Agricultural machine operation planning method and system based on unmanned aerial vehicle

Abstract

The invention discloses an agricultural machinery operation planning method and system based on an unmanned aerial vehicle, and the method comprises the steps: obtaining a high-resolution remote sensing image of a target farmland region, and carrying out the denoising preprocessing of the remote sensing image through a pre-constructed convolutional neural network model, and obtaining a preprocessed image; generating a preliminary farmland operation flight path planning scheme by adopting a deep reinforcement learning algorithm and combining performance parameters and energy consumption constraint conditions of the unmanned aerial vehicle according to crop growth stage information acquired from the crop image; in the farmland operation process, farmland environment parameter data are collected in real time through an airborne sensor, and the flight attitude and track parameters of the unmanned aerial vehicle are dynamically adjusted by adopting a Kalman filtering algorithm so as to ensure that the collection quality of remote sensing images meets farmland monitoring requirements.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to a UAV-based agricultural machinery operation planning method and system. BACKGROUND

[0002] When using a UAV to plan agricultural field operations, the technical challenge of how to efficiently and accurately complete the surveying and regionalization of the agricultural field arises. Due to the complex and variable topography of the agricultural field, the soil fertility, humidity, and other conditions of different regions differ greatly, and the growth of crops is not the same. Therefore, when the UAV performs aerial photography tasks, it needs to dynamically adjust the flight height, speed, route, and other parameters according to the actual situation of the agricultural field to obtain clear and comprehensive agricultural field image data. At the same time, when identifying and processing the agricultural field image, due to the complex agricultural field environment and variable lighting conditions, there may be a large amount of noise and interference factors in the image. How to accurately extract the boundary information of the agricultural field region and finely classify and regionalize according to the crop type, growth stage, and other characteristics is a technical challenge that needs to be solved. In addition, due to the large differences in operation requirements of different agricultural field regions, how to reasonably plan the operation path and operation parameters of the UAV according to the regionalization results and the growth of crops is also a complex optimization problem that needs to consider multiple factors such as the performance parameters of the UAV, operation efficiency, energy consumption, etc., and solve it through intelligent algorithms to realize the intelligent and accurate management of agricultural field operations. SUMMARY

[0003] The technical problem to be solved by the present application is to provide a UAV-based agricultural machinery operation planning method and system,

[0004] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0005] A UAV-based agricultural machinery operation planning method, comprising:

[0006] Step S1, obtaining a preprocessed image from a high-resolution remote sensing image of a target agricultural field region;

[0007] Step S2, obtaining a segmentation image containing clear agricultural field boundary information from the preprocessed image;

[0008] Step S3, obtaining a region image containing crop type regionalization results from the segmentation image;

[0009] Step S4, generating a crop image containing crop growth stage and humidity distribution information from the region image;

[0010] Step S5, generating a preliminary agricultural field operation flight path planning scheme from the crop image;

[0011] Step S6, according to the preliminary farmland operation flight path planning scheme, the flight attitude and track parameters of the unmanned aerial vehicle are dynamically adjusted.

[0012] As preferred, in step S1, the convolutional neural network model constructed in advance is used for denoising preprocessing of the remote sensing image to obtain a preprocessed image.

[0013] As preferred, in step S2, according to the preprocessed image, an edge detection algorithm is used to extract the farmland boundary feature, and combined with the preset soil fertility distribution data, an adaptive threshold segmentation method is used for fine segmentation of the farmland region to obtain a segmented image containing clear farmland boundary information.

[0014] As preferred, in step S3, according to the farmland boundary information extracted from the segmented image, a region growing algorithm is used to preliminarily divide each farmland region, and combined with the pre-established crop type feature database, the crop type of each region is judged to obtain a region image containing the crop type region division result.

[0015] As preferred, in step S4, according to the region image, a pre-trained support vector machine classification model is used to identify the growth stage of each region crop, and combined with the real-time humidity data collected by the farmland humidity sensor, a crop image containing the growth stage and humidity distribution information of the crop is generated.

[0016] As preferred, in step S5, according to the crop growth stage information obtained from the crop image, a deep reinforcement learning algorithm is used, combined with the performance parameters and energy consumption constraints of the unmanned aerial vehicle, to generate a preliminary farmland operation flight path planning scheme.

[0017] As preferred, in step S6, a Kalman filter algorithm is used to dynamically adjust the flight attitude and track parameters of the unmanned aerial vehicle.

[0018] The application also provides an unmanned aerial vehicle-based agricultural machinery operation planning system, comprising:

[0019] The first processing device is used for obtaining a preprocessed image according to the high-resolution remote sensing image of the target farmland region.

[0020] The second processing device is used for obtaining a segmented image containing clear farmland boundary information according to the preprocessed image.

[0021] The third processing device is used for obtaining a region image containing the crop type region division result according to the segmented image.

[0022] The fourth processing device is used for generating a crop image containing the growth stage and humidity distribution information of the crop according to the region image.

[0023] a fifth processing device for generating a preliminary flight path planning scheme for farmland operations based on the crop images;

[0024] The sixth processing device is used to dynamically adjust the flight attitude and track parameters of the UAV according to the preliminary farmland operation flight path planning scheme.

[0025] This hair has the following beneficial effects:

[0026] The present invention first acquires high-resolution remote sensing images, extracts farmland boundaries through convolutional neural network denoising and edge detection algorithms, and performs refined segmentation in combination with soil fertility data. Crop types are then divided using a region growing algorithm and a crop feature database, followed by identification of growth stages using a support vector machine, and integration of real-time humidity data to generate a comprehensive crop image. Based on this information, a deep reinforcement learning algorithm is used to plan the optimal flight path, taking into account drone performance and energy consumption constraints. During the operation, the present invention also uses onboard sensors to collect environmental parameters in real time, and uses a Kalman filter algorithm to dynamically adjust flight attitude and trajectory to ensure the quality of remote sensing data. The present invention realizes refined monitoring of farmland and intelligent operation planning, thereby improving agricultural production efficiency and resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of the agricultural machinery operation planning method based on drones of the present invention. DETAILED DESCRIPTION

[0028] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0029] Example 1:

[0030] like Figure 1 As shown, an embodiment of the present invention provides an agricultural machinery operation planning method based on a drone, comprising:

[0031] S101. Acquire a high-resolution remote sensing image of a target farmland area, and use a pre-built convolutional neural network model to perform denoising preprocessing on the remote sensing image to obtain a preprocessed image.

[0032] High-resolution remote sensing image data of the target farmland area is acquired, and the image data is input into a pre-trained convolutional neural network model. The convolutional neural network model adopts a structure of multiple convolutional layers and pooling layers, and performs feature extraction and noise reduction processing on the image through a convolution kernel. According to the characteristics of the image, appropriate convolution kernel size, convolution layer number and pooling layer parameters are set in the convolutional neural network model to achieve the best denoising effect. The denoised image data is output to obtain a preliminary preprocessed remote sensing image. The preliminary preprocessed remote sensing image is evaluated, and whether the denoising effect reaches the expected target is judged by calculating the signal-to-noise ratio, definition and other indicators of the image. If the denoising effect does not reach the expected target, the parameters of the convolutional neural network model are adjusted, such as increasing the number of convolutional layers, adjusting the size of the convolution kernel, etc., and the image is denoised again. The remote sensing image that reaches the expected denoising effect is used as input data for subsequent farmland analysis, providing high-quality data support for precision agriculture.

[0033] Specifically, high-resolution remote sensing image data is an important basis for precision agriculture. After obtaining the remote sensing image of the target farmland area, it needs to be denoised by a convolutional neural network model to improve the image quality. The convolutional neural network model is composed of multiple convolutional layers and pooling layers, which can effectively extract image features and reduce noise. Taking a rice field as an example, a multispectral remote sensing image with a resolution of 0.5 meters can be used. This high-resolution image can clearly show the growth status of rice, but may contain atmospheric scattering, sensor noise and other interference. To remove these noises, a 5-layer convolutional neural network model is designed. The first layer uses 32 3x3 convolutional kernels to capture the texture features of rice leaves. The second and third layers use 64 and 128 3x3 convolutional kernels respectively to further extract more complex features. The fourth and fifth layers use 256 3x3 convolutional kernels to identify large-scale features such as rice plant type. Each convolutional layer is followed by a 2x2 max pooling layer to reduce the spatial size of the feature map and improve the computational efficiency of the model. After convolutional neural network processing, a preliminary denoised remote sensing image is obtained. To evaluate the denoising effect, the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) of the image are calculated. For example, the PSNR of the original image is 25 dB, and the SSIM is 0.75, while the PSNR of the denoised image is increased to 30 dB, and the SSIM is increased to 0.85, indicating that the denoising effect is significant. If the denoising effect does not meet expectations, model parameters can be adjusted. For example, increase the number of convolutional layers to 7, or change some 3x3 convolutional kernels to 5x5 to capture more spatial information. At the same time, residual connections can be introduced to help the model learn the difference between the original image and the noise, further improving the denoising effect. Optimized remote sensing images provide high-quality data support for subsequent farmland analysis. For example, denoised images can be used for crop classification, growth condition monitoring and yield prediction. In crop classification, clear images help distinguish between different crop types such as rice and wheat. In terms of growth condition monitoring, denoised images can more accurately reflect physiological indicators such as leaf area index and chlorophyll content. For yield prediction, high-quality remote sensing images can help identify yield components such as spikelet number and grain weight, improving prediction accuracy. Through this series of processing, the original high-resolution remote sensing image is transformed into a clearer and more reliable data source, providing a solid foundation for precision agriculture decision-making. This not only improves agricultural production efficiency, but also helps optimize resource utilization, reduces environmental impact and promotes sustainable agricultural development.

[0034] S102、On the basis of the preprocessed image, an edge detection algorithm is used to extract farmland boundary features, and combined with preset soil fertility distribution data, an adaptive threshold segmentation method is used to finely segment the farmland area, obtaining a segmented image containing clear farmland boundary information.

[0035] The pre-processed image is subjected to Gaussian filtering to remove high-frequency noise in the image and obtain smooth image data, providing a good data basis for subsequent edge detection. A Canny operator is used to detect the edges of the smoothed image, and by calculating the gradient amplitude and direction of the image, the candidate pixel points of the farmland boundary are determined, and a binary edge image is obtained. According to the pre-set soil fertility distribution data, the farmland area in the edge image is divided into three fertility levels, i.e. high, medium and low, to obtain a farmland fertility distribution map. For farmland areas of different fertility levels, different adaptive thresholds are set, and by dynamically adjusting the threshold size, fine segmentation of the farmland area is realized, and a segmented image containing detailed farmland boundary information is obtained. The segmented farmland area is subjected to morphological processing, and through opening and closing operations, the breakpoints and holes of the farmland boundary are eliminated, and a continuous and complete farmland boundary contour is obtained. A region growing algorithm is used to label and number the farmland area, and according to the connectivity and similarity of the pixel points, the pixel points belonging to the same farmland are merged into a region, and a labeled farmland segmentation image is obtained. Combined with the farmland boundary contour and the labeling information, the geometric features of each farmland region are extracted, including area, perimeter, shape, etc., and a farmland attribute database is constructed to provide data support for precision agriculture management.

[0036] Specifically, Gaussian filtering is a common image smoothing method that reduces high-frequency noise by weighted averaging. In agricultural remote sensing image processing, Gaussian filtering can effectively remove noise caused by sensor or atmospheric interference, providing a clear data foundation for subsequent edge detection. For example, for a remote sensing image containing multiple farmland plots, after applying Gaussian filtering, the boundaries between plots will become smoother, facilitating accurate identification of farmland contours. The Canny edge detection operator is a multi-stage edge detection technique that can effectively identify strong and weak edges in an image. In farmland boundary detection, the Canny operator first calculates the gradient magnitude and direction of the image, then determines the edge pixels through non-maximum suppression and double thresholding. This method can accurately locate the farmland boundary, even in complex terrain or under different crop growth conditions. The pre-setting of soil fertility distribution data is crucial for farmland management. By combining the edge image with fertility data, farmland areas can be divided into high, medium, and low fertility levels. This grading method helps farmers target fertilization and irrigation, improving resource utilization efficiency. For example, for a farm containing multiple plots, high-fertility areas may be suitable for growing crops that require a large amount of fertilizer, while low-fertility areas may require additional soil improvement measures. Adaptive threshold segmentation technology can dynamically adjust the threshold according to the local features of the image, which is particularly effective in processing images with uneven lighting or multiple soil types. By setting different threshold ranges for different fertility levels, farmland boundaries can be more accurately divided. For example, in a mountainous farmland with significant lighting changes, the threshold values for shaded areas and sunny areas may differ significantly. Morphological processing is an important step in image processing, used to optimize segmentation results. Opening operation can remove small protrusions, while closing operation can fill small holes. In farmland boundary processing, these operations can eliminate boundary breaks and holes caused by noise or local anomalies, resulting in a continuous and complete farmland contour. For example, for a farmland partially obscured by trees or buildings, morphological processing can help reconstruct the complete boundary. The region growing algorithm is a segmentation method based on pixel similarity, suitable for labeling and numbering adjacent farmland regions. This algorithm starts from a seed point and gradually merges similar adjacent pixels into the same region. In farmland segmentation, this method can effectively distinguish different plots, even if their boundaries are not very obvious. For example, in a large farm with multiple crops, the region growing algorithm can divide the farmland into different management units based on crop type or growth conditions. The extraction of farmland geometric features provides important data support for precision agriculture management. Area information can be used to estimate yield and develop fertilization plans, perimeter data can help plan irrigation systems, and shape features may affect the efficiency of mechanized operations. For example, by analyzing an irregularly shaped farmland, farmers can decide whether to perform land consolidation to improve farming efficiency.These geometric features combined with soil fertility data can provide comprehensive decision support for farmers to achieve precision and personalized farmland management.

[0037] S103, according to the crop boundary information extracted from the segmented image, a region growing algorithm is used to preliminarily divide each farmland region, and a pre-established crop type feature database is combined to judge the crop type of each region, so as to obtain a region image containing the crop type region division result.

[0038] According to the segmented image, the farmland boundary information is extracted, the farmland boundary coordinate point set is obtained, the region growing algorithm is used, the farmland boundary coordinate point is taken as the seed point, and the growth is expanded to the inside of the farmland until the complete farmland region is covered, so that the preliminarily divided farmland sub-region is obtained; for each farmland sub-region, the texture, color and other features thereof are extracted, and a feature vector is constructed; the farmland sub-region feature vector is matched with each crop type feature template in the pre-established crop feature database in similarity, a support vector machine classification algorithm is used to judge the crop type of the farmland sub-region; if the similarity exceeds a preset threshold, the farmland sub-region is marked as the corresponding crop type; if the similarity does not exceed the threshold, it is marked as an unknown type; the crop type marking results of all farmland sub-regions are fused to generate a region image, different colors in the image represent different crop type regions; morphological processing is performed on the region image to eliminate isolated small regions and smooth the crop region boundary, so that a final farmland crop distribution map is obtained.

[0039] Specifically, the extraction of field boundary information is a crucial step in precision agriculture management. By obtaining a set of boundary coordinate points, the location and shape of each field can be precisely determined. For example, for an irregularly shaped rice field, hundreds of coordinate points may be obtained, which, when connected, form the outline of the field. The region growing algorithm plays an important role in the division of field sub-regions. Starting from the boundary points, the algorithm gradually expands inward until it covers the entire field. Assuming there is a corn field, starting from its boundary, the algorithm will gradually include pixels of similar color and texture into the same region, eventually forming the complete corn field outline. The extraction of field sub-region features involves multiple aspects. Taking a cotton field as an example, attention may be paid to its unique texture features, such as the regularity of leaf arrangement; color features may include the green tone of cotton plants and the white color of mature cotton. These features are combined into a multi-dimensional vector for subsequent crop identification. The support vector machine (SVM) classification algorithm performs well in crop type judgment. By comparing with the pre-built crop feature database, SVM can effectively distinguish different crops. For example, although both wheat and rice appear green, SVM can accurately classify them based on their growth density, leaf shape, and other subtle differences. The similarity threshold setting is crucial to the classification result. If set too high, a large number of fields may be labeled as unknown types; if set too low, it may introduce false classifications. Usually, the best threshold is found through repeated trials, such as setting it to 0.85, which can ensure classification accuracy without producing too many unknown types. The generation of region images provides an intuitive display of crop distribution in the field. Different colors represent different crops, such as green for rice, yellow for wheat, and purple for vineyards. This visualization helps agricultural management personnel quickly understand the crop distribution in the entire region. Morphological processing plays a key role in the optimization of the final image. By eliminating isolated small regions, "noise" that may be caused by classification errors can be removed. For example, in a large corn field, there may be a few pixels that are mistakenly classified as soybeans, which will be merged into the surrounding corn region through morphological processing. Boundary smoothing processing can make the boundaries of crop regions more natural. In actual fields, the transition between crops is often gradual, rather than abrupt. Through smoothing, this natural transition can be better simulated, making the final crop distribution map of the field more realistic and providing more reliable basis for precision agriculture management.

[0040] S104、On the basis of the region image, a pre-trained support vector machine classification model is used to identify the growth stage of each region crop, and real-time humidity data collected by the field humidity sensor is combined to generate a crop image containing crop growth stage and humidity distribution information.

[0041] Step 1: Obtain a region image that has completed the division of the farmland region. Step 2: For each region in the region image, use a pre-trained support vector machine classification model to identify the crop growth stage, obtaining the crop growth stage information for each region. Step 3: Obtain the humidity data collected by the farmland humidity sensor in real time, and match the humidity data with the regions in the region image according to the sensor location, obtaining the humidity distribution information for each region. Step 4: Fuse the crop growth stage information obtained in step 2 and the humidity distribution information obtained in step 3 to generate a crop image containing the crop growth stage and humidity distribution. Step 5: Visualize the crop image in the form of a heat map to display the crop growth stage and humidity distribution, with different colors representing different growth stages and humidity ranges. Step 6: Based on the growth stage and humidity distribution information in the crop image, combine the crop growth model to determine whether the growth of crops in each region is normal, and if there are abnormal regions, mark them. Step 7: Output the marked crop image as a decision basis for farmland management, guiding agricultural production activities such as irrigation, fertilization, etc., to optimize crop growth conditions and improve crop yield and quality.

[0042] Specifically, after obtaining the region images, crop growth stage identification needs to be performed for each region. This can be achieved through a pre-trained support vector machine classification model. For example, for rice, the growth stages can be divided into seedling stage, tillering stage, heading stage, and mature stage. The model analyzes the color, texture, and other features in the image to determine the growth stage of each region. Next, the farmland humidity data needs to be obtained. This can be achieved through a network of humidity sensors placed in the farmland. Each sensor has its specific geographic location coordinates, and through these coordinates, the humidity data can be matched with the regions in the region image. For example, if a sensor is located at coordinates (100, 200), the humidity data it collects can be assigned to the region at the corresponding location in the image. By fusing the crop growth stage information and humidity distribution information, a crop image can be generated. This process can be achieved through image processing techniques. For example, different colors can be used to represent different growth stages, while the intensity of the color can be used to represent the level of humidity. For example, light green may represent a region in the seedling stage with low humidity, while dark green may represent a region in the tillering stage with high humidity. Visualizing the crop image can take the form of a heat map. A heat map is a visual data representation method that clearly shows the distribution of data. Here, red can be used to represent regions with high humidity, blue to represent regions with low humidity, and different patterns or textures to represent different growth stages. Based on the information in the crop image, combined with the crop growth model, the growth status of each region can be determined. For example, if the rice in a certain region is in the tillering stage but the humidity is significantly lower than the normal requirement for that stage, it can be determined that this region may have abnormal growth. These abnormal regions can be highlighted in the image through specific markers, such as yellow borders. Finally, the labeled crop image serves as the basis for farmland management decisions. For example, for regions with low humidity, irrigation frequency can be increased; for regions with slow growth, fertilizer application or fertilizer formulation may need to be adjusted. Through such precise farmland management, crop growth conditions can be optimized, and crop yield and quality can be improved. This data-based precision agriculture management method not only improves agricultural production efficiency but also reduces resource waste and environmental pollution, achieving sustainable development of agriculture.

[0043] S105、According to the crop growth stage information obtained from the crop image, a deep reinforcement learning algorithm is used to generate a preliminary farmland operation flight path planning scheme, combined with the performance parameters of the unmanned aerial vehicle and the energy consumption constraints.

[0044] According to the acquired crop image, the growth stage information of the crop is extracted by using an image segmentation algorithm, and the growth stage parameter of the crop is obtained. According to the acquired unmanned aerial vehicle performance parameters, including the maximum flight speed, the maximum load capacity, the maximum flight height and the like of the unmanned aerial vehicle, an unmanned aerial vehicle performance constraint condition model is established. According to the acquired unmanned aerial vehicle energy consumption parameters, including the battery capacity, the unit time energy consumption and the like of the unmanned aerial vehicle, an unmanned aerial vehicle energy consumption constraint condition model is established. The crop growth stage parameter, the unmanned aerial vehicle performance constraint condition model and the unmanned aerial vehicle energy consumption constraint condition model are taken as inputs of a deep reinforcement learning algorithm, and a preliminary farmland operation flight path is generated by using the deep reinforcement learning algorithm. According to the farmland operation task requirements, including the operation area, the operation time requirement and the like, the preliminary farmland operation flight path is optimized, and an optimized farmland operation flight path is obtained. According to the optimized farmland operation flight path, in combination with the farmland topographic information, the farmland operation flight path is processed by using a path smoothing algorithm, and a smoothed farmland operation flight path planning scheme is obtained. The smoothed farmland operation flight path planning scheme is converted into a control instruction of the unmanned aerial vehicle, and is transmitted to the unmanned aerial vehicle, so as to guide the unmanned aerial vehicle to perform the farmland operation task according to the planned flight path.

[0045] Specifically, the flight path planning for agricultural operations is a complex task that involves multiple technical fields and data sources. First, the crop growth stage information is extracted through image segmentation algorithms, which can use methods such as convolutional neural networks such as U-Net or MaskR-CNN. For example, for a rice field, different growth stages such as seedling stage, tillering stage, and heading stage can be identified and quantified as growth stage parameters. The UAV performance constraint model considers various performance indicators of the UAV. For example, a certain agricultural UAV has a maximum flight speed of 10 meters per second, a maximum payload of 10 kilograms, and a maximum flight altitude of 500 meters. These parameters will directly affect the efficiency and range of the UAV's operation. At the same time, the UAV energy consumption constraint model is also crucial. Assuming that the battery capacity of this UAV is 10000 milliampere-hours and the unit time energy consumption is 500 milliampere-hours / minute, which will determine the maximum flight time and operation area of the UAV. Deep reinforcement learning algorithms play a key role in generating the initial flight path for agricultural operations. Methods such as Q-learning or policy gradient can be used, with crop growth stage parameters, UAV performance and energy consumption constraints as state inputs, through repeated trial and error and reward mechanisms, to learn the optimal flight strategy. For example, the algorithm may learn that in the early growth stage of crops, a lower flight altitude and slower speed should be adopted to ensure accurate fertilization. According to the specific requirements of the agricultural operation task, the flight path is optimized. Assuming that a 100-acre rice field needs to be fertilized within 4 hours, the optimization algorithm will consider the payload and flight speed of the UAV, and may choose a serpentine or spiral flight path to maximize coverage and reduce repetition. Path smoothing is an important step to ensure flight safety and improve operation efficiency. Through methods such as Bezier curves or spline interpolation, sharp turns can be converted into smooth curves, reducing energy loss when the UAV turns. At the same time, combined with the information of the agricultural terrain, such as avoiding tall trees or buildings, to ensure the feasibility of the flight path. Finally, the smoothed flight path is converted into UAV control instructions. This includes a series of GPS coordinate points, flight altitudes, speeds, and operation instructions (how to turn on the fertilization device). These instructions are transmitted to the UAV through wireless communication to guide it to accurately perform the agricultural operation task. Through this series of steps, efficient and accurate flight path planning for agricultural operations can be achieved. This not only improves agricultural production efficiency, but also optimizes resource utilization and reduces environmental impact. For example, precise fertilization can reduce fertilizer runoff and reduce the risk of water pollution. At the same time, this intelligent operation method also provides farmers with more scientific field management decision support, which helps to improve crop yield and quality.

[0046] The crop growth stage parameters, UAV performance constraint model, and UAV energy consumption constraint model are input into the deep reinforcement learning algorithm to generate the initial flight path for agricultural operations.

[0047] According to the crop growth stage parameter, obtain the farmland operation task type and operation parameters required for the current crop growth stage; according to the unmanned aerial vehicle performance constraint condition model, obtain the maximum flight speed, maximum load capacity, maximum flight height and other performance parameters of the unmanned aerial vehicle; according to the unmanned aerial vehicle energy consumption constraint condition model, obtain the unmanned aerial vehicle energy consumption parameters under different flight speeds, flight heights and load capacities; take the crop growth stage parameter, farmland operation task type and operation parameter, unmanned aerial vehicle performance parameter and unmanned aerial vehicle energy consumption parameter as the input of the deep reinforcement learning algorithm; through the deep reinforcement learning algorithm, take the farmland operation coverage rate maximization, flight time minimization and energy consumption minimization as the objective function, generate a preliminary farmland operation flight path; if the farmland operation coverage rate does not reach the preset threshold, adjust the reward function weight of the deep reinforcement learning algorithm, regenerate the farmland operation flight path until the farmland operation coverage rate reaches the preset threshold; according to the generated farmland operation flight path, combine the farmland topographic and geomorphic information, smooth and optimize the flight path to obtain the final farmland operation flight path

[0048] Specifically, crop growth stage parameters form the basis for farmland operation planning, determining the required task types and specific parameters. For example, during the tillering stage of rice growth, fertilization and pest control are necessary. At this stage, a drone may need to carry a certain weight of fertilizer or pesticide and precisely spray at a low altitude and speed. The drone performance constraint model takes into account the physical limitations of the equipment. For example, a certain agricultural drone model has a maximum flight speed of 10 meters per second, a maximum payload of 10 kilograms, and a maximum flight altitude of 100 meters. These parameters directly impact operational efficiency and coverage. The energy consumption constraint model reflects the drone's endurance under different operating conditions. Generally, larger payloads and faster flight speeds result in higher energy consumption. For example, a fully loaded flight at 8 meters per second might consume 2000 mAh of power per hour, while an unladen, slow flight might consume only 1000 mAh per hour. Deep reinforcement learning algorithms optimize flight paths through trial and error. The algorithm might initially generate a simple "S"-shaped path with an 80% coverage rate. By adjusting the reward function and increasing the weight of coverage, the algorithm can generate more complex "U"-shaped paths, increasing coverage to 95% while also balancing flight time and energy consumption. Information about farmland topography is crucial for path optimization. On flat terrain, a regular grid-like path can be used. In hilly areas, a curved path based on contour lines is needed to avoid obstacles and ensure uniform spraying. For example, on a 100-mu (approximately 164 acres) terraced field, the optimized path might take on an ascending spiral, ensuring effective operation while maximizing energy conservation. This refined path planning not only improves the efficiency and accuracy of farmland operations but also significantly reduces resource waste. For example, in a 500-mu (approximately 164 acres) rice pest control operation, the optimized path can save 20% of pesticide use and 30% of flight time compared to traditional methods, reducing environmental pollution and extending the drone's operating time. This intelligent approach is gradually transforming modern agriculture, driving it towards greater efficiency and environmental friendliness.

[0049] S106. During farmland operations, farmland environmental parameter data is collected in real time through airborne sensors, and the flight attitude and track parameters of the UAV are dynamically adjusted using the Kalman filter algorithm to ensure that the acquisition quality of remote sensing images meets the needs of farmland monitoring.

[0050] The environment parameter data of the farmland is acquired by the airborne sensor, the acquired data is preprocessed to remove abnormal values and noise interference, and effective environment parameter data is obtained. According to the preprocessed environment parameter data, the flight attitude and track parameters of the unmanned aerial vehicle are combined to construct a Kalman filter model, and the flight state of the unmanned aerial vehicle is estimated and predicted. According to the estimation and prediction results of the Kalman filter model, it is judged whether the flight attitude and track parameters of the unmanned aerial vehicle meet the quality requirements of remote sensing image acquisition, and if not, dynamic adjustment is carried out. The PID control algorithm is adopted, the adjustment amount of the flight attitude and track parameters of the unmanned aerial vehicle is calculated according to the estimation and prediction results of the Kalman filter model, and the flight state of the unmanned aerial vehicle is controlled and optimized in real time. The high-definition camera on board is used to collect remote sensing images of the farmland under the optimized flight attitude and track parameters, and high-quality farmland image data is obtained. An image stitching algorithm is used to stitch the collected multiple remote sensing images to obtain a complete farmland panoramic image, which provides data support for farmland monitoring. According to the stitched farmland panoramic image, the crop growth model and the expert knowledge base are combined to analyze and evaluate the growth, diseases and pests of the farmland, and provide decision basis for precise management of the farmland.

[0051] Specifically, the acquisition and preprocessing of farmland environmental parameter data are the foundation of precision agriculture. Airborne sensors can collect data such as temperature, humidity, and light, and remove outliers through methods such as median filtering and eliminate noise interference through techniques such as wavelet transform to ensure data quality. For example, the original data collected by the temperature sensor may contain sudden values, and the sliding window method can be used to remove abnormal points to obtain a smooth temperature curve. The Kalman filter model plays an important role in estimating the flight state of the UAV. This model combines environmental parameters and UAV attitude information to predict the flight state at the next time. For example, when the UAV is affected by crosswind, Kalman filtering can estimate the change in yaw angle, providing a basis for subsequent attitude adjustment. The quality of remote sensing image acquisition is closely related to the flight state of the UAV. If the predicted flight height exceeds the preset range, the system will trigger a dynamic adjustment mechanism. The PID control algorithm calculates the adjustment amount based on the error, such as when the height is too high, reducing the throttle output to make the UAV descend to the appropriate height. The high-definition camera collects farmland images under the optimized flight state. For example, in sunny weather, the UAV maintains a height of 50 meters and a speed of 5 meters per second, and takes a 4K resolution image every 2 seconds to ensure image clarity and overlap. Image stitching is a key step in constructing a panoramic view of the farmland. Feature point matching algorithms such as SIFT can identify common areas between adjacent images, perspective transformation can align images, and fusion algorithms can eliminate seam marks. A 100-acre farmland may require hundreds of aerial images to be stitched together to generate a high-resolution orthographic image. Farmland monitoring and analysis use the stitched panoramic view, combined with crop growth models and expert knowledge. For example, through the calculation of the vegetation index NDVI, the growth of crops can be evaluated; using deep learning models such as convolutional neural networks, early symptoms of pests and diseases can be identified. These analysis results provide decision support for farmland management, such as determining the best time to fertilize or local pest control measures. The entire process embodies the technical chain of precision agriculture, from data collection, processing to analysis and decision-making, each link is closely connected. Through this method, farmers can timely grasp the farmland conditions and achieve precision management, improving yield and resource utilization efficiency. At the same time, this technology can also help monitor large-scale farmland, providing important basis for regional agricultural planning and food security.

[0052] PID control algorithm is adopted to calculate the adjustment amount of UAV flight attitude and track parameters based on the estimation and prediction results of Kalman filter model, and to realize real-time control and optimization of UAV flight state.

[0053] According to the flight state data of the unmanned aerial vehicle, the Kalman filtering algorithm is used to estimate and predict the attitude and position of the unmanned aerial vehicle, and the current flight state parameters of the unmanned aerial vehicle are obtained. The flight state parameters of the unmanned aerial vehicle estimated and predicted by the Kalman filtering model are compared with the preset target flight state parameters, and the deviation amount of the attitude and track of the unmanned aerial vehicle is calculated. According to the calculated attitude and track deviation amount of the unmanned aerial vehicle, the PID control algorithm is used to calculate the adjustment amount of the attitude and track parameters of the unmanned aerial vehicle, and the flight control instruction of the unmanned aerial vehicle is obtained. The flight control instruction of the unmanned aerial vehicle calculated by the PID control algorithm is transmitted to the flight control system of the unmanned aerial vehicle, and the flight attitude and track of the unmanned aerial vehicle are adjusted and controlled in real time. In the process of flight of the unmanned aerial vehicle, the flight state data of the unmanned aerial vehicle is continuously obtained and transmitted to the Kalman filtering model for real-time estimation and prediction, and the current flight state parameters of the unmanned aerial vehicle are obtained. It is judged whether the current flight state parameters of the unmanned aerial vehicle meet the preset flight performance index, if not, the adjustment amount is calculated according to the deviation amount by the PID control algorithm, and the adjustment instruction is transmitted to the flight control system of the unmanned aerial vehicle for real-time adjustment. Through the process of continuous iteration of Kalman filtering estimation and prediction, PID control algorithm calculation of adjustment amount, and flight control system execution of adjustment instruction, the real-time optimization control of the flight attitude and track of the unmanned aerial vehicle is realized, and the flight state of the unmanned aerial vehicle is ensured to meet the preset performance index requirements.

[0054] Specifically, the Kalman filter algorithm is a recursive estimation method used to estimate the state of a dynamic system. In UAV flight control, this algorithm can effectively fuse data from various sensors such as GPS, gyroscopes, and accelerometers to obtain more accurate flight state estimates. For example, when a UAV is flying over a farmland, GPS signals may become unstable due to tree obstruction, at this time Kalman filter can combine other sensor data to make more reliable estimates of the UAV's position. In the prediction step, Kalman filter uses the state and control input at the last time to predict the state at the current time. For example, based on the position, velocity and acceleration of the UAV at the last second, the position at the next second can be predicted. This prediction takes into account the dynamics model of the UAV, making the state estimation more accurate. Comparing the estimated result with the target state is a key step in the control system. Assuming the target height of the UAV is 50 meters, and the current height estimated by Kalman filter is 48 meters, then the height deviation is 2 meters. This deviation will be used as input to the PID controller. PID control algorithm is a classic feedback control method, composed of proportional, integral and derivative parts. In UAV control, PID can calculate appropriate control amount according to the deviation. For example, for a height deviation of 2 meters, the PID controller may output an instruction to increase the thrust by 10% to make the UAV rise to the target height. The execution of the control instruction involves the actuator of the UAV, such as the motor and the rudder. When receiving the instruction to increase the thrust by 10%, the flight control system of the UAV will adjust the motor speed accordingly, so that the propeller generates more lift. The real-time adjustment process is a continuous closed-loop control. For example, in the task of monitoring farmland, the UAV needs to maintain a stable flight height and speed to obtain high-quality images. If suddenly encountering an upward airflow, the UAV may deviate from the planned route. At this time, Kalman filter will quickly detect this deviation, and PID controller will calculate the correction instruction, such as reducing the thrust or adjusting the attitude, to offset the influence of the airflow. Through this cycle of continuous state estimation, deviation calculation and control adjustment, the UAV can maintain stable flight in complex farmland environment. This not only ensures the quality of remote sensing data collection, but also improves flight safety. For example, when there are tall trees on the edge of the farmland, the UAV can accurately adjust the height and heading to avoid collision risk. The advantage of this control method lies in its adaptability and robustness. No matter facing wind changes, load changes or sensor noise, the system can quickly respond and make appropriate adjustments. This is particularly important for agricultural UAVs, which often need to work in different weather conditions and terrain environments, thus providing reliable data support for precision agriculture.

[0055] Example 2:

[0056] The application also provides an unmanned aerial vehicle-based agricultural machinery operation planning system, comprising:

[0057] A first processing device is used to obtain a pre-processed image based on a high-resolution remote sensing image of a target farmland area;

[0058] The second processing device is used to obtain a segmented image containing clear farmland boundary information based on the preprocessed image;

[0059] a third processing device for obtaining, based on the segmented image, a region image containing a result of the crop type region division;

[0060] a fourth processing device for generating a crop image containing crop growth stage and moisture distribution information based on the regional image;

[0061] a fifth processing device for generating a preliminary flight path planning scheme for farmland operations based on the crop images;

[0062] The sixth processing device is used to dynamically adjust the flight attitude and track parameters of the UAV according to the preliminary farmland operation flight path planning scheme.

[0063] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A method for planning agricultural machinery operations based on drones, characterized in that: include: Step S1, obtaining a pre-processed image based on a high-resolution remote sensing image of a target farmland area; Step S2: obtaining a segmented image containing clear farmland boundary information based on the preprocessed image; Step S3: obtaining a regional image including the crop type region division result based on the segmented image; Step S4: generating a crop image including crop growth stage and moisture distribution information based on the regional image; Step S5: Generate a preliminary farmland operation flight path planning scheme based on the crop image; Step S6: Dynamically adjust the flight attitude and track parameters of the UAV according to the preliminary farmland operation flight path planning scheme.

2. The agricultural machinery operation planning method based on UAV according to claim 1, characterized in that: In step S1, a pre-built convolutional neural network model is used to perform denoising preprocessing on the remote sensing image to obtain a preprocessed image.

3. The agricultural machinery operation planning method based on UAV according to claim 1, characterized in that: In step S2, based on the preprocessed image, an edge detection algorithm is used to extract the farmland boundary features, and combined with the preset soil fertility distribution data, the farmland area is finely segmented through an adaptive threshold segmentation method to obtain a segmented image containing clear farmland boundary information.

4. The agricultural machinery operation planning method based on UAV according to claim 3, characterized in that: In step S3, the farmland areas are preliminarily divided using a region growing algorithm based on the farmland boundary information extracted from the segmented image, and the crop type of each area is determined in combination with a pre-established crop type feature database to obtain a regional image containing the crop type area division results.

5. The agricultural machinery operation planning method based on UAV according to claim 4, characterized in that: In step S4, the growth stage of crops in each region is identified based on the regional image using a pre-trained support vector machine classification model, and combined with the real-time humidity data collected by the farmland humidity sensor to generate a crop image containing crop growth stage and humidity distribution information.

6. The agricultural machinery operation planning method based on UAV according to claim 5, characterized in that: In step S5, based on the crop growth stage information obtained from the crop image, a deep reinforcement learning algorithm is used, combined with the performance parameters and energy consumption constraints of the UAV, to generate a preliminary flight path planning scheme for farmland operations.

7. The agricultural machinery operation planning method based on UAV according to claim 6, characterized in that: In step S6, the Kalman filter algorithm is used to dynamically adjust the flight attitude and track parameters of the UAV.

8. An agricultural machinery operation planning system based on drones, characterized in that: include: A first processing device is used to obtain a pre-processed image based on a high-resolution remote sensing image of a target farmland area; The second processing device is used to obtain a segmented image containing clear farmland boundary information based on the preprocessed image; a third processing device for obtaining, based on the segmented image, a region image containing a result of the crop type region division; a fourth processing device for generating a crop image containing crop growth stage and moisture distribution information based on the regional image; a fifth processing device for generating a preliminary flight path planning scheme for farmland operations based on the crop images; The sixth processing device is used to dynamically adjust the flight attitude and track parameters of the UAV according to the preliminary farmland operation flight path planning scheme.

Images