Crop whole-period yield estimation method and system based on remote sensing satellite and unmanned aerial vehicle

Abstract

The invention relates to the technical field of crop yield estimation, and particularly discloses a crop whole-period yield estimation method and system based on a remote sensing satellite and an unmanned aerial vehicle, and the method comprises the steps: obtaining a remote sensing image of a crop region, determining a collection path based on the remote sensing image, and obtaining an unmanned aerial vehicle image based on the collection path; identifying the unmanned aerial vehicle image, predicting the crop yield, and synchronously calculating the crop yield variation; environment information is acquired and recorded, and a neural network model between the environment information and crop yield variation is trained; and verifying the error rate of the neural network model according to the recognition result of the unmanned aerial vehicle image, and adjusting the use frequency of the unmanned aerial vehicle according to the error rate. The remote sensing image is firstly subjected to coarse recognition, then the collection path of the unmanned aerial vehicle is determined, the unmanned aerial vehicle image is obtained, and fine recognition is performed; in the fine recognition process, the rough recognition process of the remote sensing image is optimized, the application frequency of the unmanned aerial vehicle is reduced as much as possible, and the recognition efficiency is very high while the precision is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of crop yield estimation, and in particular to a method and system for crop full-period yield estimation based on remote sensing satellites and unmanned aerial vehicles. Background Art

[0002] Crop yield estimation is the prediction of crop yield before harvest. Its purpose is to provide yield information as early as possible before harvest, as a basis for formulating plans for harvesting, warehousing, transportation, marketing, processing, etc.

[0003] Most of the existing crop yield estimation methods use manual estimation, and both the data collection process and the estimation process are completed manually. However, the crop area is outdoor and the conditions are relatively harsh. In some cases, there are certain risks. Therefore, the data collection process based on remote sensing images and drones is gradually becoming popular.

[0004] The acquisition process of remote sensing images is easy, but the accuracy is not particularly high. The accuracy of drone images is high, but the acquisition process is difficult. How to further improve the intelligence level of the data collection process in the crop yield estimation task is the technical problem that the technical solution of the present invention wants to solve. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for estimating crop yield throughout the entire period based on remote sensing satellites and unmanned aerial vehicles, so as to solve the problems raised in the above-mentioned background technology.

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

[0007] A method for estimating crop yields during the entire crop period based on remote sensing satellites and unmanned aerial vehicles, the method comprising:

[0008] Acquire the region boundary of the crop region, and acquire the zoning information within the region boundary; the zoning information includes the contour and the region type;

[0009] A remote sensing image of a crop area is obtained, a collection path is determined based on the remote sensing image, and a drone is used to obtain drone images along the collection path; in the process of obtaining the drone image, the drone performs local processing on the obtained video, and feeds back a panoramic image after processing, which is called a drone image; the local processing process includes image merging and image screening; the number of the drone is at least one;

[0010] Identify the drone image, predict crop yield, and simultaneously calculate the crop yield change;

[0011] Obtain and record environmental information, and train a neural network model between environmental information and crop yield changes;

[0012] The error rate of the neural network model is verified according to the recognition result of the drone image, and the use frequency of the drone is adjusted according to the error rate; the use frequency is proportional to the error rate.

[0013] As a further solution of the present invention, the step of acquiring remote sensing images of the crop area, determining an acquisition path based on the remote sensing images, and acquiring drone images along the acquisition path by the drone includes:

[0014] Acquire a remote sensing image of the crop area, partition the remote sensing image according to the contours in the zoning information, and calibrate each sub-area obtained by the partitioning according to the area type in the zoning information;

[0015] Determine the number of path points in each sub-area according to the area type, and randomly select path points in the sub-area;

[0016] Determine the collection path of the sub-area based on the selected points, and connect the endpoints of the collection paths of adjacent sub-areas as the final path;

[0017] The final path is sent to the drone, which acquires the drone image.

[0018] As a further solution of the present invention: the process of the drone acquiring the drone image is:

[0019] The drone takes video in real time and records the position and time, and converts the video into an image sequence; the images in the image sequence contain position tags and time tags;

[0020] Query the location tag of the image and classify the image according to the location tag;

[0021] Query the time tags in each type of image and sort the images according to the time tags;

[0022] Selecting and combining images from various types of images according to a preset time point to obtain a combined image; wherein the selection rule is that the time difference between the time label of the image and the time point is the smallest;

[0023] The combined image is screened to obtain a drone image.

[0024] As a further solution of the present invention: the step of screening the combined image to obtain the drone image includes:

[0025] Perform frequency domain conversion on the combined image to obtain a feature image;

[0026] Extract the characteristic value in the characteristic image as the dependent variable, obtain the time point corresponding to the combined image as the independent variable, and establish the curve;

[0027] Derivative the change curve, and select a smooth segment according to the derivative result; the proportion of the number of points in the smooth segment whose absolute value of the derivative is less than a preset value is greater than a preset proportion;

[0028] A time point is selected in each smooth segment, and the combined image at the time point is read as the UAV image.

[0029] As a further solution of the present invention: the steps of identifying the drone image, predicting the crop yield, and synchronously calculating the crop yield change include:

[0030] Reading a convolution kernel from a pre-established convolution kernel table and traversing the matching drone image; the convolution kernel table includes a time period item, a convolution kernel item, and a yield item; the time period represents a growth period, and the convolution kernel represents an image feature corresponding to a certain yield;

[0031] When the match is successful, the corresponding yield is read as the predicted crop yield;

[0032] Calculate the difference between crop yield and the previous crop yield to get the change in crop yield.

[0033] As a further solution of the present invention: the step of verifying the error rate of the neural network model according to the recognition result of the drone image and adjusting the use frequency of the drone according to the error rate includes:

[0034] The recognition results of the drone images are taken as the true results;

[0035] Based on the real results, calculate the error rate of the results of the neural network model;

[0036] adjusting the frequency of use of the drone based on the error rate;

[0037] The relationship between usage frequency and error rate is:

[0038] In the formula, f is the operating frequency, F is the preset parameter, called the limit frequency, α is the preset correction coefficient, σ is the error rate, ΔJ is the difference between the true result and the result of the neural network model, and J is the true result.

[0039] The technical solution of the present invention also provides a crop yield estimation system based on remote sensing satellites and unmanned aerial vehicles, the system comprising:

[0040] A zoning information acquisition module is used to acquire the regional boundary of the crop area and acquire the zoning information within the regional boundary; the zoning information is the outline and the region type;

[0041] The data acquisition module is used to acquire remote sensing images of the crop area, determine an acquisition path based on the remote sensing images, and acquire drone images along the acquisition path by the drone; in the process of acquiring the drone images, the drone performs local processing on the acquired video, and feeds back a panoramic image after processing, which is called a drone image; the local processing process includes image merging and image screening; the number of drones is at least one;

[0042] A data recognition module is used to recognize the drone image, predict crop yield, and simultaneously calculate the crop yield change;

[0043] A model training module is used to obtain and record environmental information and train a neural network model between environmental information and crop yield changes;

[0044] The frequency update module is used to verify the error rate of the neural network model according to the recognition result of the drone image, and adjust the use frequency of the drone according to the error rate; the use frequency is proportional to the error rate.

[0045] As a further solution of the present invention: the data acquisition module includes:

[0046] An image partitioning unit is used to obtain a remote sensing image of a crop area, partition the remote sensing image according to the contours in the zoning information, and calibrate each sub-area obtained by the partitioning according to the area type in the zoning information;

[0047] A point selection unit, used to determine the number of path points in each sub-area according to the area type, and randomly select path points in the sub-area;

[0048] A path connection unit, used to determine the collection path of the sub-area based on the selected points, and connect the endpoints of the collection paths of adjacent sub-areas as the final path;

[0049] The image acquisition unit is used to send the final path to the drone, and the drone acquires the drone image.

[0050] As a further solution of the present invention: the process of the drone acquiring the drone image is:

[0051] The drone takes video in real time and records the position and time, and converts the video into an image sequence; the images in the image sequence contain position tags and time tags;

[0052] Query the location tag of the image and classify the image according to the location tag;

[0053] Query the time tags in each type of image and sort the images according to the time tags;

[0054] Selecting and combining images from various types of images according to a preset time point to obtain a combined image; wherein the selection rule is that the time difference between the time label of the image and the time point is the smallest;

[0055] The combined image is screened to obtain a drone image.

[0056] As a further solution of the present invention: the combined image is screened to obtain the contents of the drone image including:

[0057] Perform frequency domain conversion on the combined image to obtain a feature image;

[0058] Extract the characteristic value in the characteristic image as the dependent variable, obtain the time point corresponding to the combined image as the independent variable, and establish the curve;

[0059] Derivative the change curve, and select a smooth segment according to the derivative result; the proportion of the number of points in the smooth segment whose absolute value of the derivative is less than a preset value is greater than a preset proportion;

[0060] A time point is selected in each smooth segment, and the combined image at the time point is read as the UAV image.

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention first performs coarse recognition through remote sensing images, and then determines the collection path of the drone, obtains the drone image, and performs fine recognition; in the fine recognition process, the coarse recognition process of the remote sensing image is optimized to minimize the application frequency of the drone, while ensuring accuracy and very high recognition efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0063] Figure 1 This is a flowchart of the crop yield estimation method based on remote sensing satellites and drones.

[0064] Figure 2 This is a block diagram of the composition of the crop full-season yield estimation system based on remote sensing satellites and drones. DETAILED DESCRIPTION

[0065] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0066] Figure 1The flowchart of the method for estimating crop yield in the whole period based on remote sensing satellites and unmanned aerial vehicles is as follows. In an embodiment of the present invention, a method for estimating crop yield in the whole period based on remote sensing satellites and unmanned aerial vehicles comprises:

[0067] Step S100: obtaining a region boundary of a crop region, and obtaining zoning information within the region boundary; the zoning information includes an outline and a region type;

[0068] A crop area is an area where crop yield estimation is required. The boundaries of the area are known data. Zoning information is the planning information of each location within the crop area. It is public information and can be obtained with the help of map services. Zoning information is used to represent which crop is planted at which location. It should be noted that there are situations where crops are not planted in the zoning information, which is used to represent all locations where crops are not planted.

[0069] Step S200: acquiring a remote sensing image of a crop area, determining a collection path based on the remote sensing image, and acquiring a drone image along the collection path by a drone; in the process of acquiring the drone image, the drone performs local processing on the acquired video, and feeds back a panoramic image after processing, which is called a drone image; the local processing process includes image merging and image screening; the number of the drone is at least one;

[0070] By acquiring remote sensing images of the crop area and identifying the remote sensing images, the acquisition path can be determined. Under normal permissions, the accuracy of remote sensing images that ordinary people can obtain is very low. Therefore, the remote sensing images in this application are only used to determine the working path of the drone, that is, the above-mentioned acquisition path. After the acquisition path is determined, the acquisition path is sent to the drone. The drone will move along the acquisition path, continuously acquire video, and finally feed back drone images.

[0071] The video processing of the drone is handled locally by the drone. This method seems to increase the energy consumption of the drone, but in fact, it reduces more energy consumption of data transmission.

[0072] Step S300: identifying the drone image, predicting crop yield, and synchronously calculating the crop yield change;

[0073] After receiving the drone image feedback from the drone, the drone image is recognized. The recognition method can adopt the convolution recognition process. The convolution recognition process is to first obtain an image with a known yield, and then extract features from the image. The extracted image features are used as the convolution kernel. During recognition, the convolution kernel is regarded as known data and can be used directly to predict crop yields. This method is a yield prediction process based on actual images and has high accuracy.

[0074] Step S400: Acquire and record environmental information, and train a neural network model between the environmental information and crop yield variation;

[0075] Environmental information includes soil composition and weather conditions, which are factors that affect crop yields. There is a corresponding relationship between environmental information and crop yields. More precisely, environmental information affects the change in crop yields. Crop yields are a single predicted data. Whenever environmental information changes, the predicted crop yield will change. Therefore, constructing a corresponding relationship between environmental information and the change in crop yields is more in line with the actual situation.

[0076] Step S500: verifying the error rate of the neural network model according to the recognition result of the drone image, and adjusting the use frequency of the drone according to the error rate; the use frequency is proportional to the error rate;

[0077] In the actual application stage, the recognition result of the drone image is obtained, and the recognition result of the drone image is taken as an accurate result, and then the error rate of the neural network model is verified. The lower the error rate, the higher the accuracy of the neural network model. At this time, the application frequency of the neural network model will be higher, and correspondingly, the use frequency of the drone will be lower; the use frequency of the drone is used to adjust step S200, that is, the application frequency of step 200; in addition, the number of drones involved in step S200 is generally not unique. In order to ensure that as much data as possible is obtained, multiple drones will be set to obtain drone images. The collection paths of multiple drones are the same, and the working process is also the same.

[0078] Regarding step S200, the step of acquiring remote sensing images of the crop area, determining an acquisition path based on the remote sensing images, and acquiring drone images along the acquisition path by the drone includes:

[0079] Acquire a remote sensing image of the crop area, partition the remote sensing image according to the contours in the zoning information, and calibrate each sub-area obtained by the partitioning according to the area type in the zoning information;

[0080] Determine the number of path points in each sub-area according to the area type, and randomly select path points in the sub-area;

[0081] Determine the collection path of the sub-area based on the selected points, and connect the endpoints of the collection paths of adjacent sub-areas as the final path;

[0082] The final path is sent to the drone, which acquires the drone image.

[0083] With the help of existing satellite services to obtain remote sensing images of crop areas, combined with known zoning information, the remote sensing images can be partitioned, and each sub-area (determined by the outline) and the area type of the sub-area can be determined in the remote sensing image; the area type is used to characterize what kind of crops are planted, and different crops have different importance. Therefore, the amount of detection resources required is different. The higher the importance, the more detection resources need to be invested, and the more path points need to be set in the corresponding sub-area. The more path points there are, the longer the acquisition path is (under normal circumstances). Specifically, in order to ensure the randomness of the acquisition path, a random sampling scheme is provided. The present invention randomly selects a number of path points in the sub-area, and uses these selected points as waypoints to generate a path as the acquisition path of the sub-area; finally, the acquisition paths of adjacent sub-areas are connected end to end to obtain a path passing through all sub-areas, which is called the final path; the final path is sent to the drone, and the drone can acquire the drone image.

[0084] It should be noted that the number of drones is not unique, and once the collection path is determined, all drones use the same collection path.

[0085] Among them, the process of drone acquiring drone images is:

[0086] The drone takes video in real time and records the position and time, and converts the video into an image sequence; the images in the image sequence contain position tags and time tags;

[0087] Query the location tag of the image and classify the image according to the location tag;

[0088] Query the time tags in each type of image and sort the images according to the time tags;

[0089] Selecting and combining images from various types of images according to a preset time point to obtain a combined image; wherein the selection rule is that the time difference between the time label of the image and the time point is the smallest;

[0090] The combined image is screened to obtain a drone image.

[0091] The video taken by the drone contains audio, but the audio is considered to have no analysis value in the technical solution of the present invention. Therefore, only the image information is retained, that is, the video is converted into an image sequence; the images in the image sequence contain location tags and time tags. This is because the range of a single shot is difficult to completely cover the entire area, so multiple images need to be spliced.

[0092] Specifically, the splicing process is described as follows:

[0093] The images are classified according to the locations where they were taken, and for each category of images, the images are sorted according to the time when they were acquired, so that each category of images arranged in time order can be obtained.

[0094] Then, the administrator pre-sets a time interval, such as 0.5s. Multiple time points can be selected based on the set time interval. With the time point as the center, an image can be selected from each type of image. The images can be combined according to the corresponding position tags of these images to obtain a combined image.

[0095] The obtained combined images correspond to time points. For crop areas, the variation is extremely small. Therefore, most of the combined images are similar. They are screened and the retained combined images are called drone images, which are then fed back to the execution platform of this method.

[0096] The step of screening the combined image to obtain the drone image includes:

[0097] Perform frequency domain conversion on the combined image to obtain a feature image;

[0098] Extract the characteristic value in the characteristic image as the dependent variable, obtain the time point corresponding to the combined image as the independent variable, and establish the curve;

[0099] The variation curve is derived, and a smooth segment is selected according to the derivative result; the proportion of the number of points in the smooth segment whose absolute value of the derivative is less than a preset value is greater than a preset proportion;

[0100] A time point is selected in each smooth segment, and the combined image at the time point is read as the UAV image.

[0101] The above content limits the screening scheme, the principle of which is to first perform frequency domain conversion on the image, then extract the eigenvalues, and then eliminate duplicate images based on the eigenvalues. The retained images can be used as drone images.

[0102] Specifically, after the image is converted in the frequency domain, the important content in the image (pixels whose color values ​​are close to the average color value) is concentrated in the upper left corner of the image. Therefore, by intercepting the feature image at the upper left endpoint, a small image that can reflect the overall parameters of the feature image can be obtained. The mean of the small image can be calculated as the eigenvalue; each panoramic image corresponds to an eigenvalue, and all eigenvalues ​​are counted to obtain a change curve.

[0103] The variation curve is differentiated and divided into multiple smooth segments according to the size of the derivative. A smooth segment represents a period of time in which the image changes very little. At this time, the image analysis is more meaningful. An image is selected in each smooth segment as the drone image.

[0104] It is worth mentioning that when extracting features from the image after frequency domain conversion, the size of the small image is determined by the staff, that is, the acquisition step length in the above content

[0105] Regarding step S300, the steps of identifying the drone image, predicting crop yield, and synchronously calculating the crop yield change include:

[0106] Reading a convolution kernel from a pre-established convolution kernel table and traversing the matching drone image; the convolution kernel table includes a time period item, a convolution kernel item, and a yield item; the time period represents a growth period, and the convolution kernel represents an image feature corresponding to a certain yield;

[0107] When the match is successful, the corresponding yield is read as the predicted crop yield;

[0108] Calculate the difference between crop yield and the previous crop yield to get the change in crop yield.

[0109] The staff will pre-count samples and obtain image features of different periods corresponding to different crop yields as the convolution kernel table. In the recognition stage, they will query the current growth period, read the corresponding convolution kernel, and perform matching recognition on the drone image. The corresponding crop yield of the convolution kernel that matches successfully is the predicted yield. The difference between the crop yield and the previous crop yield is calculated to obtain the change in crop yield.

[0110] Regarding step S500, the step of verifying the error rate of the neural network model according to the recognition result of the drone image and adjusting the use frequency of the drone according to the error rate includes:

[0111] The recognition results of the drone images are taken as the true results;

[0112] Based on the real results, calculate the error rate of the results of the neural network model;

[0113] adjusting the frequency of use of the drone based on the error rate;

[0114] The relationship between usage frequency and error rate is:

[0115] In the formula, f is the operating frequency, F is the preset parameter, called the limit frequency, α is the preset correction coefficient, σ is the error rate, ΔJ is the difference between the true result and the result of the neural network model, and J is the true result.

[0116] The result of the drone image recognition based on the convolution kernel is considered to be an accurate result, that is, the real result in the above content. The error rate of the result of the neural network model is calculated based on the real result, and then the use frequency of the drone is adjusted according to the error rate; the use frequency has a preset limit frequency. When the error rate is calculated, if the error rate is small, the limit frequency is also small. If the error rate is large enough and exceeds 100%, then the limit frequency is used, where the value range of α is also zero to one.

[0117] Figure 2 The present invention is a block diagram of a crop yield estimation system based on remote sensing satellites and unmanned aerial vehicles. In an embodiment of the present invention, a crop yield estimation system based on remote sensing satellites and unmanned aerial vehicles, the system 10 includes:

[0118] The zoning information acquisition module 11 is used to acquire the region boundary of the crop region and acquire the zoning information within the region boundary; the zoning information is the outline and the region type;

[0119] The data acquisition module 12 is used to acquire remote sensing images of the crop area, determine an acquisition path based on the remote sensing images, and acquire drone images along the acquisition path by the drone; in the process of acquiring the drone images, the drone performs local processing on the acquired video, and feeds back a panoramic image after processing, which is called a drone image; the local processing process includes image merging and image screening; the number of drones is at least one;

[0120] A data recognition module 13 is used to recognize the drone image, predict crop yield, and simultaneously calculate the crop yield change;

[0121] A model training module 14 is used to obtain and record environmental information and train a neural network model between the environmental information and the crop yield change;

[0122] The frequency updating module 15 is used to verify the error rate of the neural network model according to the recognition result of the drone image, and adjust the use frequency of the drone according to the error rate; the use frequency is proportional to the error rate.

[0123] Furthermore, the data acquisition module 12 includes:

[0124] An image partitioning unit is used to obtain a remote sensing image of a crop area, partition the remote sensing image according to the contours in the zoning information, and calibrate each sub-area obtained by the partitioning according to the area type in the zoning information;

[0125] A point selection unit, used to determine the number of path points in each sub-area according to the area type, and randomly select path points in the sub-area;

[0126] A path connection unit, used to determine the collection path of the sub-area based on the selected points, and connect the endpoints of the collection paths of adjacent sub-areas as the final path;

[0127] The image acquisition unit is used to send the final path to the drone, and the drone acquires the drone image.

[0128] Specifically, the process of drones acquiring drone images is as follows:

[0129] The drone takes video in real time and records the position and time, and converts the video into an image sequence; the images in the image sequence contain position tags and time tags;

[0130] Query the location tag of the image and classify the image according to the location tag;

[0131] Query the time tags in each type of image and sort the images according to the time tags;

[0132] Selecting and combining images from various types of images according to a preset time point to obtain a combined image; wherein the selection rule is that the time difference between the time label of the image and the time point is the smallest;

[0133] The combined image is screened to obtain a drone image.

[0134] The combined image is screened to obtain the contents of the drone image, including:

[0135] Perform frequency domain conversion on the combined image to obtain a feature image;

[0136] Extract the characteristic value in the characteristic image as the dependent variable, obtain the time point corresponding to the combined image as the independent variable, and establish the curve;

[0137] The variation curve is derived, and a smooth segment is selected according to the derivative result; the proportion of the number of points in the smooth segment whose absolute value of the derivative is less than a preset value is greater than a preset proportion;

[0138] A time point is selected in each smooth segment, and the combined image at the time point is read as the UAV image.

[0139] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for estimating crop yields during the entire crop period based on remote sensing satellites and unmanned aerial vehicles, characterized in that: The method comprises: Acquire the region boundary of the crop region, and acquire the zoning information within the region boundary; the zoning information includes the contour and the region type; A remote sensing image of a crop area is obtained, a collection path is determined based on the remote sensing image, and a drone is used to obtain drone images along the collection path; in the process of obtaining the drone image, the drone performs local processing on the obtained video, and feeds back a panoramic image after processing, which is called a drone image; the local processing process includes image merging and image screening; the number of the drone is at least one; Identify the drone image, predict crop yield, and simultaneously calculate the crop yield change; Obtain and record environmental information, and train a neural network model between environmental information and crop yield changes; The error rate of the neural network model is verified according to the recognition result of the drone image, and the use frequency of the drone is adjusted according to the error rate; the use frequency is proportional to the error rate.

2. The method for estimating crop yields during the entire crop period based on remote sensing satellites and unmanned aerial vehicles according to claim 1, characterized in that: The step of acquiring a remote sensing image of the crop area, determining a collection path based on the remote sensing image, and acquiring the drone image along the collection path by the drone comprises: Acquire a remote sensing image of the crop area, partition the remote sensing image according to the contours in the zoning information, and calibrate each sub-area obtained by the partitioning according to the area type in the zoning information; Determine the number of path points in each sub-area according to the area type, and randomly select path points in the sub-area; Determine the collection path of the sub-area based on the selected points, and connect the endpoints of the collection paths of adjacent sub-areas as the final path; The final path is sent to the drone, which acquires the drone image.

3. The method for estimating crop yields during the entire crop period based on remote sensing satellites and unmanned aerial vehicles according to claim 2, characterized in that: The process of drone acquiring drone images is: The drone takes video in real time and records the position and time, and converts the video into an image sequence; the images in the image sequence contain position tags and time tags; Query the location tag of the image and classify the image according to the location tag; Query the time tags in each type of image and sort the images according to the time tags; Selecting and combining images from various types of images according to a preset time point to obtain a combined image; wherein the selection rule is that the time difference between the time label of the image and the time point is the smallest; The combined image is screened to obtain a drone image.

4. The method for estimating crop yields during the entire crop period based on remote sensing satellites and unmanned aerial vehicles according to claim 3, characterized in that: The step of screening the combined image to obtain the drone image includes: Perform frequency domain conversion on the combined image to obtain a feature image; Extract the characteristic value in the characteristic image as the dependent variable, obtain the time point corresponding to the combined image as the independent variable, and establish the curve; The variation curve is derived, and a smooth segment is selected according to the derivative result; the proportion of the number of points in the smooth segment whose absolute value of the derivative is less than a preset value is greater than a preset proportion; A time point is selected in each smooth segment, and the combined image at the time point is read as the UAV image.

5. The method for estimating crop yields during the entire crop period based on remote sensing satellites and unmanned aerial vehicles according to claim 1, characterized in that: The steps of identifying the drone image, predicting crop yield, and synchronously calculating the crop yield change include: Reading a convolution kernel from a pre-established convolution kernel table and traversing the matching drone image; the convolution kernel table includes a time period item, a convolution kernel item, and a yield item; the time period represents a growth period, and the convolution kernel represents an image feature corresponding to a certain yield; When the match is successful, the corresponding yield is read as the predicted crop yield; Calculate the difference between crop yield and the previous crop yield to get the change in crop yield.

6. The method for estimating crop yields during the entire crop period based on remote sensing satellites and unmanned aerial vehicles according to claim 1, characterized in that: The step of verifying the error rate of the neural network model according to the recognition result of the drone image and adjusting the use frequency of the drone according to the error rate includes: The recognition results of the drone images are taken as the true results; Based on the real results, calculate the error rate of the results of the neural network model; adjusting the frequency of use of the drone based on the error rate; The relationship between usage frequency and error rate is: In the formula, f is the operating frequency, F is the preset parameter, called the limit frequency, α is the preset correction coefficient, σ is the error rate, ΔJ is the difference between the true result and the result of the neural network model, and J is the true result.

7. A crop yield estimation system based on remote sensing satellites and drones, characterized in that: The system comprises: A zoning information acquisition module is used to acquire the regional boundary of the crop area and acquire the zoning information within the regional boundary; the zoning information is the outline and the region type; The data acquisition module is used to acquire remote sensing images of the crop area, determine an acquisition path based on the remote sensing images, and acquire drone images along the acquisition path by the drone; in the process of acquiring the drone images, the drone performs local processing on the acquired video, and feeds back a panoramic image after processing, which is called a drone image; the local processing process includes image merging and image screening; the number of drones is at least one; A data recognition module is used to recognize the drone image, predict crop yield, and simultaneously calculate the crop yield change; A model training module is used to obtain and record environmental information and train a neural network model between environmental information and crop yield changes; The frequency update module is used to verify the error rate of the neural network model according to the recognition result of the drone image, and adjust the use frequency of the drone according to the error rate; the use frequency is proportional to the error rate.

8. The crop yield estimation system based on remote sensing satellites and unmanned aerial vehicles according to claim 7 is characterized in that: The data acquisition module comprises: An image partitioning unit is used to obtain a remote sensing image of a crop area, partition the remote sensing image according to the contours in the zoning information, and calibrate each sub-area obtained by the partitioning according to the area type in the zoning information; A point selection unit, used to determine the number of path points in each sub-area according to the area type, and randomly select path points in the sub-area; A path connection unit, used to determine the collection path of the sub-area based on the selected points, and connect the endpoints of the collection paths of adjacent sub-areas as the final path; The image acquisition unit is used to send the final path to the drone, and the drone acquires the drone image.

9. The crop yield estimation system based on remote sensing satellites and unmanned aerial vehicles according to claim 8, characterized in that: The process of drone acquiring drone images is: The drone takes video in real time and records the position and time, and converts the video into an image sequence; the images in the image sequence contain position tags and time tags; Query the location tag of the image and classify the image according to the location tag; Query the time tags in each type of image and sort the images according to the time tags; Selecting and combining images from various types of images according to a preset time point to obtain a combined image; wherein the selection rule is that the time difference between the time label of the image and the time point is the smallest; The combined image is screened to obtain a drone image.

10. The crop yield estimation system based on remote sensing satellites and unmanned aerial vehicles according to claim 9, characterized in that: The combined image is screened to obtain the contents of the drone image including: Perform frequency domain conversion on the combined image to obtain a feature image; Extract the characteristic value in the characteristic image as the dependent variable, obtain the time point corresponding to the combined image as the independent variable, and establish the curve; The variation curve is derived, and a smooth segment is selected according to the derivative result; the proportion of the number of points in the smooth segment whose absolute value of the derivative is less than a preset value is greater than a preset proportion; A time point is selected in each smooth segment, and the combined image at the time point is read as the UAV image.

Images