Fertilizing method for plant growth

Abstract

The invention belongs to the technical field of plant fertilization, and particularly relates to a plant growth fertilization method which comprises the following steps: S1, regularly shooting a high-resolution image of a kapok planting base by an unmanned aerial vehicle; s2, preprocessing the image; s3, identifying the growth state of the kapok by using a computer vision technology; s4, calculating the fertilizing amount according to the identified growth state; s5, transmitting a calculation result to a fertilization device, and enabling the fertilization device to perform fixed-point precise fertilization on the kapok; the invention provides the kapok growth state recognition and precise fertilization method based on the computer vision technology and unmanned aerial vehicle monitoring, the fertilization operation of the kapok can be efficiently and precisely carried out, the growth quality and yield of the kapok are improved, and the method has wide application prospects and economic benefits.

Description

Technical Field

[0001] The invention belongs to the technical field of plant fertilization, and in particular relates to a plant growth fertilization method. Background Art

[0002] As the city flower of Guangzhou, Kapok not only has ornamental value, but also has important ecological and cultural significance. However, with the acceleration of urbanization and the impact of climate change, the growth environment and health of Kapok face many challenges. Traditional Kapok growth management methods mainly rely on manual inspections and fertilization, which have limitations such as high labor costs, inaccurate fertilization, low efficiency, and lack of data. In order to solve these problems, the development of modern agricultural technology has provided new solutions for the scientific management of Kapok. Using drone imaging technology and computer vision technology, real-time monitoring and precise management of the growth status of Kapok can be achieved. Summary of the invention

[0003] The purpose of the present invention is to provide a method for fertilizing plant growth to solve the problems existing in the background technology.

[0004] In order to achieve the above technical objectives, the technical solution adopted by the present invention is as follows:

[0005] A method for fertilizing plant growth, comprising the following steps:

[0006] S1: UAVs regularly take high-resolution images of kapok planting bases;

[0007] S2: preprocess the image;

[0008] S2 includes removing image noise from the collected image using a Wiener filtering method;

[0009] The S2 also includes enhancing the image contrast by a histogram equalization method.

[0010] The captured images are processed by computer vision algorithms to identify the growth status of kapok flowers, including the leaf color and flower size of kapok flowers. The Wiener filter denoising function is used. The specific formula is:

[0011]

[0012] in, represents the estimated image after processing in the domain, that is, the image after denoising and restoration. G(u, v) is the representation of the observed image in the domain, usually an image containing noise. S f (u, v) is the power spectral density of the original noise-free image, S η (u, v) is the noise power spectral density.

[0013] S3: Using computer vision technology to identify the growth status of kapok flowers;

[0014] The S3 includes using a deep learning algorithm to extract the growth characteristics of kapok flowers from the preprocessed image, and the deep learning algorithm includes a YOLOv5 model for extracting the size and quantity information of kapok flowers, and also includes a ResNet (residual network) for extracting the color changes and signs of diseases and insect pests of kapok flowers.

[0015] The S3 method for identifying the growth status of kapok flowers comprises:

[0016] S3.1: Dataset preparation: collect and record a large amount of image data of kapok flowers in different growth states, including excellent, good, average and poor states, represented by values ​​between 0 and 1;

[0017] S3.2: Feature extraction. Since the flowers in the image may be of different sizes, the YOLOv5 algorithm is used to predict outputs of different scales for each input image, and the output of each scale is designed to identify objects of different sizes;

[0018] S3.3: Substitute the input image into the core formula of ResNet to calculate the features:

[0019] y=F(x,{Wi})+x

[0020] Among them, x represents the input feature; F(x,{Wi}) represents the residual mapping, the specific formula is as follows, and y is the output feature;

[0021] F(x,{Wi})=ReLU(BN(W2*ReLU(BN(W1*x))))

[0022] Where x is the input feature vector, W1 and W2 are weight matrices, BN is the batch normalization operation, ReLU is the ReLU activation function, and F(x, Wi) is the residual mapping function.

[0023] S3.4: Input the extracted features into the GBRT machine learning model to obtain the predicted growth.

[0024] The specific steps of the GBRT machine learning model include:

[0025] S3.4.1: Feature selection: Select the size, number, color and texture of flowers as input features;

[0026] The flower size is calculated by the target detection algorithm to calculate the diameter and area of ​​the flower, and then measure the growth status of each flower; the number of flowers is determined by detecting the center points of all flowers in the image and counting them. The color features are analyzed using two color spaces, RGB and HSV. The distribution of the main colors and the changes in hue are extracted through the color histogram. These color information can reflect the health status and nutritional level of the plant. Texture features are extracted through the gray-level co-occurrence matrix (GLCM) and local binary patterns (LBP). These features can reveal the subtle structure of the flower surface and possible signs of pests and diseases. In addition, in order to avoid overfitting, all features are standardized, and the formulas for the gray-level co-occurrence matrix and local binary pattern are respectively

[0027] P(i,j|d,θ)=the mumber of times(i,j)occur with distance d anddirectionθ

[0028] Where i, j are grayscale values, representing the grayscale level of the pixel in the image, d is the distance between pixels, and θ is the direction between pixels.

[0029]

[0030] where g c is the gray value of the center pixel, g p is the grayscale value of the neighborhood pixel, P is the number of pixels in the neighborhood, s(x) represents the sign function, 2 p is the weight value.

[0031] S3.4.2: Model training: The GBRT model is trained on the labeled data set, and the model performance is gradually optimized by boosting the tree. The Bayesian algorithm is introduced to optimize the hyperparameters of the model, and 5-fold cross validation is implemented.

[0032] S3.4.3: Prediction: Input the new image features into the trained GBRT model and output a continuous value between 0 and 1 as the predicted kapok growth status index Q.

[0033] Using the coefficient of determination R 2 , we also introduced the mean square error (MSE) and mean absolute percentage error (MAPE) to further evaluate the generalization performance of the model. The specific calculation formulas of MSE, MAPE and R2 are as follows:

[0034]

[0035] Where n is the number of samples, y i is the actual value, is the predicted value, is the sample mean.

[0036] S4: Calculate the amount of fertilizer to be applied according to the identified growth status;

[0037] The calculation formula of the fertilizer application amount of S4 is:

[0038] W=B×F×λ

[0039] Where W is the actual amount of fertilizer applied; B is the base amount of fertilizer applied to each kapok tree: 50 ml of compound fertilizer per tree; and F is the adjustment coefficient calculated based on the growth status index;

[0040] Among them, the adjustment factor F is determined by the following function:

[0041] F = 1 + (0.5 × (Q-0.5))

[0042] Wherein, Q is the kapok growth state index predicted in step S3.

[0043] S5: The calculated results are transmitted to the fertilization device, so that the fertilization device can carry out targeted and precise fertilization on the kapok flowers.

[0044] The present invention provides a method for identifying the growth status of kapok flowers and applying precise fertilization based on computer vision technology and unmanned aerial vehicle monitoring, which can efficiently and accurately carry out fertilization operations on kapok flowers, improves the growth quality and yield of kapok flowers, and has broad application prospects and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The present invention is further illustrated by means of the non-limiting examples given in the accompanying drawings.

[0046] Figure 1 A schematic diagram of a plant growth fertilization method of the present invention Figure 1 ;

[0047] Figure 2 A schematic diagram of a plant growth fertilization method of the present invention Figure 2 . DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0049] Example 1: Figure 1-2 As shown, a plant growth fertilization method of the present invention comprises the following steps:

[0050] S1: Use drones to regularly collect plant images in the kapok planting area. Use drones to regularly take high-resolution images of the flower planting area.

[0051] Specifically include:

[0052] Drone configuration and deployment: Select a DJI Phantom drone, install a high-definition camera and multispectral sensor, preset shooting parameters, and configure the GPS positioning system and flight control software.

[0053] Flight route planning: Plan the flight route of the drone based on the distribution characteristics of the kapok planting areas to ensure full coverage of all kapok areas.

[0054] Scheduled Flights and Imaging: Set up a regular flight schedule (once a week) for the drone and fly along a predetermined route to acquire high-definition images and multispectral data.

[0055] Data storage and transmission: The image data taken by the drone is stored in a secure storage device and backed up and preliminarily processed.

[0056] S2: Preprocess the image. Use computer vision algorithms to process the captured image and identify the growth status of the kapok flower (such as leaf color, flower size, etc.). Specifically include:

[0057] Image preprocessing: resize the image, use image enhancement techniques to improve image quality, and apply Wiener filter to remove noise.

[0058] Preferably, the image is denoised using a Wiener filter with the following formula:

[0059]

[0060] in, Represents the estimated image after processing in the domain, that is, the image after denoising and restoration. G(u,v) is the representation of the observed image in the domain, usually an image containing noise. S f (u, v) is the power spectral density of the original noise-free image. is the noise power spectral density. S η (u, v) can effectively reduce noise while retaining image details through the Wiener filter, and reduce image blur caused by drone movement.

[0061] S3: Use computer vision technology to identify the growth status of Kapok flowers. Use the GBRT machine learning model to perform regression prediction on the extracted features, including:

[0062] Dataset preparation: Collect a large amount of image data of kapok flowers in different growth states, including excellent, good, average or poor. Each image in the dataset has a corresponding label, which represents the growth state index of the kapok flower (a value between 0 and 1, 0 represents a very good growth state, 1 represents a very poor growth state).

[0063] Feature extraction: YOLOv5 algorithm is used for object detection. For each input image, YOLOv5 predicts multiple outputs at different scales, each of which is designed to identify objects of different sizes, thereby achieving comprehensive detection of kapok trees, flowers, and leaves. ResNet can be used to extract deeper features of kapok flowers, such as color changes, signs of pests and diseases, etc. The core formula of ResNet is:

[0064] y=F(x,{Wi})+x

[0065] Where x represents the input feature; F(x,{Wi}) represents the residual mapping function, which is the result of parameter calculation of the convolution layer, activation function, etc.; y is the output feature; F(x,{Wi})+x is to add the input directly to the output of the main path through a shortcut connection or skip connection. This shortcut connection is usually an identity mapping. If the input and output dimensions are inconsistent, the dimensions are adjusted through 1x1 convolution;

[0066] F(x,{Wi})=ReLU(BN(W2*ReLU(BN(W1*x))))

[0067] Where x is the input feature vector, W1 and W2 are weight matrices, BN is the batch normalization operation, and ReLU is the ReLU activation function.

[0068] Feature selection: Select flower size, quantity, color, and texture as input features.

[0069] The flower size is calculated by the target detection algorithm to calculate the diameter and area of ​​the flower, and then measure the growth status of each flower; the number of flowers is determined by detecting the center points of all flowers in the image and counting them. The color features are analyzed using two color spaces, RGB and HSV. The distribution of the main colors and the changes in hue are extracted through the color histogram. These color information can reflect the health status and nutritional level of the plant. Texture features are extracted through the gray-level co-occurrence matrix (GLCM) and local binary patterns (LBP). These features can reveal the subtle structure of the flower surface and possible signs of pests and diseases. In addition, in order to avoid overfitting, all features are standardized, and the formulas for the gray-level co-occurrence matrix and local binary pattern are respectively

[0070] P(i,j|d,θ)=the number of times(i,j)occur with distance dand directionθ

[0071] Where i, j are grayscale values, representing the grayscale level of the pixel in the image, d is the distance between pixels, and θ is the direction between pixels.

[0072]

[0073] where g c is the gray value of the center pixel, g p is the grayscale value of the neighborhood pixel, P is the number of pixels in the neighborhood, s(x) represents the sign function, 2 p is the weight value.

[0074] Model training: The GBRT model is trained using a labeled dataset, and the model performance is gradually optimized by boosting the tree. The Bayesian algorithm is introduced to optimize the hyperparameters of the model. To ensure the stability of the experimental results, a 5-fold cross validation is implemented. Specifically, the dataset is randomly divided into 5 parts, one of which is used as the test set in turn, and the remaining 4 are used as training sets. This process is repeated 5 times, each time using a different combination of training sets. The performance of the model is calculated by R 2 The coefficient of determination is used to evaluate the model's accuracy in fitting the data. 2 The value of the coefficient is between 0 and 1. The closer the value is to 1, the stronger the model's ability to explain the data and the higher the quality of fit. The goal of this study is to use the R 2 The average of the scores is used to evaluate the predictive performance of the model.

[0075] Prediction: Input the features of the new image into the trained GBRT model and output a continuous value between 0 and 1 as the predicted kapok growth status index Q.

[0076] Preferably, in this example, in addition to using the determination coefficient R 2 , we also introduced mean square error (MSE) and mean absolute percentage error (MAPE) to further evaluate the generalization performance of the model. MSE calculates the average of the squares of the difference between the predicted value and the actual value, and is mainly used to detect the degree of deviation between the model prediction and the actual observed data. MAPE also has a step of comparing the predicted value with the original data. The lower these two indicators are, it usually indicates that the model has higher prediction accuracy and better fitting effect. MSE, MAPE and R 2 The specific calculation formula is as follows:

[0077]

[0078] Where n is the number of samples, y i is the actual value, is the predicted value, is the sample mean.

[0079] S4 calculates the precise amount of fertilizer to be applied based on the growth status of the kapok flower to ensure that the kapok flower has the best growth conditions. The calculation formula for the amount of fertilizer to be applied is:

[0080] W=B×F×λ

[0081] Where W is the actual amount of fertilizer applied; B is the baseline amount of fertilizer applied to each kapok tree, which is 50 ml of compound fertilizer per tree. F is the adjustment coefficient calculated based on the growth status index, and F ranges from 0.75 (low growth state, more fertilizer is required) to 1.25 (high growth state, less fertilizer is required). λ is the environmental adjustment factor, which is used to adjust the amount of fertilizer according to the current environmental conditions. For example, lower humidity or higher temperature may require increased nutrient supplementation, so λ can vary from 1.2 (unfavorable environment, more fertilizer is needed) to 0.8 (favorable environment, less fertilizer can be applied).

[0082] Among them, the adjustment factor F is determined by the following function:

[0083] F = 1 + (0.5 × (Q-0.5))

[0084] Specifically, Q is the kapok growth state index predicted in step S3.

[0085] S5: Carry out precise fertilization at designated locations according to the calculation results. Specifically including:

[0086] Robot positioning and navigation: Use GPS, IMU (inertial measurement unit) and odometer combination to achieve global positioning and navigation of the robot. Based on the geographic coordinates generated by drone image data, plan the robot's movement path to ensure that the robot can accurately reach the designated fertilization location.

[0087] Path planning: A* algorithm is used for path planning to avoid obstacles and optimize the path.

[0088] Real-time path adjustment: When the robot is moving, it uses ultrasonic sensors to detect obstacles ahead and make dynamic path adjustments.

[0089] Precise fertilization control: According to the fertilization amount and position output by the fertilization calculation module, the robot accurately controls the start and stop of the fertilization device and the amount of fertilization through its control module. The fertilization device is equipped with an electric valve and a precision flow meter to ensure that the fertilizer is applied according to the set amount.

[0090] The present invention provides a method for identifying the growth status of kapok flowers and applying precise fertilization based on computer vision technology and unmanned aerial vehicle monitoring, which can efficiently and accurately carry out fertilization operations on kapok flowers, improves the growth quality and yield of kapok flowers, and has broad application prospects and economic benefits.

[0091] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A method for fertilizing plant growth, characterized in that: The following steps are involved: S1: UAVs regularly take high-resolution images of kapok planting bases; S2: preprocess the image; S3: Using computer vision technology to identify the growth status of kapok flowers; S4: Calculate the amount of fertilizer to be applied according to the identified growth status; S5: The calculated results are transmitted to the fertilization device, so that the fertilization device can carry out targeted and precise fertilization on the kapok flowers.

2. A method for fertilizing plant growth according to claim 1, characterized in that: S2 includes removing image noise from the collected image using a Wiener filtering method; The S2 also includes enhancing the image contrast by a histogram equalization method.

3. A method for fertilizing plant growth according to claim 2, characterized in that: The captured images are processed by computer vision algorithms to identify the growth status of kapok flowers, including the leaf color and flower size of kapok flowers. The Wiener filter denoising function is used. The specific formula is: in, represents the estimated image after processing in the domain, that is, the image after denoising and restoration. G(u, v) is the representation of the observed image in the domain, usually an image containing noise. S f (u,v) is the power spectral density of the original noise-free image, S η (u,v) is the noise power spectral density.

4. A method for fertilizing plant growth according to claim 1, characterized in that: The S3 includes using a deep learning algorithm to extract the growth characteristics of kapok flowers from the preprocessed image, and the deep learning algorithm includes a YOLOv5 model for extracting the size and quantity information of kapok flowers, and also includes a ResNet (residual network) for extracting the color changes and signs of diseases and insect pests of kapok flowers.

5. A method for fertilizing plant growth according to claim 4, characterized in that: The S3 method for identifying the growth status of kapok flowers comprises: S3.1: Dataset preparation: collect and record a large amount of image data of kapok flowers in different growth states, including excellent, good, average and poor states, represented by values ​​between 0 and 1; S3.2: Feature extraction, using the YOLOv5 algorithm to predict outputs of different scales for each input image, and designing the output of each scale to identify objects of different sizes; S3.3: Substitute the input image into the core formula of ResNet to calculate the features: y=F(x,{Wi})+x Where x represents the input feature; F(x,{Wi}) represents the residual mapping, and y is the output feature; F(x,{Wi})=ReLU(BN(W2*ReLU(BN(W1*x)))) Where x is the input feature vector, W1 and W2 are weight matrices, Bn is the batch normalization operation, ReLU is the ReLU activation function, and F(x, Wi) is the residual mapping function; S3.4: Input the extracted features into the GBRT machine learning model to obtain the predicted growth.

6. A method for fertilizing plant growth according to claim 5, characterized in that: The specific steps of the GBRT machine learning model include: S3.4.1: Feature selection: Select the size, number, color and texture of flowers as input features; S3.4.2: Model training: The GBRT model is trained on the labeled data set, and the model performance is gradually optimized by boosting the tree. The Bayesian algorithm is introduced to optimize the hyperparameters of the model, and 5-fold cross validation is implemented. S3.4.3: Prediction: Input the new image features into the trained GBRT model and output a continuous value between 0 and 1 as the predicted kapok growth status index Q.

7. A method for fertilizing plant growth according to claim 6, characterized in that: Using the coefficient of determination R 2 , mean square error (MSE) and mean absolute percentage error (MAPE) to further evaluate the generalization performance of the model. The specific calculation formulas for MSE, MAPE and R2 are as follows: Where n is the number of samples, y i is the actual value, is the predicted value, is the sample mean.

8. A method for fertilizing plant growth according to claim 6, characterized in that: The calculation formula of the fertilizer application amount of S4 is: W=B×F×λ Where W is the actual amount of fertilizer applied; B is the base amount of fertilizer applied to each kapok tree: 50 ml of compound fertilizer per tree; and F is the adjustment coefficient calculated based on the growth status index; Among them, the adjustment factor F is determined by the following function: F = 1 + (0.5 × (Q-0.5)) Wherein, Q is the kapok growth state index predicted in step S3.

Images