges, a model is built to predict changes in nitrogen, phosphorus, and potassium content;

[0020] Model validation: Verify the established model, evaluate its prediction accuracy and stability, and ensure the reliability and applicability of the model.

[0021] Preferably, the identification model module for predicting the degree of rice fertilizer deficiency includes a sample collection and measurement unit and a model design unit;

[0022] Sample collection and measurement unit: Field sampling, including collecting rice field images taken by drones and corresponding rice plant samples in the plots, and obtaining their actual nitrogen, phosphorus and potassium content as standard values ​​through laboratory measurement;

[0023] Model design unit: Using the support vector machine method, a recognition model for predicting the degree of rice fertilizer deficiency is constructed based on the spectral features in drone images and the nutrient content data measured in the laboratory.

[0024] Preferably, in the model training and verification refinement module, the sample data set is divided into a training set and a test set, 80% of the data is used for training, and 20% of the data is used for testing, to verify the generalization performance of the model on unknown data.

[0025] Preferably, in order to solve the problem that there is a huge difference between the number of healthy rice and fertilizer-deficient rice in the sample categories, the traditional cross entropy loss function easily leads to the model's preference to predict the majority class; Focal Loss introduces the modulation factor γ and the category weight ε to assign lower weights to easy-to-classify samples and higher weights to difficult-to-classify samples, so as to balance the learning weights of samples in each category.

[0026] Preferably, in the real-time monitoring module, the multispectral camera carried by the drone continuously performs periodic and all-round aerial photography of the rice fields to obtain high-definition image data containing multiple growth stages, different perspectives and rich spectral information. These multispectral or hyperspectral image data can reveal the biochemical characteristics related to the chlorophyll content, water status and the content of key nutrients such as nitrogen, phosphorus and potassium in rice leaves; the trained rice fertilizer deficiency recognition model is used to quickly process and deeply analyze the latest image data collected by the drone in real time; the model makes a comprehensive judgment on the spectral characteristics and vegetation index information in the image, and estimates the distribution status and degree of deficiency of nitrogen, phosphorus and potassium nutrients in each area of ​​the entire rice field in real time.

[0027] Technical effects and advantages of the present invention:

[0028] 1. The present invention calculates the vegetation index and other related spectral characteristic parameters by performing in-depth analysis on the multispectral images taken by drones. These parameters can highly sensitively reflect the deficiency of essential nutrients such as nitrogen, phosphorus, and potassium in crops such as rice, achieving high-precision identification from macro to micro, overcoming the limitations and lags of visual judgment or conventional chemical testing methods, and facilitating early detection and early intervention of crop nutritional imbalance problems.

[0029] 2. Based on the precise monitoring data obtained by the present invention, an intelligent fertilization prescription map generation algorithm is designed, which can guide the drone to implement precise fertilization operations according to the actual nutritional needs of crops. It can not only avoid environmental damage and waste of resources caused by blind fertilization, but also prevent the impact of insufficient fertilization on crop yield and quality, and realize on-demand fertilization and precise regulation.

[0030] 3. The present invention significantly reduces agricultural production costs by reducing unnecessary manual field work and avoiding additional losses caused by inaccurate fertilization. At the same time, this precise fertilization method helps reduce the loss of fertilizers into water bodies and soil, mitigates the negative impact of agricultural activities on the environment, and is in line with the development strategy of green and sustainable agriculture.

[0031] 4. The present invention constructs a precision agriculture decision support system that integrates multi-source data, machine learning and big data analysis technologies, which can provide agricultural producers with scientific and reasonable fertilization strategy recommendations, enhance the accuracy and predictability of agricultural management, meet the needs of refined and intelligent management of modern agriculture, and thus promote the entire agricultural industry chain to develop in an efficient and environmentally friendly direction.

[0032] 5. The present invention integrates UAV multispectral remote sensing technology to cover a large area of ​​farmland in a short time, capture and continuously monitor the growth status and nutritional status of crops in real time, greatly improve the monitoring efficiency, and solve the problem that traditional methods cannot grasp the overall nutritional dynamics of farmland in real time and comprehensively. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of system module connection of the present invention.

[0034] Figure 2 It is a schematic diagram of the method flow of the present invention.

[0035] Figure 3 Schematic diagram of the labelImg data annotation tool of the present invention.

[0036] Figure 4 A schematic diagram of an XML file for storing annotation information of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] See also Figure 1As shown, the present invention relates to an image recognition system for real-time monitoring and precise fertilization of rice fertilizer deficiency, which includes a data acquisition module, an image data preprocessing module, a feature extraction module, a recognition model module for building a prediction module for the degree of rice fertilizer deficiency, a model training and verification refinement module, and a real-time monitoring module.

[0039] The data acquisition module is connected to the image data preprocessing module, the image data preprocessing module is connected to the feature extraction module, the feature extraction module is connected to the recognition model module for constructing a prediction of the degree of rice fertilizer deficiency, the recognition model module for constructing a prediction of the degree of rice fertilizer deficiency is connected to the model training and verification refinement module, and the model training and verification refinement module is connected to the real-time monitoring module.

[0040] The data acquisition module is used to acquire image data using an acquisition device and transmit the acquired image data to the image preprocessing module;

[0041] In the data acquisition module, drones equipped with multispectral or hyperspectral cameras are used to conduct periodic, all-round, and multi-angle aerial photography of rice fields to obtain high-quality image data containing multiple time stages, multiple perspectives, and multiple spectral information; these cameras can capture electromagnetic waves in different bands, some of which are closely related to the chlorophyll content, water conditions, and biochemical properties of important nutrients such as nitrogen, phosphorus, and potassium in rice leaves.

[0042] The image data preprocessing module is used to perform radiation correction, geometric correction and atmospheric correction preprocessing steps on the original image;

[0043] In the image data preprocessing module, the steps in data processing are to offset the interference of solar radiation, terrain undulation, atmospheric scattering and other factors on the spectral signal to ensure the consistency and accuracy of the data;

[0044] Radiation correction refers to converting the radiation energy in the original image into radiometry or reflectivity to eliminate the radiation effects of solar radiation and the earth's surface itself. By correcting the response of the sensor, images from different sensors and at different times can be made comparable, providing a consistent benchmark for subsequent data analysis;

[0045] Geometric correction refers to correcting the geometric form of the image, eliminating the influence of terrain undulations and sensor tracks, so that the position and shape of the objects in the image are consistent with the actual scene; by resampling, removing distortion and registering the image, the geometric changes in the image can be corrected to improve the accuracy and geometric precision of the data;

[0046] Atmospheric correction refers to eliminating the influence of atmospheric scattering on the image spectrum and extracting the true reflectivity of the ground object. The goal of atmospheric correction is to remove the atmospheric scattering components in the original image and reveal the spectral characteristics of the ground object itself. By using atmospheric remote sensing models and estimates of atmospheric parameters, the image can be atmospherically corrected to improve the credibility and application value of remote sensing data.

[0047] The feature extraction module is used to extract vegetation index and spectral features from the processed data;

[0048] In the feature extraction module, the vegetation index includes a normalized vegetation index and a vegetation restoration index;

[0049] The calculation method of normalized difference vegetation index is as follows:

[0050] Among them, G NDVI It is expressed as the normalized vegetation index, N is the reflectance value of the near-infrared band, and R is the reflectance value of the red light band;

[0051] The calculation method of vegetation restoration index is as follows:

[0052] Among them, G PRI It is expressed as the vegetation restoration index, L is expressed as the reflectance of the green band, and B is expressed as the reflectance of the blue band;

[0053] The specific method for extracting spectral features is as follows:

[0054] Spectral feature extraction: By decomposing the hyperspectral data into multiple narrow band spectral curves, the spectral features of each band are extracted;

[0055] Feature selection: Spectral analysis is used to identify the areas most sensitive to changes in nitrogen, phosphorus, and potassium content by characteristic wavelengths and spectral intervals, and feature selection is performed using machine learning algorithms;

[0056] Model building: Based on the selected characteristic wavelengths and spectral ranges, a model is built to predict changes in nitrogen, phosphorus, and potassium content;

[0057] Model validation: Verify the established model, evaluate its prediction accuracy and stability, and ensure the reliability and applicability of the model.

[0058] Through the above steps, the sensitive characteristics of changes in nitrogen, phosphorus and potassium contents in hyperspectral data are analyzed and utilized.

[0059] The method for constructing a recognition model module for predicting the degree of rice fertilizer deficiency is used to construct a recognition model module for predicting the degree of rice fertilizer deficiency, including sample collection, actual measurement and model design;

[0060] The identification model module for predicting the degree of rice fertilizer deficiency includes a sample collection and measurement unit and a model design unit;

[0061] Sample collection and measurement unit: Field sampling, including collecting rice field images taken by drones and corresponding rice plant samples in the plots, and obtaining their actual nitrogen, phosphorus and potassium content as standard values ​​through laboratory measurement;

[0062] Model design unit: Using the support vector machine method, a recognition model for predicting the degree of rice fertilizer deficiency is constructed based on the spectral features in drone images and the nutrient content data measured in the laboratory.

[0063] The model training and verification refinement module is used to divide the data set, annotate and preprocess the images, and train the model;

[0064] In the model training and verification refinement module, the sample data set is divided into a training set and a test set, 80% of the data is used for training and 20% of the data is used for testing to verify the generalization performance of the model on unknown data;

[0065] Use the labelImg image annotation tool to finely annotate rice field images, accurately delineate the bounding boxes of rice plants, and store the annotation information as an XML file to ensure that the target rice can be identified based on this location information during model training;

[0066] Standardization processing is performed on the annotated field rice images. The standardization processing includes image scaling, cropping, rotation, enhancement and weakening operations to ensure the uniformity of image size and meet the input requirements of the deep learning network model. It also improves the standardization and normalization level of the image, which is conducive to the model's efficient data processing.

[0067] Targetedly strengthen the rice feature expression within the marked frame, while weakening similar background features inside and outside the frame to ensure that the features of the target object are prominent; in the weakening process, ensure that the features outside the frame are significantly weaker than those inside the frame to improve the accuracy of model recognition;

[0068] In order to make up for the possible limitations of image samples under the actual growth state of rice in the field, image data augmentation technology is used to generate a large number of new training samples through geometric transformation, color gamut transformation, clarity transformation, noise injection and local erasure to enhance the adaptability and robustness of the model to various complex situations.

[0069] Among them, geometric transformation includes rotation, mirroring, translation, cropping, scaling, and distortion; color gamut transformation refers to simulating brightness changes under different lighting conditions; clarity transformation refers to adjusting image clarity using kernel filtering technology; noise injection refers to simulating various imaging quality problems; local erasing refers to simulating occlusion effects;

[0070] The present invention adopts two types of loss functions to process classification tasks and positioning tasks respectively; wherein the two types of loss functions include classification task loss function and positioning task loss function;

[0071] In view of the large difference between the number of healthy rice and fertilized rice in the sample category, the traditional cross entropy loss function easily leads to the model's preference for predicting the majority class. Focal Loss introduces the modulation factor γ and the class weight ε to assign lower weights to easy-to-classify samples and higher weights to difficult-to-classify samples, so as to balance the learning weights of samples in each category. The specific calculation method of Focal Loss is as follows:

[0072] F L =-ε t (1-P t ) γ log(P t ), where F L Expressed as the classification task loss function, P t is the classification probability of the model for the sample, ε t is the class weight of the sample, and γ is an adjustable parameter used to adjust the loss value of easy-to-classify samples;

[0073] When samples are easy to classify, P t Close to 1,1-P t Close to 0, at this time, due to the existence of the exponential term, the value of Focal Loss will be greatly reduced, thereby reducing the penalty for easy-to-classify samples. For difficult-to-classify samples, due to P t Close to 0,1-P t When it is close to 1, the value of the exponential term will be close to 1. At this time, the value of Focal Loss will not decrease, but will remain high, thereby increasing the penalty for difficult-to-classify samples.

[0074] In summary, Focal Loss introduces an adjustable parameter γ to reduce the loss value of easily classified samples and increase the penalty for difficult-to-classify samples, thereby paying more attention to difficult-to-classify samples and improving the performance of the model under class imbalance.

[0075] The method of positioning task loss function is as follows:

[0076] In the task of rice plant bounding box positioning, the traditional mean square error cannot effectively handle the relative position relationship between bounding boxes, so an improved version of the IoU loss function is used;

[0077] On the basis of IoU, a penalty term for the distance between the center point of the bounding box is added to encourage the model to reduce the distance between the center point of the predicted box and the true box. The specific calculation method is:

[0078] Among them, p(b, b gt ) is the Euclidean distance between the center points of the predicted box and the real box, c is the diagonal length of the minimum enclosing box that can cover both boxes at the same time, and IoU is the intersection over union ratio;

[0079] The specific calculation method of IoU is:

[0080] Among them, IoU is expressed as the intersection-over-union ratio, A∩B is expressed as the intersection area of ​​the predicted box and the real box, and A∪B is expressed as the union area of ​​the predicted box and the real box;

[0081] The bounding box aspect ratio constraint is further added to the DIoU loss function to encourage the model to better maintain the consistency of the bounding box shape during the optimization process; the calculation method of CIoU is as follows:

[0082] Among them, p(b, b gt ) is the Euclidean distance between the center points of the predicted box and the real box, c is the diagonal length of the minimum enclosing box that can cover both boxes at the same time, IoU is the intersection over union ratio, v is the measure of aspect ratio inconsistency, and α is the parameter that measures the similarity of the aspect ratios of the predicted box and the real box;

[0083] GIoU not only focuses on the overlapping area of ​​the bounding box, but also considers the non-overlapping area, so that the model can more comprehensively optimize the bounding box position and size during training. The calculation method of GIoU is as follows:

[0084] Among them, A∪B represents the union area of ​​the predicted box and the real box, IoU represents the intersection over union ratio, and C represents the minimum closed area formed by the combination of the predicted box and the real box;

[0085] In the actual training process, the weights of different loss functions are adjusted according to the actual situation to seek the best balance of overall model performance. In addition, through hyperparameter tuning, cross-validation, and grid search methods, the optimal hyperparameter combination is found to maximize model performance; at the same time, the model performance during the training process is closely monitored. If there are signs of overfitting, the model complexity, optimization strategy, or early stopping strategy should be adjusted in time to avoid overfitting; finally, after the model achieves ideal performance on the training set, it is verified using an independent test set. By comparing the model's prediction results on the test set with the actual nitrogen, phosphorus, and potassium content, the model's prediction accuracy, practicality, and stability are quantitatively evaluated.

[0086] The real-time monitoring module is used to continuously and periodically and comprehensively photograph the rice fields using a multi-spectral camera carried by the drone, and transmit the data to the control terminal in real time;

[0087] In the real-time monitoring module, the multispectral camera carried by the drone continuously conducts periodic and all-round aerial photography of the rice fields to obtain high-definition image data containing multiple growth stages, different perspectives and rich spectral information. These multispectral or hyperspectral image data can reveal the biochemical characteristics related to the chlorophyll content, water status and the content of key nutrients such as nitrogen, phosphorus and potassium in rice leaves; using the trained rice fertilizer deficiency recognition model, the latest image data collected by the drone in real time is quickly processed and deeply analyzed; the model makes a comprehensive judgment on the spectral characteristics and vegetation index information in the image, and estimates the distribution status and deficiency degree of nitrogen, phosphorus and potassium nutrients in each area of ​​the entire rice field in real time;

[0088] Based on the nutrient distribution and fertilizer deficiency data of rice fields obtained through real-time monitoring, the system automatically generates targeted precision fertilization recommendations and prescription maps; based on the differences in nutrient element requirements of different plots, differentiated and refined fertilization plans are formulated, such as guiding farmers to apply corresponding fertilizers according to the specific requirements of nitrogen, phosphorus and potassium; farmers can implement fixed-point and quantitative fertilization operations based on this precise fertilization plan, reduce ineffective fertilization and excessive fertilization, and achieve fertilization on demand, thereby effectively improving agricultural production efficiency and reducing production costs, while also helping to improve soil quality and promote the dual increase in rice yield and quality.

[0089] See also Figure 2 As shown, in this embodiment, it should be specifically explained that the present invention provides a method for real-time monitoring of rice fertilizer deficiency and precise fertilization by image recognition, comprising the following steps:

[0090] Step S1: using an acquisition device to acquire image data;

[0091] Step S2: Perform radiation correction, geometric correction and atmospheric correction preprocessing steps on the original image;

[0092] Step S3: extracting vegetation index and spectral features from the processed data;

[0093] Step S4: constructing a recognition model module for predicting the degree of rice fertilizer deficiency, including sample collection, field measurement and model design;

[0094] Step S5: Divide the data set, annotate and preprocess the images, and train the model;

[0095] Step S6: Use the multispectral camera carried by the drone to continuously and periodically and omnidirectionally take aerial photos of the rice fields, and transmit the data to the control terminal in real time.

[0096] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A real-time monitoring and precise fertilization system for rice fertilizer deficiency based on image recognition. It is characterized in that include: Data acquisition module: used to acquire image data using acquisition equipment and transmit the acquired image data to the image preprocessing module; Image data preprocessing module: used to perform radiation correction, geometric correction and atmospheric correction preprocessing steps on the original image; Feature extraction module: used to extract vegetation index and spectral features from the processed data; Constructing a recognition model module for predicting the degree of rice fertilizer deficiency: It is used to construct a recognition model module for predicting the degree of rice fertilizer deficiency, including sample collection, field measurement and model design; Model training and verification refinement module: used to divide the data set, annotate and preprocess the images, and train the model; Real-time monitoring module: It is used to use the multi-spectral camera carried by the drone to continuously conduct periodic and all-round aerial photography of the rice fields, and transmit the data to the control end in real time.

2. According to claim 1, a rice fertilizer deficiency real-time monitoring and precise fertilization system based on image recognition, Features: In the data acquisition module, drones equipped with multispectral or hyperspectral cameras are used to conduct periodic, all-round, and multi-angle aerial photography of rice fields to obtain high-quality image data containing multiple time stages, multiple perspectives, and multiple spectral information; these cameras can capture electromagnetic waves in different bands, some of which are closely related to the chlorophyll content, water conditions, and biochemical properties of important nutrients such as nitrogen, phosphorus, and potassium in rice leaves.

3. According to the image recognition system for real-time monitoring and precise fertilization of rice fertilizer deficiency in claim 1, Features: In the image data preprocessing module, the steps in data processing are to offset the interference of solar radiation, terrain undulation, atmospheric scattering and other factors on the spectral signal to ensure the consistency and accuracy of the data.

4. According to claim 1, a rice fertilizer deficiency real-time monitoring and precise fertilization system based on image recognition, Features: In the feature extraction module, the vegetation index includes a normalized vegetation index and a vegetation restoration index; The specific method for extracting spectral features is as follows: Spectral feature extraction: By decomposing the hyperspectral data into multiple narrow band spectral curves, the spectral features of each band are extracted; feature Selection: Spectroscopic analysis of characteristic wavelengths and spectral intervals that are most sensitive to changes in nitrogen, phosphorus, and potassium content, and feature selection using machine learning algorithms; Model building: Based on the selected characteristic wavelengths and spectral ranges, a model is built to predict changes in nitrogen, phosphorus, and potassium content; Model validation: Verify the established model, evaluate its prediction accuracy and stability, and ensure the reliability and applicability of the model.

5. According to claim 1, a rice fertilizer deficiency real-time monitoring and precise fertilization system based on image recognition, Features: The identification model module for predicting the degree of rice fertilizer deficiency includes a sample collection and measurement unit and a model design unit; Sample collection and measurement unit: Field sampling, including collecting rice field images taken by drones and corresponding rice plant samples in the plots, and obtaining their actual nitrogen, phosphorus and potassium content as standard values ​​through laboratory measurement; Model design unit: Using the support vector machine method, a recognition model for predicting the degree of rice fertilizer deficiency is constructed based on the spectral features in drone images and the nutrient content data measured in the laboratory.

6. The image recognition system for real-time monitoring and precise fertilization of rice fertilizer deficiency according to claim 1, Features: In the model training and verification refinement module, the sample data set is divided into a training set and a test set, 80% of the data is used for training and 20% of the data is used for testing to verify the generalization performance of the model on unknown data; Use the labelImg image annotation tool to finely annotate rice field images, accurately delineate the bounding boxes of rice plants, and store the annotation information as an XML file to ensure that the target rice can be identified based on this location information during model training; Standardization processing is performed on the annotated field rice images. The standardization processing includes image scaling, cropping, rotation, enhancement and weakening operations to ensure the uniformity of image size and meet the input requirements of the deep learning network model. It also improves the standardization and normalization level of the image, which is conducive to the model's efficient data processing. Targetedly strengthen the rice feature expression within the marked frame, while weakening similar background features inside and outside the frame to ensure that the features of the target object are prominent; in the weakening process, ensure that the features outside the frame are significantly weaker than those inside the frame to improve the accuracy of model recognition; In order to make up for the possible limitations of image samples under the actual growth status of rice in the field, image data augmentation technology is used to generate a large number of new training samples through geometric transformation, color gamut transformation, clarity transformation, noise injection and local erasing, so as to enhance the adaptability and robustness of the model to various complex situations.

7. The image recognition system for real-time monitoring and precise fertilization of rice fertilizer deficiency according to claim 6, Features: In view of the large difference between the number of healthy rice and fertilized rice in the sample category, the traditional cross entropy loss function easily leads to the model's preference for predicting the majority class; FocalLoss introduces the modulation factor γ and the category weight ε to assign lower weights to easy-to-classify samples and higher weights to difficult-to-classify samples to balance the learning weights of samples in each category. The specific calculation method of FocalLoss is as follows: F L =-ε t (1-P t ) γ log(P t ), where F L Expressed as the classification task loss function, P t is the classification probability of the model for the sample, ε t is the category weight of the sample, and γ is an adjustable parameter used to adjust the loss value of easy-to-classify samples.

8. The image recognition system for real-time monitoring and precise fertilization of rice fertilizer deficiency according to claim 1, Features: In the real-time monitoring module, the multispectral camera carried by the drone continuously conducts periodic and all-round aerial photography of the rice fields to obtain high-definition image data containing multiple growth stages, different perspectives and rich spectral information. These multispectral or hyperspectral image data can reveal the biochemical characteristics related to the chlorophyll content, water status and the content of key nutrients such as nitrogen, phosphorus and potassium in rice leaves; using the trained rice fertilizer deficiency recognition model, the latest image data collected by the drone in real time is quickly processed and deeply analyzed; the model makes a comprehensive judgment on the spectral characteristics and vegetation index information in the image, and estimates the distribution status and degree of deficiency of nitrogen, phosphorus and potassium nutrients in each area of ​​the entire rice field in real time.

9. The method for real-time monitoring and precise fertilization of rice fertilizer deficiency by image recognition according to claim 1, using the system for real-time monitoring and precise fertilization of rice fertilizer deficiency by image recognition according to any one of claims 1 to 8, Features: Step S1: using an acquisition device to acquire image data; Step S2: Perform radiation correction, geometric correction and atmospheric correction preprocessing steps on the original image; Step S3: Extract vegetation index and spectral features from the processed data Step S4: constructing a recognition model module for predicting the degree of rice fertilizer deficiency, including sample collection, field measurement and model design; Step S5: Divide the data set, annotate and preprocess the images, and train the model; Step S6: Use the multispectral camera carried by the drone to continuously and periodically and omnidirectionally take aerial photos of the rice fields, and transmit the data to the control terminal in real time.

Citation Information

Cited By

  • Plant vigor intelligent monitoring system and method based on multi-dimensional perception compound eye imaging

    CN119198720A

  • Intelligent plant vitality monitoring system and method based on multi-dimensional perception compound eye imaging

    CN119198720B

  • Spectrum-based rice ammonia phosphorus and potassium deficiency and yield estimation method and system

    CN120450161A

  • A spectral-based method and system for predicting the deficiency of ammonia, phosphorus, and potassium in rice and its yield.

    CN120450161B

  • Accurate, timely and lossless cyperus esculentus fertilizer supplementing method for weak fertilizer conservation soil

    CN120959024A