Intelligent agriculture management method and system based on artificial intelligence and Internet of Things

Abstract

The invention discloses an intelligent agricultural management method and system based on artificial intelligence and Internet of Things, and the system comprises a crop information and environmental parameter obtaining module which is used for obtaining multispectral images and infrared thermal images of crops in different pollination states through an unmanned plane, and obtaining environmental parameters through a sensor; the crop disease type identification module is used for constructing a crop disease type identification model and judging the disease type of the crop of which the health state is disease; the crop pollination spectral feature extraction module is used for identifying spectral feature wavebands and extracting spectral feature data; the crop pollination state recognition module is used for constructing a crop pollination state recognition model and judging a crop pollination state; and the crop pollination execution module is used for carrying out supplementary pollination on the unsuccessfully pollinated crops by using an unmanned aerial vehicle pollination mode. According to the invention, crops with failed pollination can be rapidly identified in a large-scale planting area, supplementary pollination is carried out through the unmanned aerial vehicle, and the comprehensiveness and timeliness of pollination are ensured.

Description

Technical Field

[0001] The present invention relates to the field of agricultural management, and in particular to a smart agricultural management method and system based on artificial intelligence and the Internet of Things. Background Art

[0002] In recent years, as global agriculture evolves towards efficiency, sustainability, and intelligence, the precise assessment and management of crop pollination status has become increasingly important in agricultural management. Pollination status is directly related to crop yield and quality. Therefore, accurately identifying crop pollination status and effectively managing the pollination process, particularly in large-scale cropping areas, is a major challenge facing agricultural technology. However, in practical applications, traditional pollination status identification methods have significant limitations, primarily due to a lack of effective mechanisms to address the complex influences of multiple environmental factors and changes in crop physiological status. Traditional agricultural management relies on farmer experience and simple observation methods, typically considering only a single or limited number of factors, such as pollination time and flower appearance, when determining whether a crop has been successfully pollinated. Such methods are insufficient in the complex and ever-changing farmland environment. In modern agricultural production, the environmental conditions and physiological status of crops are dynamically changing. These changes can directly affect the spectral characteristics of crops, making the assessment of pollination status based on spectral data even more complex. For example, changes in environmental factors such as light, temperature, and humidity, as well as changes in crop water and nutritional status, can lead to variations in spectral reflectance. Due to these interfering factors, relying solely on spectral data to assess pollination status often leads to misjudgments, failing to accurately distinguish spectral changes caused by pollination success from other environmental or physiological factors. Furthermore, crop health and physiological changes also significantly influence spectral data. During the crop growth cycle, different growth stages, diseases, or nutrient deficiencies can cause significant changes in spectral characteristics. For example, when a crop is suffering from disease or water deficiency, the reflectance spectrum of its leaves will change, manifesting as an increase or decrease in reflectance in specific bands. These spectral changes caused by physiological changes can mimic the spectral signatures of pollination failure, making it difficult for traditional methods to accurately distinguish these contributing factors, which in turn affects pollination status assessment. Current technology still has shortcomings in handling these complex influences. While modern technology can collect multispectral and infrared data using drones and sensors, effectively distinguishing the true causes of spectral changes remains an unresolved challenge in data analysis and comprehensive assessment. In particular, when multiple factors are at play, a single spectral data analysis method struggles to provide reliable pollination status assessments. This limitation leads to inefficiency and uncertainty in crop pollination status management, impacting the ultimate yield and quality of the crop. To meet these challenges, there is an urgent need for an intelligent method that can comprehensively consider the influence of multiple factors to accurately determine the pollination status of crops. Summary of the Invention

[0003] To address the problems of the above-mentioned prior art, the present invention provides a smart agricultural management method and system based on artificial intelligence and the Internet of Things, aiming to provide a judgment on the pollination status of crops when multiple factors act together.

[0004] The first embodiment of the present invention provides a smart agricultural management method based on artificial intelligence and the Internet of Things, which mainly includes:

[0005] Use drones to obtain multispectral images and infrared thermal images of crops in different pollination states to determine the health status of crops, and use sensors to obtain environmental parameters;

[0006] Based on images of crops with different disease types, a crop disease type recognition model is constructed to determine the disease type of crops whose healthy state is diseased;

[0007] Based on the median filtering and SIFT alignment processing of the multispectral image, the eigenvalue decomposition of the covariance matrix is ​​used to identify the spectral feature bands and extract the spectral feature data;

[0008] Based on crop status data, environmental parameters, spectral feature data and pollination status, a crop pollination status recognition model is constructed to determine the crop pollination status;

[0009] The crop pollination status recognition model is used to determine the crop pollination status within the crop planting area, and drone pollination is used to supplement the pollination of crops that have failed to be pollinated.

[0010] Furthermore, the method of obtaining multispectral images and infrared thermal images of crops in different pollination states by drones to judge the health status of crops and obtaining environmental parameters by sensors includes:

[0011] Using drones equipped with multispectral imaging equipment, multispectral images of crops in different pollination states are obtained, including successful and failed pollination. Using drones equipped with infrared thermal imaging equipment, thermal imaging data of crops is obtained. Based on the infrared thermal imaging data of crops and the preset temperature range, abnormal temperature areas are identified to determine the health status of crops, including health and disease. The crop planting record database is used to obtain the variety information and growth time of crops in the current planting area. A sensor network is deployed in the farmland to monitor and record environmental parameters in real time, including light intensity, temperature and humidity.

[0012] Furthermore, the crop disease type recognition model is constructed based on the images of crops with different disease types to determine the disease type of crops whose health status is diseased, including:

[0013] Acquire images of crops with different disease types and label the images with the disease type; use convolutional neural networks to train models based on the images of crops with different disease types and build a crop disease type recognition model; use drone-mounted cameras to acquire images of crops that are diseased in a healthy state, use the crop disease type recognition model to determine the disease type of crops that are diseased in a healthy state, and identify the location of crops that are diseased in a healthy state; use drone-mounted drug spraying devices to apply pesticides to crops that are diseased in a healthy state based on the disease type and location of the crops.

[0014] Furthermore, the method of performing median filtering and SIFT alignment processing on the multispectral image, using eigenvalue decomposition of the covariance matrix, identifying spectral feature bands, and extracting spectral feature data includes:

[0015] Use a median filter to remove noise from the multispectral image, and use the SIFT feature point matching algorithm to detect and match key feature points in images of different bands, and align the images of different bands through affine transformation; integrate the reflectance data of each pixel point in different bands into a vector to form a multispectral data matrix. The size of the multispectral data matrix is ​​n×m, where n is the number of pixels and m is the number of bands. Each row of the multispectral data matrix represents the reflectance data of a pixel point in all bands; standardize the data of each band by subtracting the mean of the reflectance data of each band and dividing it by the standard deviation; according to the multispectral data matrix, use the standard formula Calculate the covariance matrix between each wavelength, where X k is the kth row of the multispectral data matrix, is the mean vector of the multispectral data matrix; use NumPy to perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors; select eigenvectors with eigenvalues ​​greater than the preset eigenvalues ​​as principal components, and form an eigenvector matrix with the eigenvectors corresponding to the selected principal components; multiply the multispectral data matrix and the selected eigenvector matrix to obtain the multispectral data matrix after dimensionality reduction as principal component data; identify the spectral feature bands whose absolute values ​​in each eigenvector are greater than the preset absolute value threshold, and determine the positions of the spectral feature bands; based on the identified spectral feature bands, extract the average reflectance of the spectral feature bands in the original multispectral image data, and use the average reflectance of the spectral feature bands and the positions of the spectral feature bands as spectral feature data.

[0016] Furthermore, the crop pollination status recognition model is constructed based on the crop status data, environmental parameters, spectral characteristic data and pollination status of the crop to determine the crop pollination status, including:

[0017] Crop status data, environmental parameters, spectral feature data, and pollination status labels were integrated into a comprehensive dataset, serving as the pollination dataset. Crop status data included crop health status, disease type, crop variety information, and crop growth time. The pollination dataset was checked for missing values ​​or outliers, and missing values ​​were filled with the mean or samples containing outliers were deleted. All numerical features were standardized. The pollination dataset was divided into a training set and a test set. A support vector machine algorithm was used to train a model based on the training set to construct a crop pollination status recognition model. The trained crop pollination status recognition model was used to predict the test set, and its accuracy, precision, recall, and F1 score were calculated to evaluate its performance. If the performance of the crop pollination status recognition model did not meet expectations, the model was optimized by adjusting penalty parameters, kernel function parameters, improving data features, and increasing the data size. The optimized crop pollination status recognition model was used to determine crop pollination status and identify successful or failed pollination.

[0018] Furthermore, the method of using the crop pollination status recognition model to determine the crop pollination status in the crop planting area and using drone pollination to supplement pollination of crops that have failed to be pollinated includes:

[0019] Based on the crop status data, environmental parameters, and spectral feature data of the crops, a crop pollination status recognition model is used to identify the pollination status of crops in the planting area, determine the crops that have failed to pollinate, and mark the locations of the crops that have failed to pollinate; a drone equipped with pollination equipment is used to supplement the pollination of the identified crops that have failed to pollinate; after the supplementary pollination by the drone, the multispectral image, infrared image, and crop image of the crop are acquired again by the drone, and the crop status data, environmental parameters, and spectral feature data of the crop are acquired in real time, and the crop pollination status is re-identified using the crop pollination status recognition model, and the crops that have failed to pollinate are supplemented with pollination again; if the number of supplementary pollinations is greater than the preset threshold, and the crop pollination status is still pollination failure, the location of the crop that has failed to pollinate is sent to the person in charge of the crop planting area, who is notified to perform manual intervention, determine the cause of the pollination failure, and perform artificial pollination.

[0020] The second embodiment of the present invention provides a smart agricultural management system based on artificial intelligence and the Internet of Things, which mainly includes:

[0021] A smart agricultural management system based on artificial intelligence and the Internet of Things, mainly including:

[0022] The crop information and environmental parameter acquisition module is used to obtain multispectral images and infrared thermal images of crops in different pollination states through drones to determine the health status of crops and obtain environmental parameters through sensors;

[0023] The crop disease type recognition module is used to build a crop disease type recognition model based on images of crops with different disease types, and determine the disease type of crops with diseased health status;

[0024] The crop pollination spectral feature extraction module is used to identify spectral feature bands and extract spectral feature data based on the median filtering and SIFT alignment processing of the multispectral image using the eigenvalue decomposition of the covariance matrix;

[0025] The crop pollination status recognition module is used to build a crop pollination status recognition model based on crop status data, environmental parameters, spectral feature data and pollination status to determine the crop pollination status;

[0026] The crop pollination execution module is used to use the crop pollination status recognition model to determine the crop pollination status in the crop planting area, and use drone pollination to supplement the pollination of crops that have failed to be pollinated.

[0027] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0028] The present invention provides a smart agricultural management method and system based on artificial intelligence and the Internet of Things. By acquiring multispectral and infrared images through drones, the present invention can accurately monitor the health of crops, promptly identify diseases, and improve the accuracy of crop management. Through comprehensive analysis of multiple environmental and crop physiological factors, different sources of spectral variation can be effectively distinguished, avoiding the misjudgment problems of traditional methods and improving the reliability of pollination status judgment. Furthermore, the present invention can quickly identify crops that have failed to pollinate in large-scale planting areas and perform supplemental pollination through drones, ensuring the comprehensiveness and timeliness of pollination, thereby significantly improving crop yield and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a flow chart of a smart agricultural management method based on artificial intelligence and the Internet of Things of the present invention;

[0030] Figure 2 Schematic diagram of a smart agricultural management method based on artificial intelligence and the Internet of Things of the present invention;

[0031] Figure 3 This is a schematic diagram of a smart agricultural management system based on artificial intelligence and the Internet of Things of the present invention; DETAILED DESCRIPTION

[0032] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] like Figure 1-2 In this embodiment, a smart agricultural management method based on artificial intelligence and the Internet of Things may specifically include:

[0034] Step S101: Use drones to obtain multispectral images and infrared thermal images of crops in different pollination states, determine the health status of the crops, and obtain environmental parameters through sensors.

[0035] Using drones equipped with multispectral imaging equipment, multispectral images of crops are captured under different pollination conditions, including successful and failed pollination. Drones equipped with infrared thermal imaging equipment also capture thermal imaging data of crops. Based on the crop infrared thermal imaging data and a preset temperature range, abnormal temperature areas are identified and the crop's health status, including healthy and diseased, is determined. A crop planting record database is used to obtain information on crop varieties and growth periods in the current planting area. A sensor network is deployed in farmland to monitor and record environmental parameters, including light intensity, temperature, and humidity, in real time.

[0036] For example, drones equipped with multispectral imaging equipment can be used to regularly fly and capture multispectral images of corn crops in farmland under both successful and failed pollination conditions. Drones equipped with infrared thermal imaging equipment can capture thermal imaging data of corn crops. Based on the infrared thermal imaging data and a preset temperature range of 34°C-35°C, abnormal temperature areas in the thermal image that fall outside the preset temperature range are identified to determine the health of the crops. Crops with temperatures within the preset temperature range are marked as healthy, while crops with abnormal temperatures are marked as diseased. A crop planting record database is used to obtain information on the current corn variety planted in the farmland and its growth time. Records indicate that corn has been planted in this area for 60 days. A sensor network deployed in the farmland monitors and records environmental parameters in real time, including an illumination of 8,000 lux, a temperature of 28°C, and a humidity of 60%.

[0037] Step S102 : constructing a crop disease type recognition model based on images of crops with different disease types, and determining the disease type of crops whose healthy state is diseased.

[0038] Images of crops with different disease types are acquired and labeled with the disease type. Based on these images of crops with different disease types, a convolutional neural network is used to train a model and build a crop disease type recognition model. Using drone-mounted cameras, images of healthy but diseased crops are acquired. The crop disease type recognition model is used to determine the disease type and location of these healthy crops. Based on the disease type and location, pesticides are applied to these healthy crops using a drone-mounted sprayer.

[0039] For example, images of corn crops infected with different diseases, including corn leaf spot, rust, and aphid infestation, were collected and annotated in detail. These annotated images were then used to train a convolutional neural network, building a model capable of identifying corn disease types. A high-resolution drone-mounted camera captured images of healthy corn crops in the field, both diseased and healthy. Using this crop disease identification model, the specific disease types of these diseased crops were accurately determined. For example, if the incidence of corn leaf spot reached 30% in one area, while the incidence of rust was 20% in another. The drone's positioning system precisely marked the coordinates of the diseased crops. Subsequently, based on the identified disease type and location, the drone's onboard sprayer precisely applied pesticides, spraying specific pesticides on crops infected with corn leaf spot and rust, respectively, to control the spread and damage of the diseases.

[0040] Step S103 : Based on the median filtering and SIFT alignment processing of the multispectral image, the eigenvalue decomposition of the covariance matrix is ​​used to identify the spectral feature bands and extract the spectral feature data.

[0041] Use a median filter to remove noise from the multispectral image, and use the SIFT feature point matching algorithm to detect and match key feature points in images of different bands, and align images of different bands through affine transformation. Integrate the reflectance data of each pixel point in different bands into a vector to form a multispectral data matrix. The size of the multispectral data matrix is ​​n×m, where n is the number of pixels and m is the number of bands. Each row of the multispectral data matrix represents the reflectance data of a pixel point in all bands. The data of each band is standardized by subtracting the mean of the reflectance data of each band and dividing it by the standard deviation. According to the multispectral data matrix, use the standard formula Calculate the covariance matrix between each wavelength, where X k is the kth row of the multispectral data matrix, is the mean vector of the multispectral data matrix. Use NumPy to perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors. Select eigenvectors with eigenvalues ​​greater than the preset eigenvalues ​​as principal components, and form an eigenvector matrix with the eigenvectors corresponding to the selected principal components. Multiply the multispectral data matrix and the selected eigenvector matrix to obtain the multispectral data matrix after dimensionality reduction as the principal component data. Identify the spectral feature bands whose absolute values ​​in each eigenvector are greater than the preset absolute value threshold, and determine the position of the spectral feature bands. Based on the identified spectral feature bands, extract the average reflectance of the spectral feature bands in the original multispectral image data, and use the average reflectance of the spectral feature bands and the position of the spectral feature bands as spectral feature data.

[0042] For example, multispectral images of crops in a certain farmland are collected by using a multispectral imaging device equipped on a drone. There is noise in the images obtained in different bands. After removing the noise from these images using a median filter, the SIFT feature point matching algorithm is applied to detect and match the key feature points in images of different bands, and the images of different bands are aligned by affine transformation. The reflectance data of each pixel point in different bands are integrated into a vector to form a multispectral data matrix of size 1000×10, where 1000 represents the number of pixels and 10 represents the number of bands. The data of each band is standardized by subtracting the mean from the reflectance data of each band and then dividing it by the standard deviation. Use the standard formula Calculate the covariance matrix of the multispectral data matrix, where X k is the kth row of the multispectral data matrix, and X is the mean vector of the multispectral data matrix. The covariance matrix is ​​then subjected to eigenvalue decomposition using NumPy to obtain eigenvalues ​​and eigenvectors. Eigenvectors with eigenvalues ​​greater than 1 are selected as principal components, and these eigenvectors are combined into an eigenvector matrix. The multispectral data matrix is ​​multiplied by the selected eigenvector matrix to obtain a reduced-dimensional multispectral data matrix, which serves as the principal component data. Spectral feature bands with an absolute value greater than 0.5 in each eigenvector are identified, and the locations of these spectral feature bands are determined. Finally, based on the identified spectral feature bands, the average reflectance of these bands is extracted from the original multispectral image data, and the average reflectance and position of the spectral feature bands are used as the spectral feature data. For example, spectral feature bands in bands 3, 5, and 7 are identified, and the average reflectances of these bands are calculated to be 0.45, 0.52, and 0.48, respectively. Their positions in the multispectral data matrix are determined to be the 3rd, 5th, and 7th columns.

[0043] Step S104: constructing a crop pollination status recognition model based on crop status data, environmental parameters, spectral feature data, and pollination status of the crop to determine the crop pollination status.

[0044] Crop status data, environmental parameters, spectral feature data, and pollination status labels are integrated into a comprehensive dataset, known as the pollination dataset. Crop status data includes crop health status, disease type, crop variety information, and crop growth time. The pollination dataset is checked for missing values ​​or outliers, and missing values ​​are filled with the mean or samples with outliers are deleted. All numerical features are normalized. The pollination dataset is divided into a training set and a test set. A support vector machine algorithm is used to train a model based on the training set to construct a crop pollination status recognition model. The trained crop pollination status recognition model is used to predict the test set. The model's accuracy, precision, recall, and F1 score are calculated to evaluate its performance. If the performance of the crop pollination status recognition model does not meet expectations, the model is optimized by adjusting penalty parameters, kernel function parameters, improving data features, and increasing the data size. The optimized crop pollination status recognition model is used to determine crop pollination status and identify successful or failed pollination.

[0045] For example, there is a large planting area where different varieties of corn are planted. Multispectral and infrared thermal imaging equipment carried by drones is used to regularly collect spectral characteristic data and environmental parameters of the crops. Specific crop status data is obtained from the crop planting record database, including crop health status, disease type, variety information and growth time. For example, the data obtained shows that the corn in a certain area is variety A, with a growth time of 90 days, health status is divided into healthy and diseased, disease type is leaf spot, and environmental parameters include light intensity of 5000 lux, temperature of 25 ℃, humidity 60%, and the average reflectances of the spectral feature bands 3, 5, and 7 were 0.45, 0.52, and 0.48, respectively. This data was integrated into a comprehensive pollination dataset, which includes spectral features, ambient temperature, humidity, illuminance, crop variety information, growing time, and pollination status labels, including pollination success or failure. During data preprocessing, some data contained missing or outliers. The missing values ​​for illuminance and temperature were imputed using the mean, and several samples with abnormally high humidity were deleted. To unify the data scale, all numerical features were normalized to a mean of 0 and a standard deviation of 1. The dataset was divided into training and test sets with an 80:20 ratio. A support vector machine algorithm was used to train the model on the training set to construct a crop pollination status recognition model. The initial penalty parameter C was set to 1, and the RBF kernel was used as the kernel function. After training, the crop pollination status recognition model was used to predict the test set to evaluate its performance. The performance evaluation results showed that the model achieved an accuracy of 85%, a precision of 82%, a recall of 80%, and an F1 score of 81%, which did not meet expectations. Optimizing the penalty parameter C and the RBF kernel parameter gamma through grid search—for example, setting C to 10 and gamma to 0.01—significantly improved model performance. Improving data features by adding more spectral bands with significant differences and more detailed environmental parameters enabled the model to more accurately capture changes in pollination status. Increasing the sample size, particularly during early growth and during periods of high disease prevalence, enhanced the model's generalization. The optimized model achieved significant improvements on the test set, reaching 92% accuracy, 90% precision, 88% recall, and 89% F1 score, meeting expectations. The optimized crop pollination status recognition model was used to analyze real-time data, successfully determining the pollination status of crops and effectively identifying areas of successful and failed pollination.

[0046] Step S105: Use the crop pollination status recognition model to determine the crop pollination status in the crop planting area, and use drone pollination to supplement the pollination of crops that have failed to be pollinated.

[0047] Based on crop status data, environmental parameters, and spectral signature data, a crop pollination status recognition model is used to identify the pollination status of crops within a planting area, identify crops that have failed to pollinate, and mark their locations. Re-pollination is then performed on identified crops using a drone-mounted pollination device. After re-pollination, the drone acquires multispectral, infrared, and crop images of the crops, along with real-time crop status data, environmental parameters, and spectral signature data. The crop pollination status recognition model is then used to re-identify the pollination status of crops, and re-pollination is performed on crops that have failed to pollinate. If the number of re-pollination attempts exceeds a preset threshold and the crop's pollination status remains failed, the location of the failed crop is sent to the person in charge of the crop planting area, who is notified to intervene manually to determine the cause of the pollination failure and perform artificial pollination.

[0048] For example, in a 500-mu cornfield, a multispectral imaging device carried by a drone was used to capture multispectral images of the corn under different pollination conditions. The resulting spectral signature data showed average reflectances of 0.45, 0.52, and 0.48 for the 3rd, 5th, and 7th bands, respectively. Environmental parameters collected by the sensor network included light intensity of 5000 lux, temperature of 25°C, and humidity of 60%. The crop status data included corn variety A, 90 days of growth, health status (healthy or diseased), leaf spot disease, and environmental parameters including light intensity of 5000 lux, temperature of 25°C, and humidity of 60%. A trained crop pollination status recognition model was used for analysis to identify crops with failed pollination. The 5% of areas where pollination failed were identified and marked, with their specific locations recorded. Using the drone-mounted pollination device, re-pollination was performed on these failed crops based on the marked locations. After pollination is complete, the drone will fly again to capture updated multispectral, infrared, and crop images, acquiring the latest crop status data, environmental parameters, and spectral signature data to reassess the effectiveness of the supplemental pollination. After supplemental pollination, the pollination failure rate detected again has dropped to 1%, and the drone will continue to perform further supplemental pollination on these crops. If the number of supplemental pollination attempts exceeds a preset threshold of three and the pollination status of these crops remains unsuccessful, the location of these failed crops will be automatically sent to the farm manager. Upon receiving this notification, the farm manager will conduct manual intervention at the designated location to investigate the specific cause of the failure. If insect infestation is found to be the cause of the pollination failure, manual pollination or other corrective measures will be implemented to ensure successful pollination of these crops.

[0049] like Figure 3 In this embodiment, a smart agricultural management system based on artificial intelligence and the Internet of Things may specifically include:

[0050] The crop information and environmental parameter acquisition module is used to obtain multispectral images and infrared thermal images of crops through the multispectral imaging equipment and infrared thermal imaging equipment carried by the drone, determine the spectral characteristics and thermal distribution of crops under different pollination states, judge the health status of crops, and obtain environmental parameters through sensors, including temperature, humidity, and light intensity.

[0051] The crop disease type recognition module is used to collect images of crops with different disease types, build and use a crop disease type recognition model. If the health status of the crop is identified as a disease through infrared thermal imaging, the crop disease type recognition model determines the specific disease type of the crop based on visual features.

[0052] The crop pollination spectral feature extraction module is used to preprocess multispectral images, remove noise from the images using a median filter, and align images of different bands using the SIFT feature point matching algorithm to ensure spatial alignment between images of each band. It then identifies the main spectral feature bands through eigenvalue decomposition of the covariance matrix and extracts key spectral feature data.

[0053] The crop pollination status recognition module is used to build a crop pollination status recognition model based on the crop status data, environmental parameters, spectral feature data and pollination status of the crop to determine whether the crop has been pollinated successfully.

[0054] The crop pollination execution module is used to use the crop pollination status recognition model to determine the crop pollination status within the crop planting area, and use drones equipped with pollination equipment to accurately fly to the target area according to the marked pollination failure locations, and supplement the pollination of crops that have failed to pollinate.

[0055] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the concept of this application. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A smart agricultural management method based on artificial intelligence and the Internet of Things, characterized in that: The method comprises: Use drones to obtain multispectral images and infrared thermal images of crops in different pollination states to determine the health status of crops, and use sensors to obtain environmental parameters; Based on images of crops with different disease types, a crop disease type recognition model is constructed to determine the disease type of crops whose healthy state is diseased; Based on the median filtering and SIFT alignment processing of the multispectral image, the eigenvalue decomposition of the covariance matrix is ​​used to identify the spectral feature bands and extract the spectral feature data; Based on crop status data, environmental parameters, spectral feature data and pollination status, a crop pollination status recognition model is constructed to determine the crop pollination status; The crop pollination status recognition model is used to determine the crop pollination status within the crop planting area, and drone pollination is used to supplement the pollination of crops that have failed to be pollinated.

2. The method according to claim 2, wherein: The method of obtaining multispectral images and infrared thermal images of crops in different pollination states by drones to judge the health status of crops and obtain environmental parameters through sensors includes: Using drones equipped with multispectral imaging equipment, multispectral images of crops in different pollination states are obtained, including successful and failed pollination. Using drones equipped with infrared thermal imaging equipment, thermal imaging data of crops is obtained. Based on the infrared thermal imaging data of crops and the preset temperature range, abnormal temperature areas are identified to determine the health status of crops, including health and disease. The crop planting record database is used to obtain the variety information and growth time of crops in the current planting area. A sensor network is deployed in the farmland to monitor and record environmental parameters in real time, including light intensity, temperature and humidity.

3. The method according to claim 2, wherein: The method of constructing a crop disease type recognition model based on images of crops with different disease types to determine the disease type of crops whose healthy state is diseased includes: Acquire images of crops with different disease types and label the images with the disease type; use convolutional neural networks to train models based on the images of crops with different disease types and build a crop disease type recognition model; use drone-mounted cameras to acquire images of crops that are diseased in a healthy state, use the crop disease type recognition model to determine the disease type of crops that are diseased in a healthy state, and identify the location of crops that are diseased in a healthy state; use drone-mounted drug spraying devices to apply pesticides to crops that are diseased in a healthy state based on the disease type and location of the crops.

4. The method according to claim 2, wherein: The method of performing median filtering and SIFT alignment processing on the multispectral image, using eigenvalue decomposition of the covariance matrix, identifying spectral feature bands, and extracting spectral feature data includes: Use a median filter to remove noise from the multispectral image, and use the SIFT feature point matching algorithm to detect and match key feature points in images of different bands, and align the images of different bands through affine transformation; integrate the reflectance data of each pixel point in different bands into a vector to form a multispectral data matrix. The size of the multispectral data matrix is ​​n×m, where n is the number of pixels and m is the number of bands. Each row of the multispectral data matrix represents the reflectance data of a pixel point in all bands; standardize the data of each band by subtracting the mean of the reflectance data of each band and dividing it by the standard deviation; according to the multispectral data matrix, use the standard formula Calculate the covariance matrix between each wavelength, where X k is the kth row of the multispectral data matrix, is the mean vector of the multispectral data matrix; use NumPy to perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors; select eigenvectors with eigenvalues ​​greater than the preset eigenvalues ​​as principal components, and form an eigenvector matrix with the eigenvectors corresponding to the selected principal components; multiply the multispectral data matrix and the selected eigenvector matrix to obtain the multispectral data matrix after dimensionality reduction as principal component data; identify the spectral feature bands whose absolute values ​​in each eigenvector are greater than the preset absolute value threshold, and determine the positions of the spectral feature bands; based on the identified spectral feature bands, extract the average reflectance of the spectral feature bands in the original multispectral image data, and use the average reflectance of the spectral feature bands and the positions of the spectral feature bands as spectral feature data.

5. The method according to claim 2, wherein: The method of constructing a crop pollination status recognition model based on crop status data, environmental parameters, spectral feature data, and pollination status to determine the crop pollination status includes: The crop status data, environmental parameters, spectral feature data and pollination status labels of crops are integrated into a comprehensive dataset as the pollination dataset. The crop status data includes crop health status, disease type, crop variety information and crop growth time. The pollination dataset is checked for missing values ​​or outliers, and the missing values ​​are filled with the mean or samples containing outliers are deleted, and all numerical features are standardized. The pollination dataset is divided into a training set and a test set. Based on the training set, the support vector machine algorithm is used for model training to construct a crop pollination status recognition model. The trained crop pollination status recognition model is used to predict the test set, and the accuracy, precision, recall rate and F1 score of the crop pollination status recognition model are calculated to evaluate the performance of the crop pollination status recognition model. If the performance of the crop pollination status recognition model does not meet expectations, the crop pollination status recognition model is optimized by adjusting the penalty parameter, kernel function parameter, improving data features, and increasing the data volume. The optimized crop pollination status recognition model is used to judge the crop pollination status and identify whether the crop pollination is successful or failed.

6. The method according to claim 2, wherein: The method of using a crop pollination status recognition model to determine the crop pollination status within a crop planting area and using drone pollination to supplement pollination of crops that have failed to be pollinated includes: Based on the crop status data, environmental parameters, and spectral feature data of the crops, a crop pollination status recognition model is used to identify the pollination status of crops in the planting area, determine the crops that have failed to pollinate, and mark the locations of the crops that have failed to pollinate; a drone equipped with pollination equipment is used to supplement the pollination of the identified crops that have failed to pollinate; after the supplementary pollination by the drone, the multispectral image, infrared image, and crop image of the crop are acquired again by the drone, and the crop status data, environmental parameters, and spectral feature data of the crop are acquired in real time, and the crop pollination status is re-identified using the crop pollination status recognition model, and the crops that have failed to pollinate are supplemented with pollination again; if the number of supplementary pollinations is greater than the preset threshold, and the crop pollination status is still pollination failure, the location of the crop that has failed to pollinate is sent to the person in charge of the crop planting area, who is notified to perform manual intervention, determine the cause of the pollination failure, and perform artificial pollination.

7. A smart agricultural management system based on artificial intelligence and the Internet of Things, characterized by: The system comprises: The crop information and environmental parameter acquisition module is used to obtain multispectral images and infrared thermal images of crops in different pollination states through drones to determine the health status of crops and obtain environmental parameters through sensors; The crop disease type recognition module is used to build a crop disease type recognition model based on images of crops with different disease types, and determine the disease type of crops with diseased health status; The crop pollination spectral feature extraction module is used to identify spectral feature bands and extract spectral feature data based on the median filtering and SIFT alignment processing of the multispectral image using the eigenvalue decomposition of the covariance matrix; The crop pollination status recognition module is used to build a crop pollination status recognition model based on crop status data, environmental parameters, spectral feature data and pollination status to determine the crop pollination status; The crop pollination execution module is used to use the crop pollination status recognition model to determine the crop pollination status in the crop planting area, and use drone pollination to supplement the pollination of crops that have failed to be pollinated.

Images