A high-standard farmland big data intelligent management platform and method based on the Internet of Things
Through the Internet of Things technology combined with GPS, U-Net network, ant colony algorithm and SVM classifier, a high-standard farmland big data intelligent management platform is built, which solves the problem of inefficiency in high-standard farmland management and realizes efficient and accurate farmland management .
Patent Information
- Application Number
- CN202410742895.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-11
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing high-standard farmland management technology lacks informatization and efficiency, resulting in inefficient management and inability to effectively utilize the Internet of Things and big data technology.
Adopting a high-standard farmland big data intelligent management platform based on the Internet of Things, through GPS positioning, U-Net network, ant colony algorithm, bat algorithm and SVM classifier, field path identification, irrigation system laying, drone fertilization and Functions such as pest identification and inspection personnel monitoring improve management efficiency and accuracy.
It has achieved efficient management of high-standard farmland areas, improved the accuracy of field path identification, irrigation water resource utilization, drone fertilization cost savings, pest identification accuracy and patrol quality, and improved overall management efficiency and accuracy.
Smart Images

Figure CN118313800B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of management or supervision systems, and specifically to a high-standard farmland big data intelligent management platform and method based on the Internet of Things. Background Art
[0002] China has a large task area for the transformation and upgrading of high-standard farmland, and the quality of farmland construction needs to be further improved, especially in the management of high-standard farmland, which lacks information and efficient technologies.
[0003] The Chinese invention patent application with publication number CN117078214A discloses a data management method for high-standard farmland construction projects, but it does not use modern technologies such as the Internet of Things and big data, and the relevant information of high-standard farmland cannot be uploaded to the management platform, and the information processing is backward and inefficient. Summary of the invention
[0004] 1. Technical issues to be resolved
[0005] In view of the shortcomings of the prior art, the present invention provides a high-standard farmland big data intelligent management platform based on the Internet of Things, which has the advantages of high efficiency and real-time information update, and solves the problem of low efficiency in high-standard farmland management.
[0006] (II) Technical solution
[0007] The present invention is implemented by the following technical scheme: a high-standard farmland big data intelligent management method based on the Internet of Things, the method comprising the following steps:
[0008] S1. Select high-standard farmland, use GPS positioning to obtain the GPS coordinates of each vertex of the high-standard farmland, and connect the GPS coordinates in sequence to obtain the high-standard farmland area;
[0009] S2, identifying the field roads in the high-standard farmland area, and marking them in the high-standard farmland area, to obtain the high-standard farmland area divided by the field roads;
[0010] S3, laying irrigation systems and drainage devices in the high-standard farmland areas divided by the field paths to obtain high-standard farmland areas after preliminary treatment;
[0011] S4, planning a drone fertilization route, and performing drone fertilization on the high-standard farmland area after the preliminary treatment to obtain a treated high-standard farmland area;
[0012] S5. In the treated high-standard farmland area, feature extraction and SVM classifier are used to identify pests, and after locating the pest area, a drone carrying pesticide is used to spray the pesticide in the pest area to obtain a final high-standard farmland area;
[0013] S6. The inspectors inspect the final high-standard farmland area and use the spatial index R-tree to monitor whether the inspectors cross the boundary.
[0014] Preferably, the S1 comprises the following steps:
[0015] S11, along the edge of the high-standard farmland, using the GPS to collect the coordinates of each vertex of the high-standard farmland to obtain the coordinates of each vertex, and obtain a coordinate set;
[0016] S12. Connect the coordinates in the coordinate set to obtain a high-standard farmland area.
[0017] Preferably, S2 comprises the following steps:
[0018] S21. In the U-Net network, the input feature map is set. There are 4 downsampling modules in the encoding stage. Each downsampling module contains 2 3×3 convolutional layers with 0 padded pixels and length 1 and 1 maximum pooling layer with 2 padded pixels. After the feature map is processed by 1 downsampling module, the feature map size becomes ,in is the number of image rows, is the number of image columns; the feature map is processed by four downsampling modules in sequence and then by two 3×3 convolutional layers with 0 padded pixels and a stride of 1 before entering the decoding stage;
[0019] S22, there are 4 upsampling modules in the decoding stage, each of which contains 2 3×3 deconvolution layers with 0 padded pixels and a stride of 1, 1 jump connection and 2 3×3 deconvolution layers with a stride of 1; after being processed by 1 upsampling module, the deconvolution layer is first used to double the size of the feature map, the number of channels is halved, and then it is spliced with the feature map generated in the encoding stage through a jump connection; before the jump connection, the feature map size is cropped, and the spliced feature map is convolved twice, and the feature map size becomes ; After 4 upsampling modules, the semantic segmentation model is finally obtained;
[0020] S23, perform semantic segmentation training on the semantic segmentation model, set represents the regressed pixel value, Indicates location aisle The pixel score on Represents the number of classes and outputs the regression pixel value in the feature map. The calculation formula is as follows:
[0021]
[0022] Setting the energy function Represents the cross entropy loss function for regressing pixel values, Indicates location The weight of Indicates location The label value of Indicates location The softmax loss function is Represents parameters; sets a training error threshold. When the training error is less than the training error threshold, stops the semantic segmentation training and obtains the semantic segmentation training model. The calculation formula is as follows:
[0023]
[0024] S24, after the semantic segmentation model training is completed, a semantic segmentation training model is obtained, a high-standard farmland area image is obtained from the high-standard farmland area, the high-standard farmland area image is input into the semantic segmentation training model to obtain the regression probability of each pixel, a critical threshold is set, and when the regression probability is greater than the critical threshold, the regression probability is retained, otherwise the regression probability is discarded; the retained regression probability is fitted into a corresponding mask area, the mask area is binarized by a scanning method to obtain edge information points, the edge information points are weighted summed to obtain the road center position, and the field path in the high-standard farmland is identified after fitting the road center position;
[0025] S25. Mark the field paths in the high-standard farmland area to obtain the high-standard farmland area divided by the field paths.
[0026] The invention realizes semantic segmentation training through the U-Net network, uses the semantic segmentation training model to determine the pixels of field paths in high-standard farmland area images, and fits the center position of roads that meet the requirements, thereby identifying field paths; uses a neural network to process images, and the obtained recognition results are highly accurate, which is convenient for subsequent operations.
[0027] Preferably, S3 comprises the following steps:
[0028] S31. Laying an irrigation system in the high-standard farmland area. The specific steps are as follows:
[0029] S311, Setting Moment Ants from the location Climb to the next position , initially releases pheromones and inspirational information , express The probability of transition at a certain moment, It is a pheromone-inducing factor. is the expectation heuristic factor, Indicates location Climb to the next position The distance express The moment expectation heuristic function is calculated as follows:
[0030]
[0031]
[0032] S312, ants crawl to the location , release pheromones in the path passed, pheromones evaporate over time, and ants will release new pheromones; set the maximum number of iterations in the crawling process, when the number of iterations is greater than the maximum number of iterations, stop the iteration, the formula is as follows,
[0033]
[0034]
[0035]
[0036] in, represents the pheromone volatility coefficient, and , Indicates that the ant is from position Climb to the next position The optimal amount of pheromone released, Indicates the intensity of pheromone release. represents the optimal path length for ants to crawl, express Release pheromones all the time. ;
[0037] S313, the optimal path length of the ants obtained after the iteration , set the first threshold ,when When It is the optimal route length of the ditch irrigation road in one of the high-standard farmland areas divided by field roads;
[0038] S314, repeat S311 and S312, and record the optimal route lengths of all ditch irrigation roads in the high-standard farmland area divided by field paths as a data set , Indicates the number of high-standard farmland areas divided by field paths;
[0039] S315. Use buried pipelines to supply water to high-standard farmland, and use furrow irrigation for irrigation. The net flow calculation formula for the furrow irrigation channel in one of the high-standard farmland areas divided by field paths is as follows:
[0040]
[0041] in, represents the net flow of the ditch irrigation channel, is the irrigation area of the ditch irrigation channel, Indicates the watering modulus;
[0042] S316, set the loss coefficient of the ditch irrigation channel to The gross flow calculation formula of the ditch irrigation channel in one of the high-standard farmland areas divided by field paths is as follows:
[0043]
[0044] in, It represents the gross flow of the ditch irrigation channel. Indicates The optimal route length of the irrigation channel in the high-standard farmland area divided by field paths, , ;
[0045] S317, the optimal route degree of all ditch irrigation roads in the high-standard farmland area divided by the field path is recorded as a data set , the calculation formula of high-standard farmland flow is as follows,
[0046]
[0047] in, It indicates the total flow required for high-standard farmland;
[0048] S32. Drainage devices are set up in the high-standard farmland areas divided by field paths. The specific steps are as follows:
[0049] S321. Use the empirical method to calculate the effective rainfall, and set the effective rainfall to , the rainfall is , the calculation formula is as follows,
[0050]
[0051] in, It indicates the effective utilization coefficient of rainfall. When the rainfall is less than 5mm, the effective coefficient of rainfall is 0; when the rainfall is 5mm-50mm, the effective coefficient of rainfall is 1.0-0.8; when the rainfall is greater than 50mm, the effective coefficient of rainfall is 0.8-0.7;
[0052] S322. Use the comprehensive method to calculate evaporation and transpiration. The calculation formula is as follows:
[0053]
[0054] in, represents evapotranspiration, represents the net radiation, Indicates the heat transfer of soil. It indicates the daily average humidity at a height of 2 meters. Indicates the wind speed at a height of 2 meters. represents the saturated water vapor pressure, represents the actual water vapor pressure, represents the slope of the saturated water vapor pressure curve, represents the hygrometer constant;
[0055] S323, the drainage device in the high-standard farmland area divided by the field path is provided with a valve, and the leakage amount of the high-standard farmland area divided by the field path is set to , to determine the effective rainfall, evapotranspiration and infiltration. , open the drain valve to drain water; when , close the drain valve; when , close the valve of the drainage device and start the irrigation system for irrigation;
[0056] S33. After laying out the irrigation system and drainage device in the high-standard farmland area divided by the field paths, a high-standard farmland area is obtained after preliminary processing.
[0057] This invention uses an ant colony algorithm to find the optimal route length of ditch irrigation channels in high-standard farmland areas divided by field paths, and then calculates the water flow required in the high-standard farmland areas, thereby realizing the laying of the irrigation system; by judging the relationship between leakage, evaporation and transpiration and effective rainfall, the intelligent opening and closing of drainage valves is realized, thereby realizing the laying of drainage devices.
[0058] Preferably, S4 comprises the following steps:
[0059] S41. The steps for UAV flight path planning are as follows:
[0060] S411, set the number of bats to , the frequency of the sound waves emitted by bats is , the maximum value of the sound wave frequency is The minimum frequency of the sound wave is , is a random value and , the parameter iteration formula for simulating bat echo frequency is as follows,
[0061]
[0062] S412, bats are flying randomly, bats are at time points The position and velocity formulas at are as follows:
[0063]
[0064]
[0065] in, Indicates time point The location of Indicates time point The speed at Indicates time point The location of Indicates time point The speed at represents the global optimal solution of the previous generation;
[0066] S413, during the random flight of bats, when the bats are at the global optimal solution, the bats Update the position, the calculation formula is as follows,
[0067]
[0068] in, is the updated position, is the position before the update, is a random value and , Indicates Average loudness of the generation;
[0069] S414: The update After the position, reduce the loudness , increase the echo emission rate , and obtain the further global optimal solution, the calculation formula is as follows,
[0070]
[0071]
[0072] in, Indicates The average loudness of the generation, Indicates Substitute the echo emission rate, represents the initial echo emission rate, is a random value and , is a random value and ;
[0073] Set the uniform distribution number ,when Greater than , start a local search and The location is updated Position, when Less than or equal to , continue global search;
[0074] As the number of iterations increases, the loudness will tend to 0, the echo emission rate will tend to the initial echo emission rate , update the location , until the bats stop making sounds and find the global optimal solution;
[0075] S415: Setting the second threshold ,calculate The sum of ,when hour, is the optimal path length of the UAV in the high-standard farmland area after preliminary processing, otherwise it is not the optimal path of the UAV in the high-standard farmland area after preliminary processing;
[0076] S42. The UAV takes off from a specified location, passes through the optimal path of the high-standard farmland area after preliminary treatment, and then sprays fertilizers by the UAV to obtain the treated high-standard farmland area.
[0077] This invention uses a bat algorithm to achieve the optimal path selection for drones when fertilizing high-standard farmland. The algorithm used is clever and the results obtained are accurate. The optimal path saves fertilization costs and drone usage costs.
[0078] Preferably, the S5 comprises the following steps:
[0079] S51, using a crop pest image dataset, preprocessing the crop pest image dataset, adjusting each image in the crop pest image dataset to a size of 300×255 pixels, and obtaining a dataset to be feature extracted;
[0080] S52, extracting texture, color and HOG features from the crop pest images in the dataset to be feature extracted, the specific steps are as follows:
[0081] S521, using the gray level co-occurrence matrix to extract texture features, and counting the gray values of the crop pest image The pixel points and the grayscale values with spatial relationship are The number of occurrences of the pixel point in the crop pest image; when the gray value of the crop pest image is The pixel points and the grayscale values with spatial relationship are When the number of pixel points with the same appearance in the crop pest image is the same, the gray value is and The pixel points are tested for contrast and the third threshold is set ,when When the gray value is and The pixel point is a texture classification feature pixel point, otherwise it is not a texture classification feature pixel point; combining the texture classification feature pixel points to obtain texture features;
[0082] S522, extracting color features using a color histogram, extracting single red, green, and blue channels from the crop pest image, each color corresponding to a two-dimensional matrix, and obtaining three independent two-dimensional matrices; inputting the two-dimensional matrix into a histogram function in MATLAB to obtain histogram count values of the red, green, and blue channels, and using the histogram count values of the red, green, and blue channels as the main color features of the crop pest image to obtain color features;
[0083] S523, using a color histogram to extract HOG features, dividing the crop pest image into 16×16 pixels, performing block processing on the crop pest image, each block containing 2×2 cells and a size of 8×8 pixels, determining the gradient magnitude and gradient direction of each block; dividing the gradient direction into 9 angle intervals, and generating a 9-dimensional vector from the gradient histogram of the 9 angle intervals of each block; performing normalization processing on the 9-dimensional vector and then cascading the vectors to generate a vector containing 3780 features, thereby obtaining a HOG feature;
[0084] S53, inputting the texture features, color features and HOG features into SVM, training the SVM using the texture features, color features and HOG features until the SVM converges, stopping the training, and obtaining a trained SVM classifier;
[0085] S54, obtaining the processed monitoring video of the high-standard farmland area to obtain a monitoring video image, and inputting the monitoring video image into the trained SVM classifier to achieve pest identification;
[0086] S55, after the pests are identified, the locations where the pests appear are located, and the drone carrying the insecticide takes off from a specified location to the locations of the pests and sprays the insecticide to obtain a final high-standard farmland area.
[0087] The invention uses a crop pest image dataset and extracts texture features, color features and HOG features therein, inputs them into SVM for training to obtain a classifier, and then inputs surveillance video images in high-standard farmland areas into the classifier to achieve pest recognition. The recognition results obtained have a high accuracy rate, and finally the pests are located to achieve pest elimination;
[0088] Preferably, the S6 comprises the following steps:
[0089] S61. Use the spatial index R-tree to define the final high-standard farmland area for the inspectors. The specific steps are as follows:
[0090] S611, setting a final high-standard farmland area composed of a geo-fenced area, the geo-fenced area is divided into irregular polygons, the geo-fenced area is divided into minimum rectangles, and each boundary of the irregular polygon is covered to obtain a minimum rectangular coverage area;
[0091] S612: Setting the minimum rectangular coverage area ,in Indicates the coordinates of the lower left corner of the minimum rectangular coverage area. Indicates the coordinates of the lower right corner of the minimum rectangular coverage area; the minimum rectangular coverage area of each geo-fenced area is recorded as , , ..., ,in Indicates the number of geo-fenced areas;
[0092] S613: Expand and merge the adjacent minimum rectangular coverage areas to obtain , , ..., ,in Indicates the coordinates of the lower left corner of the minimum rectangular coverage area after expansion and merging. Represents the coordinates of the lower right corner of the smallest rectangular coverage area after expansion and merging; find out and Minimum value, and The maximum value of ,in Indicates the coordinates of the lower left corner of the smallest rectangular coverage area after merging again. Represents the coordinates of the lower right corner of the smallest rectangular coverage area after merging again; As The parent node of the node is added to the spatial index R-tree;
[0093] S614, repeat S613 and make a recursive call. When the spatial index R tree obtains a root node, the inspector conducts an inspection in the final high-standard farmland area. Otherwise, the inspector does not conduct an inspection in the final high-standard farmland area, which is deemed as crossing the boundary.
[0094] The invention divides the high-standard farmland area through geographic fence areas, obtains rectangular coverage areas, merges and expands them, repeats the above operations, and finally determines the root node of the spatial index R tree to judge whether the patrol personnel are patrolling in the high-standard farmland area. This method cleverly solves the task of monitoring daily patrol work and ensures the patrol quality of the high-standard farmland area.
[0095] This embodiment also discloses a high-standard farmland big data intelligent management platform based on the Internet of Things, which specifically includes: a high-standard farmland division module, an irrigation and drainage module, a pest control and fertilization module, and a patrol personnel management module;
[0096] The high-standard farmland division module is used to obtain high-standard farmland area images and identify field paths to divide high-standard farmland;
[0097] The irrigation and drainage module is used to irrigate high-standard farmland and perform intelligent drainage;
[0098] The killing and fertilizing module is used to identify and locate pests to disinfect them, and then fertilize the high-standard farmland;
[0099] The patrol personnel management module is used to monitor the patrol routes and areas of high-standard farmland patrol personnel.
[0100] (III) Beneficial effects
[0101] The present invention provides a high-standard farmland big data intelligent management platform based on the Internet of Things. It has the following beneficial effects:
[0102] 1. The invention obtains the required high-standard farmland area image through GPS positioning, and uses the U-Net network to realize the recognition of field paths in the high-standard farmland area, and divides the high-standard farmland area for subsequent processing; then the irrigation system is laid, and the ant colony algorithm is used to obtain the best irrigation laying route, which saves the laying cost, and calculates the water flow required for the high-standard farmland area for irrigation; the drainage device takes into account the leakage, evaporation and transpiration and effective rainfall, and realizes the opening of the drainage device or irrigation through judgment; secondly, the drone fertilization route is considered, and the bat algorithm is used to plan the drone route to obtain the optimal route and fertilize, saving the cost of drone use; relevant features are extracted and input into the trained classifier, the position of pests is located, and drones are dispatched to spray insecticides to disinfect pests in high-standard farmland, realizing intelligent and high-precision pest control; finally, the patrol routes and areas of patrol personnel are monitored to ensure the quality of daily patrols of high-standard farmland.
[0103] 2. The invention realizes semantic segmentation training through the U-Net network, uses the semantic segmentation training model to determine the pixels of field paths in the high-standard farmland area image, and fits the center position of the road that meets the requirements, thereby identifying the field paths; uses the neural network to process the image, and the obtained recognition result is highly accurate, which is convenient for subsequent operations.
[0104] 3. The invention uses an ant colony algorithm to find the optimal route length of ditch irrigation channels in high-standard farmland areas divided by field paths, and then calculates the water flow required in the high-standard farmland areas, thereby realizing the laying of the irrigation system; by judging the relationship between leakage, evaporation and transpiration and effective rainfall, the intelligent opening and closing of drainage valves is realized, thereby realizing the laying of drainage devices.
[0105] 4. This invention uses a bat algorithm to achieve the optimal path selection for drones when applying fertilizers on high-standard farmland. The algorithm used is clever and the results obtained are accurate. The optimal path saves the cost of fertilization and the cost of using drones.
[0106] 5. The invention uses a crop pest image dataset, extracts texture features, color features and HOG features, inputs them into SVM for training to obtain a classifier, and then inputs surveillance video images in high-standard farmland areas into the classifier to achieve pest recognition. The recognition results obtained have a high accuracy rate, and finally the pests are located to achieve pest elimination;
[0107] 6. The invention divides the high-standard farmland area through geographic fence areas, obtains rectangular coverage areas, merges and expands them, repeats the above operations, and finally determines the root node of the spatial index R tree to determine whether the patrol personnel are patrolling in the high-standard farmland area. This method cleverly solves the task of monitoring daily patrol work and ensures the patrol quality of the high-standard farmland area. BRIEF DESCRIPTION OF THE DRAWINGS
[0108] Figure 1 A schematic diagram of the process of a high-standard farmland big data intelligent management platform based on the Internet of Things provided by the present invention. DETAILED DESCRIPTION
[0109] 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.
[0110] This implementation discloses a high-standard farmland big data intelligent management method based on the Internet of Things, which specifically includes the following contents:
[0111] S1. Select high-standard farmland, use GPS positioning to obtain the GPS coordinates of each vertex of the high-standard farmland, and connect the GPS coordinates in sequence to obtain the high-standard farmland area;
[0112] S2, identifying the field roads in the high-standard farmland area, and marking them in the high-standard farmland area, to obtain the high-standard farmland area divided by the field roads;
[0113] S3, laying irrigation systems and drainage devices in the high-standard farmland areas divided by the field paths to obtain high-standard farmland areas after preliminary treatment;
[0114] S4, planning a drone fertilization route, and performing drone fertilization on the high-standard farmland area after the preliminary treatment to obtain a treated high-standard farmland area;
[0115] S5. In the treated high-standard farmland area, feature extraction and SVM classifier are used to identify pests, and after locating the pest area, a drone carrying pesticide is used to spray the pesticide in the pest area to obtain a final high-standard farmland area;
[0116] S6. The inspectors inspect the final high-standard farmland area and use the spatial index R-tree to monitor whether the inspectors cross the boundary.
[0117] Preferably, the S1 comprises the following steps:
[0118] S11, along the edge of the high-standard farmland, using the GPS to collect the coordinates of each vertex of the high-standard farmland to obtain the coordinates of each vertex, and obtain a coordinate set;
[0119] S12. Connect the coordinates in the coordinate set to obtain a high-standard farmland area.
[0120] Preferably, S2 comprises the following steps:
[0121] S21. In the U-Net network, the input feature map is set. There are 4 downsampling modules in the encoding stage. Each downsampling module contains 2 3×3 convolutional layers with 0 padded pixels and length 1 and 1 maximum pooling layer with 2 padded pixels. After the feature map is processed by 1 downsampling module, the feature map size becomes ,in is the number of image rows, is the number of image columns; the feature map is processed by four downsampling modules in sequence and then by two 3×3 convolutional layers with 0 padded pixels and a stride of 1 before entering the decoding stage;
[0122] S22, there are 4 upsampling modules in the decoding stage, each of which contains 2 3×3 deconvolution layers with 0 padded pixels and a stride of 1, 1 jump connection and 2 3×3 deconvolution layers with a stride of 1; after being processed by 1 upsampling module, the deconvolution layer is first used to double the size of the feature map, the number of channels is halved, and then it is spliced with the feature map generated in the encoding stage through a jump connection; before the jump connection, the feature map size is cropped, and the spliced feature map is convolved twice, and the feature map size becomes ; After 4 upsampling modules, the semantic segmentation model is finally obtained;
[0123] S23, perform semantic segmentation training on the semantic segmentation model, set represents the regressed pixel value, Indicates location aisle The pixel score on Represents the number of classes and outputs the regression pixel value in the feature map. The calculation formula is as follows:
[0124]
[0125] Setting the energy function Represents the cross entropy loss function for regressing pixel values, Indicates location The weight of Indicates location The label value of Indicates location The softmax loss function is Represents parameters; sets a training error threshold. When the training error is less than the training error threshold, stops semantic segmentation training and obtains a semantic segmentation training model. The calculation formula is as follows:
[0126]
[0127] S24, after the semantic segmentation model training is completed, a semantic segmentation training model is obtained, a high-standard farmland area image is obtained from the high-standard farmland area, the high-standard farmland area image is input into the semantic segmentation training model to obtain the regression probability of each pixel, a critical threshold is set, and when the regression probability is greater than the critical threshold, the regression probability is retained, otherwise the regression probability is discarded; the retained regression probability is fitted into a corresponding mask area, the mask area is binarized by a scanning method to obtain edge information points, the edge information points are weighted summed to obtain the road center position, and the field path in the high-standard farmland is identified after fitting the road center position;
[0128] S25. Mark the field paths in the high-standard farmland area to obtain the high-standard farmland area divided by the field paths.
[0129] Preferably, S3 comprises the following steps:
[0130] S31. Laying an irrigation system in the high-standard farmland area. The specific steps are as follows:
[0131] S311, Setting Moment Ants from the location Climb to the next position , initially releases pheromones and inspirational information , express The probability of transition at a certain moment, It is a pheromone-inducing factor. is the expectation heuristic factor, Indicates location Climb to the next position The distance express The moment expectation heuristic function is calculated as follows:
[0132]
[0133]
[0134] S312, ants crawl to the location , release pheromones in the path passed, pheromones evaporate over time, and ants will release new pheromones; set the maximum number of iterations in the crawling process, when the number of iterations is greater than the maximum number of iterations, stop the iteration, the formula is as follows,
[0135]
[0136]
[0137]
[0138] in, represents the pheromone volatility coefficient, and , Indicates the ant from the position Climb to the next position The optimal amount of pheromone released, Indicates the intensity of pheromone release. represents the optimal path length for ants to crawl, express Release pheromones all the time. ;
[0139] S313, the optimal path length of the ants obtained after the iteration , set the first threshold ,when When It is the optimal route length of the ditch irrigation road in one of the high-standard farmland areas divided by field roads;
[0140] S314, repeat S311 and S312, and record the optimal route lengths of all ditch irrigation roads in the high-standard farmland area divided by field paths as a data set , Indicates the number of high-standard farmland areas divided by field paths;
[0141] S315. Use buried pipelines to supply water to high-standard farmland, and use furrow irrigation for irrigation. The net flow calculation formula for the furrow irrigation channel in one of the high-standard farmland areas divided by field paths is as follows:
[0142]
[0143] in, represents the net flow of the ditch irrigation channel, is the irrigation area of the ditch irrigation channel, Indicates the watering modulus;
[0144] S316, set the loss coefficient of the ditch irrigation channel to The gross flow calculation formula of the ditch irrigation channel in one of the high-standard farmland areas divided by field paths is as follows:
[0145]
[0146] in, It represents the gross flow of the ditch irrigation channel. Indicates The optimal route length of the irrigation channel in the high-standard farmland area divided by field paths, , ;
[0147] S317, the optimal route degree of all ditch irrigation roads in the high-standard farmland area divided by the field path is recorded as a data set , the calculation formula of high-standard farmland flow is as follows,
[0148]
[0149] in, It indicates the total flow required for high-standard farmland;
[0150] S32. Drainage devices are set up in the high-standard farmland areas divided by field paths. The specific steps are as follows:
[0151] S321. Use the empirical method to calculate the effective rainfall, and set the effective rainfall to , the rainfall is , the calculation formula is as follows,
[0152]
[0153] in, It indicates the effective utilization coefficient of rainfall. When the rainfall is less than 5mm, the effective coefficient of rainfall is 0; when the rainfall is 5mm-50mm, the effective coefficient of rainfall is 1.0-0.8; when the rainfall is greater than 50mm, the effective coefficient of rainfall is 0.8-0.7;
[0154] S322. Use the comprehensive method to calculate evaporation and transpiration. The calculation formula is as follows:
[0155]
[0156] in, represents evapotranspiration, represents the net radiation, Indicates the heat transfer of soil. It indicates the daily average humidity at a height of 2 meters. Indicates the wind speed at a height of 2 meters. represents the saturated water vapor pressure, represents the actual water vapor pressure, represents the slope of the saturated water vapor pressure curve, represents the hygrometer constant;
[0157] S323, the drainage device in the high-standard farmland area divided by the field path is provided with a valve, and the leakage amount of the high-standard farmland area divided by the field path is set to , to determine the effective rainfall, evapotranspiration and infiltration. , open the drain valve to drain water; when , close the drain valve; when , close the valve of the drainage device and start the irrigation system for irrigation;
[0158] S33. After laying out the irrigation system and drainage device in the high-standard farmland area divided by the field paths, a high-standard farmland area is obtained after preliminary processing.
[0159] Preferably, S4 comprises the following steps:
[0160] S41. The steps for UAV flight path planning are as follows:
[0161] S411, set the number of bats to , the frequency of the sound waves emitted by bats is , the maximum value of the sound wave frequency is The minimum frequency of the sound wave is , is a random value and , the parameter iteration formula for simulating bat echo frequency is as follows,
[0162]
[0163] S412, bats are flying randomly, bats are at time points The position and velocity formulas at are as follows:
[0164]
[0165]
[0166] in, Indicates time point The location of Indicates time point The speed at Indicates time point The location of Indicates time point The speed at represents the global optimal solution of the previous generation;
[0167] S413, during the random flight of bats, when the bats are at the global optimal solution, the bats Update the position, the calculation formula is as follows,
[0168]
[0169] in, is the updated position, is the position before the update, is a random value and , Indicates Average loudness of the generation;
[0170] S414: The update After the position, reduce the loudness , increase the echo emission rate , and obtain the further global optimal solution, the calculation formula is as follows,
[0171]
[0172]
[0173] in, Indicates The average loudness of the generation, Indicates Substitute the echo emission rate, represents the initial echo emission rate, is a random value and , is a random value and ;
[0174] Set the uniform distribution number ,when Greater than , start a local search and The location is updated Position, when Less than or equal to , continue global search;
[0175] As the number of iterations increases, the loudness will tend to 0, the echo emission rate will tend to the initial echo emission rate , update the location , until the bats stop making sounds and find the global optimal solution;
[0176] S415: Setting the second threshold ,calculate The sum of ,when hour, is the optimal path length of the UAV in the high-standard farmland area after preliminary processing, otherwise it is not the optimal path of the UAV in the high-standard farmland area after preliminary processing;
[0177] S42. The UAV takes off from a specified location, passes through the optimal path of the high-standard farmland area after preliminary treatment, and then sprays fertilizers by the UAV to obtain the treated high-standard farmland area.
[0178] Preferably, the S5 comprises the following steps:
[0179] S51, using a crop pest image dataset, preprocessing the crop pest image dataset, adjusting each image in the crop pest image dataset to a size of 300×255 pixels, and obtaining a dataset to be feature extracted;
[0180] S52, extracting texture, color and HOG features from the crop pest images in the dataset to be feature extracted, the specific steps are as follows:
[0181] S521, using the gray level co-occurrence matrix to extract texture features, and counting the gray values of the crop pest image The pixel points and the grayscale values with spatial relationship are The number of occurrences of the pixel point in the crop pest image; when the gray value of the crop pest image is The pixel points and the grayscale values with spatial relationship are When the number of pixel points with the same appearance in the crop pest image is the same, the gray value is and The pixel points are tested for contrast and the third threshold is set ,when When the gray value is and The pixel point is a texture classification feature pixel point, otherwise it is not a texture classification feature pixel point; combining the texture classification feature pixel points to obtain texture features;
[0182] S522, extracting color features using a color histogram, extracting single red, green, and blue channels from the crop pest image, each color corresponding to a two-dimensional matrix, and obtaining three independent two-dimensional matrices; inputting the two-dimensional matrix into a histogram function in MATLAB to obtain histogram count values of the red, green, and blue channels, and using the histogram count values of the red, green, and blue channels as the main color features of the crop pest image to obtain color features;
[0183] S523, using a color histogram to extract HOG features, dividing the crop pest image into 16×16 pixels, performing block processing on the crop pest image, each block containing 2×2 cells and a size of 8×8 pixels, determining the gradient magnitude and gradient direction of each block; dividing the gradient direction into 9 angle intervals, and generating a 9-dimensional vector from the gradient histogram of the 9 angle intervals of each block; performing normalization processing on the 9-dimensional vector and then cascading the vectors to generate a vector containing 3780 features, thereby obtaining a HOG feature;
[0184] S53, inputting the texture features, color features and HOG features into SVM, training the SVM using the texture features, color features and HOG features until the SVM converges, stopping the training, and obtaining a trained SVM classifier;
[0185] S54, obtaining the processed monitoring video of the high-standard farmland area to obtain a monitoring video image, and inputting the monitoring video image into the trained SVM classifier to achieve pest identification;
[0186] S55, after the pests are identified, the locations where the pests appear are located, and the drone carrying the insecticide takes off from a specified location to the locations of the pests and sprays the insecticide to obtain a final high-standard farmland area.
[0187] Preferably, the S6 comprises the following steps:
[0188] S61. Use the spatial index R-tree to define the final high-standard farmland area for the inspectors. The specific steps are as follows:
[0189] S611, setting a final high-standard farmland area composed of a geo-fenced area, the geo-fenced area is divided into irregular polygons, the geo-fenced area is divided into minimum rectangles, and each boundary of the irregular polygon is covered to obtain a minimum rectangular coverage area;
[0190] S612: Setting the minimum rectangular coverage area ,in Indicates the coordinates of the lower left corner of the minimum rectangular coverage area. Indicates the coordinates of the lower right corner of the minimum rectangular coverage area; the minimum rectangular coverage area of each geo-fenced area is recorded as , , ..., ,in Indicates the number of geo-fenced areas;
[0191] S613: Expand and merge the adjacent minimum rectangular coverage areas to obtain , , ..., ,in Indicates the coordinates of the lower left corner of the minimum rectangular coverage area after expansion and merging. Represents the coordinates of the lower right corner of the smallest rectangular coverage area after expansion and merging; find out and Minimum value, and The maximum value of ,in Indicates the coordinates of the lower left corner of the smallest rectangular coverage area after merging again. Represents the coordinates of the lower right corner of the smallest rectangular coverage area after merging again; As The parent node of the node is added to the spatial index R-tree;
[0192] S614, repeat S613 and make a recursive call. When the spatial index R tree obtains a root node, the inspector conducts an inspection in the final high-standard farmland area. Otherwise, the inspector does not conduct an inspection in the final high-standard farmland area, which is deemed as crossing the boundary.
[0193] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A high-standard farmland big data intelligent management method based on the Internet of Things, characterized in that: The steps include: S1. Select high-standard farmland, use GPS positioning to obtain the GPS coordinates of each vertex of the high-standard farmland, and connect the GPS coordinates in sequence to obtain the high-standard farmland area; S2, identifying the field roads in the high-standard farmland area, and marking them in the high-standard farmland area, to obtain the high-standard farmland area divided by the field roads; S3, laying irrigation systems and drainage devices in the high-standard farmland areas divided by the field paths to obtain high-standard farmland areas after preliminary treatment; S4, planning a drone fertilization route, and performing drone fertilization on the high-standard farmland area after the preliminary treatment to obtain a treated high-standard farmland area; S5. In the treated high-standard farmland area, feature extraction and SVM classifier are used to identify pests, and after locating the pest area, a drone carrying pesticide is used to spray the pesticide in the pest area to obtain a final high-standard farmland area; S6, the inspectors inspect the final high-standard farmland area and use the spatial index R-tree to monitor whether the inspectors cross the boundary; The S1 comprises the following steps: S11, along the edge of the high-standard farmland, using the GPS to collect the coordinates of each vertex of the high-standard farmland to obtain the coordinates of each vertex, and obtain a coordinate set; S12, connecting the coordinates in the coordinate set to obtain a high-standard farmland area; The S2 comprises the following steps: Perform semantic segmentation training on the semantic segmentation model and set represents the regressed pixel value, Indicates location aisle The pixel score on Represents the number of classes and outputs the regression pixel value in the feature map. The calculation formula is as follows: ; Setting the energy function Represents the cross entropy loss function for regressing pixel values, Indicates location The weight of Indicates location The label value of Indicates location The softmax loss function is Indicates parameters; A training error threshold is set. When the training error is less than the training error threshold, the semantic segmentation training is stopped to obtain a semantic segmentation training model. The calculation formula is as follows: ; After the semantic segmentation model training is completed, a semantic segmentation training model is obtained, a high-standard farmland area image is obtained from the high-standard farmland area, the high-standard farmland area image is input into the semantic segmentation training model to obtain the regression probability of each pixel, a critical threshold is set, and when the regression probability is greater than the critical threshold, the regression probability is retained, otherwise the regression probability is discarded; the retained regression probability is fitted into a corresponding mask area, the mask area is binarized by a scanning method to obtain edge information points, the edge information points are weighted and summed to obtain the center position of the road, and after fitting the center position of the road, a field path in the high-standard farmland is identified; Marking the field paths in the high-standard farmland area to obtain the high-standard farmland area divided by the field paths; The S3 comprises the following steps: S31. Laying an irrigation system in the high-standard farmland area. The specific steps are as follows: S311, Setting Moment Ants from the location Climb to the next position , initially releases pheromones and inspirational information , express The probability of transition at a certain moment, It is a pheromone-inducing factor. is the expectation heuristic factor, Indicates location Climb to the next position The distance express The moment expectation heuristic function is calculated as follows: ; ; S312, ants crawl to the location , release pheromones in the path passed, pheromones evaporate over time, and ants will release new pheromones; set the maximum number of iterations in the crawling process, when the number of iterations is greater than the maximum number of iterations, stop the iteration, the formula is as follows, ; ; ; in, represents the pheromone volatility coefficient, and , Indicates that the ant is from position Climb to the next position The optimal amount of pheromone released, Indicates the intensity of pheromone release. represents the optimal path length for ants to crawl, express Release pheromones all the time. ; S313, the optimal path length of the ants obtained after the iteration , set the first threshold ,when When It is the optimal route length of the ditch irrigation road in one of the high-standard farmland areas divided by field roads; S314, repeat S311 and S312, and record the optimal route lengths of all ditch irrigation roads in the high-standard farmland area divided by field paths as a data set , Indicates the number of high-standard farmland areas divided by field paths; S315. Use buried pipelines to supply water to high-standard farmland, and use furrow irrigation for irrigation. The net flow calculation formula for the furrow irrigation channel in one of the high-standard farmland areas divided by field paths is as follows: ; in, represents the net flow of the ditch irrigation channel, is the irrigation area of the ditch irrigation channel, Indicates the watering modulus; S316, set the loss coefficient of the ditch irrigation channel to The gross flow calculation formula of the ditch irrigation channel in one of the high-standard farmland areas divided by field paths is as follows: ; in, It represents the gross flow of the ditch irrigation channel. Indicates The optimal route length of the irrigation channel in the high-standard farmland area divided by field paths, , ; S317, the optimal route degree of all ditch irrigation roads in the high-standard farmland area divided by the field path is recorded as a data set , the calculation formula of high-standard farmland flow is as follows, ; in, It indicates the total flow required for high-standard farmland; S32. Drainage devices are set up in the high-standard farmland areas divided by field paths. The specific steps are as follows: S321. Use the empirical method to calculate the effective rainfall, and set the effective rainfall to , the rainfall is , the calculation formula is as follows, ; in, It indicates the effective utilization coefficient of rainfall. When the rainfall is less than 5mm, the effective coefficient of rainfall is 0; when the rainfall is 5mm-50mm, the effective coefficient of rainfall is 1.0-0.8; when the rainfall is greater than 50mm, the effective coefficient of rainfall is 0.8-0.7; S322. Use the comprehensive method to calculate evaporation and transpiration. The calculation formula is as follows: ; in, represents evapotranspiration, represents the net radiation, Indicates the heat transfer of soil. It indicates the daily average humidity at a height of 2 meters. Indicates the wind speed at a height of 2 meters. represents the saturated water vapor pressure, represents the actual water vapor pressure, represents the slope of the saturated water vapor pressure curve, represents the hygrometer constant; S323, the drainage device in the high-standard farmland area divided by the field path is provided with a valve, and the leakage amount of the high-standard farmland area divided by the field path is set to , to determine the effective rainfall, evapotranspiration and infiltration. , open the drain valve to drain water; when , close the drain valve; when , close the valve of the drainage device and start the irrigation system for irrigation; S33, after laying out the irrigation system and drainage device in the high-standard farmland area divided by the field path, a high-standard farmland area after preliminary treatment is obtained; The S4 comprises the following steps: S41. The steps for UAV flight path planning are as follows: S411, set the number of bats to , the frequency of the sound waves emitted by bats is , the maximum value of the sound wave frequency is The minimum frequency of the sound wave is , is a random value and , the parameter iteration formula for simulating bat echo frequency is as follows, ; S412, bats are flying randomly, bats are at time points The position and velocity formulas at are as follows: ; ; in, Indicates time point The location of Indicates time point The speed at Indicates time point The location of Indicates time point The speed at represents the global optimal solution of the previous generation; S413, during the random flight of bats, when the bats are at the global optimal solution, the bats Update the position, the calculation formula is as follows, ; in, is the updated position, is the position before the update, is a random value and , Indicates Average loudness of the generation; S414: The update After the position, reduce the loudness , increase the echo emission rate , and obtain the further global optimal solution, the calculation formula is as follows, ; ; in, Indicates The average loudness of the generation, Indicates Substitute the echo emission rate, represents the initial echo emission rate, is a random value and , is a random value and ; Set the uniform distribution number ,when Greater than , start a local search and The location is updated Position, when Less than or equal to , continue global search; As the number of iterations increases, the loudness will tend to 0, the echo emission rate will tend to the initial echo emission rate , update the location , until the bats stop making sounds and find the global optimal solution; S415: Setting the second threshold ,calculate The sum of ,when hour, is the optimal path length of the UAV in the high-standard farmland area after preliminary processing, otherwise it is not the optimal path of the UAV in the high-standard farmland area after preliminary processing; S42, the drone takes off from the specified location, passes through the optimal path of the high-standard farmland area after the preliminary treatment, and then sprays fertilizer by the drone to obtain the treated high-standard farmland area; The S5 comprises the following steps: S51, using a crop pest image dataset, preprocessing the crop pest image dataset, adjusting each image in the crop pest image dataset to a size of 300×255 pixels, and obtaining a dataset to be feature extracted; S52, extracting texture, color and HOG features from the crop pest images in the dataset to be feature extracted, the specific steps are as follows: S521, using the gray level co-occurrence matrix to extract texture features, and counting the gray values of the crop pest image The pixel points and the grayscale values with spatial relationship are The number of occurrences of the pixel point in the crop pest image; when the gray value of the crop pest image is The pixel points and the grayscale values with spatial relationship are When the number of pixel points with the same appearance in the crop pest image is the same, the gray value is and The pixel points are tested for contrast and the third threshold is set ,when When the gray value is and The pixel point is a texture classification feature pixel point, otherwise it is not a texture classification feature pixel point; combining the texture classification feature pixel points to obtain texture features; S522, extracting color features using a color histogram, extracting single red, green, and blue channels from the crop pest image, each color corresponding to a two-dimensional matrix, and obtaining three independent two-dimensional matrices; inputting the two-dimensional matrix into a histogram function in MATLAB to obtain histogram count values of the red, green, and blue channels, and using the histogram count values of the red, green, and blue channels as the main color features of the crop pest image to obtain color features; S523, using a color histogram to extract HOG features, dividing the crop pest image into 16×16 pixels, performing block processing on the crop pest image, each block containing 2×2 cells and a size of 8×8 pixels, determining the gradient magnitude and gradient direction of each block; dividing the gradient direction into 9 angle intervals, and generating a 9-dimensional vector from the gradient histogram of the 9 angle intervals of each block; performing normalization processing on the 9-dimensional vector and then cascading the vectors to generate a vector containing 3780 features, thereby obtaining a HOG feature; S53, inputting the texture features, color features and HOG features into SVM, training the SVM using the texture features, color features and HOG features until the SVM converges, stopping the training, and obtaining a trained SVM classifier; S54, obtaining the processed monitoring video of the high-standard farmland area to obtain a monitoring video image, and inputting the monitoring video image into the trained SVM classifier to achieve pest identification; S55, after the pests are identified, the location where the pests appear is located, and the drone carrying the insecticide takes off from the specified location to the location of the pests and sprays the insecticide to obtain the final high-standard farmland area; The S6 comprises the following steps: S61. Use the spatial index R-tree to define the final high-standard farmland area for the inspectors. The specific steps are as follows: S611, setting a final high-standard farmland area composed of a geo-fenced area, the geo-fenced area is divided into irregular polygons, the geo-fenced area is divided into minimum rectangles, and each boundary of the irregular polygon is covered to obtain a minimum rectangular coverage area; S612: Setting the minimum rectangular coverage area ,in Indicates the coordinates of the lower left corner of the minimum rectangular coverage area. Indicates the coordinates of the lower right corner of the minimum rectangular coverage area; the minimum rectangular coverage area of each geo-fenced area is recorded as , , ..., ,in Indicates the number of geo-fenced areas; S613: Expand and merge the adjacent minimum rectangular coverage areas to obtain , , ..., ,in Indicates the coordinates of the lower left corner of the minimum rectangular coverage area after expansion and merging. Represents the coordinates of the lower right corner of the smallest rectangular coverage area after expansion and merging; find out and Minimum value, and The maximum value of ,in Indicates the coordinates of the lower left corner of the smallest rectangular coverage area after merging again. Represents the coordinates of the lower right corner of the smallest rectangular coverage area after merging again; As The parent node of the node is added to the spatial index R-tree; S614, repeat S613 and make a recursive call. When the spatial index R tree obtains a root node, the inspector conducts an inspection in the final high-standard farmland area. Otherwise, the inspector does not conduct an inspection in the final high-standard farmland area, which is deemed as crossing the boundary.
2. A management platform for implementing the high-standard farmland big data intelligent management method based on the Internet of Things as described in claim 1, characterized in that: Specifically include: Divide into high-standard farmland module, irrigation and drainage module, disinfection and fertilization module and patrol personnel management module; The high-standard farmland division module is used to obtain high-standard farmland area images and identify field paths to divide high-standard farmland; The irrigation and drainage module is used to irrigate high-standard farmland and perform intelligent drainage; The killing and fertilizing module is used to identify and locate pests to disinfect them, and then fertilize the high-standard farmland; The patrol personnel management module is used to monitor the patrol routes and areas of high-standard farmland patrol personnel.
