Method, system and device for identifying and listing small areas of field blocks in original orthophoto images

Abstract

The application discloses a field original orthographic image field plot area identification and listing method, system and device, relates to the field plot area identification technical field, and comprises the following steps: identifying field plot area instances in a field original orthographic image according to a deep learning algorithm to obtain a field plot area instance list; performing filtering operation on the field plot area instance list; performing completion operation on the field plot area instance list after the filtering operation; determining the orthographic image, orientation attribute and information attribute of the field plot area instance according to the field plot area instance list after the completion operation, and performing listing fusion on the orthographic image, orientation attribute and information attribute to obtain fusion information of the field plot area instance. The application can automatically and efficiently identify, filter and list a large number of heterogeneous field plot areas in a field original orthographic image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of field plot identification, in particular to a field plot identification and listing method, system and device in a field original orthographic image. BACKGROUND

[0002] The field planning for the purpose of modern breeding has the characteristics of small size, large quantity and large phenotypic difference between field plots. In recent years, the field high-throughput image acquisition method based on unmanned aerial vehicle imaging platform has developed rapidly. However, due to the complexity of field design and planning and the great difference in characteristics of field plots, in order to realize the purpose of field plot information matching and boundary coordinate extraction, it is still necessary to manually frame each field plot according to the artificial judgment result. Obviously, this step has become the bottleneck step in the field analysis process. SUMMARY

[0003] The purpose of the present application is to provide a field plot identification and listing method, system and device in a field original orthographic image, which can automatically and efficiently identify, filter and list a large number of heterogeneous field plots in a field original orthographic image.

[0004] To achieve the above purpose, the present application provides the following scheme:

[0005] In a first aspect, the present application provides a field plot identification and listing method in a field original orthographic image, comprising:

[0006] According to a deep learning algorithm, identifying field plot instances in a field original orthographic image to obtain a field plot instance list;

[0007] Filtering the field plot instance list;

[0008] Completing the field plot instance list after filtering;

[0009] According to the field plot instance list after completion, determining the orthographic image, orientation attribute and information attribute of the field plot instance, and fusing the orthographic image, orientation attribute and information attribute to obtain the fusion information of the field plot instance.

[0010] In a second aspect, the present application provides a field plot identification and listing system in a field original orthographic image, comprising:

[0011] A field plot instance identification module is configured to identify field plot instances in a field original orthographic image according to a deep learning algorithm to obtain a field plot instance list;

[0012] A filtering module is configured to filter the field plot instance list;

[0013] complementing the field plot instance list after filtering operation;

[0014] The fusion information determining module is configured to determine the orthographic image, azimuth attribute and information attribute of the field plot instance according to the field plot instance list after the complementing operation, and compile and fuse the orthographic image, azimuth attribute and information attribute to obtain the fusion information of the field plot instance.

[0015] In a third aspect, the present application provides an electronic device comprising a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to perform the field plot instance identification and compilation method in the raw orthographic image of a large field according to the first aspect.

[0016] According to the embodiments of the present application, the following technical effects are provided:

[0017] The present application realizes the accurate identification and automatic matching of the orthographic image, azimuth attribute and information attribute of the field plot instance in the raw orthographic image of a large field by using the automatic operation mode to identify, filter, complement, extract the boundary points and fuse the information attribute of the field plot instance. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0019] Figure 1 The flowchart of the field plot instance identification and compilation method in the raw orthographic image of a large field provided by the embodiments of the present application;

[0020] Figure 2 The detailed process diagram of the field plot instance identification and compilation method in the raw orthographic image of a large field provided by the embodiments of the present application;

[0021] Figure 3 The process diagram of the field plot instance identification and compilation method in the raw orthographic image of a large field provided by the embodiments of the present application;

[0022] Figure 4 The field design information diagram provided by the embodiments of the present application;

[0023] Figure 5 The field plot information diagram provided by the embodiments of the present application;

[0024] Figure 6This is a schematic diagram of the structure of the field plot identification and arrangement system in the original orthophoto of the field provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Example 1

[0028] like Figure 1 As shown, this embodiment of the invention provides a method for identifying and arranging field plots in original field orthophotos, including:

[0029] Step 101: Based on the deep learning algorithm, identify the field plot instances in the original orthophoto of the field to obtain a list of field plot instances.

[0030] Step 102: Filter the list of field plot instances.

[0031] Step 103: Complete the list of field plot instances after the filtering operation.

[0032] Step 104: Based on the list of field plot instances after the completion operation, determine the orthophoto, orientation attribute, and information attribute of the field plot instances, and compile and fuse the orthophoto, orientation attribute, and information attribute to obtain the fused information of the field plot instances.

[0033] In a preferred embodiment, step 101 specifically includes:

[0034] (1) Using the optimal cropping size, cropping is performed on the x-axis and y-axis of the original Daejeon orthophoto to obtain multiple square Daejeon cropped orthophotos; among them, every two adjacent square Daejeon cropped orthophotos have the same pixel size overlap.

[0035] (2) The Mask R-CNN field cell recognition deep learning model that has been built is used to identify the field cell instances in each square field cropped orthophoto image, and the boundary coordinates of the field cell instances corresponding to each square field cropped orthophoto image are obtained.

[0036] (3) Based on the boundary coordinates of the field plot instances in each square field cropped orthophoto, the field plot instances are merged to obtain a list of field plot instances.

[0037] The process for determining the optimal cutting size is as follows:

[0038] 1) Randomly select several random pixels in the original orthophoto of Daejeon, and crop square images with any random side length around the random pixels in turn to obtain multiple cropped images; wherein, the side length of each pair of adjacent cropped images is different.

[0039] 2) Use the already constructed Mask R-CNN deep learning model for field cell recognition to perform field cell instance recognition on each of the cropped images to obtain the field cell instance recognition results.

[0040] 3) Extract the boundary coordinates of the field plot instances from the field plot instance recognition results, and calculate the maximum horizontal and vertical dimensions of each field plot instance based on the boundary coordinates of the field plot instances;

[0041] 4) Based on the maximum horizontal and vertical dimensions of all field plot instances, determine the maximum horizontal and vertical dimensions with the highest frequency.

[0042] 5) Determine the optimal cutting size based on the most frequent maximum horizontal dimension and the most frequent maximum vertical dimension.

[0043] Furthermore, based on the boundary coordinates of the field plot instances in each square field cropped orthophoto image, the field plot instances are merged to obtain a list of field plot instances, specifically including:

[0044] First, the boundary coordinates of all field plot instances in each large square field cropped orthophoto are extracted. Then, based on the boundary coordinates, the overlapping area between each pair of field plot instances in any large square field cropped orthophoto is calculated. Finally, based on the overlapping area, two field plot instances that meet the first condition are merged to obtain the merged field plot instances in each large square field cropped orthophoto.

[0045] Secondly, based on the optimal cropping size and overlap, the overlapping cross regions of the adjacent 2×2 target square field cropped orthophotos are extracted, and the field plot instances are merged in units of overlapping cross regions to obtain a list of field plot instances; the target square field cropped orthophoto is the square field cropped orthophoto after merging the field plot instances.

[0046] The first condition is: (A) m ∩A n ) / Am >α|(A m ∩A n ) / A n >α.

[0047] Among them, A m and A n Let be the m-th plot instance region and the n-th plot predicted instance region, respectively, where α is the overlap rate hyperparameter; | represents OR.

[0048] Furthermore, the field cell instances are merged based on overlapping cross regions to obtain a list of field cell instances, specifically including:

[0049] First, the target field cell instances are determined; the target field cell instances are those that overlap with the overlapping cross area; second, the target field cell instances that meet the second condition are merged to obtain a list of field cell instances.

[0050] The second condition is:

[0051] (A p ∩C i,j )∩(A q ∩C i,j ) / (A p ∩C i,j )>β|(A p ∩C i,j )∩(A q ∩C i,j ) / (A q ∩C i,j )>β;

[0052] Among them, A p and A q These represent the p-th plot instance region and the q-th plot predicted instance region, respectively; C i,j The i-th and j-th overlapping cross regions on the x-axis and y-axis are defined as the target square cropped orthophoto image, where β is the hyperparameter of overlap rate; | represents or.

[0053] As a preferred implementation, step 102 specifically includes: filtering the list of field plot instances based on the area size of the field plot instances, the distance between field plot instances, or the distribution of the long and short sides of the field plot instances.

[0054] As a preferred embodiment, step 103 specifically includes:

[0055] Based on the bounding box coordinates of the field plot instances and the gridding operation of the center point of the field plot instances, the missing positions of the field plot instances in the filtered field plot instance list are filled in.

[0056] As a preferred implementation, based on the list of field plot instances after the completion operation, the orthophoto, orientation attribute, and information attribute of the field plot instances are determined, specifically including:

[0057] (1) Based on the list of field plot instances after completion operation, determine the bounding box pixel coordinates and center point pixel coordinates of each field plot instance; (2) Based on the bounding box pixel coordinates of the field plot instances and the latitude and longitude range of the original orthophoto of the field, calculate the bounding box latitude and longitude coordinates of each field plot instance; (3) Based on the bounding box latitude and longitude coordinates of the field plot instances and the original orthophoto of the field, determine the orthophoto of each field plot; (4) Pack the center point pixel coordinates, bounding box pixel coordinates, bounding box latitude and longitude coordinates and orthophoto of each field plot instance and arrange them in row and column order to obtain the field information table; the orientation attributes include the center point pixel coordinates, bounding box pixel coordinates and bounding box latitude and longitude coordinates; (5) Match the field information table with the field design table to determine the information attributes of each field plot.

[0058] Example 2

[0059] This invention provides a method for identifying and arranging field plots in original field orthophotos, such as... Figure 2 and Figure 3 As shown, the table in the Daejeon design is as follows: Figure 4 As shown in the table, the field information table is as follows: Figure 5 As shown, the method specifically includes:

[0060] 1. Predict and compile examples of field plots.

[0061] 1.1 Estimation of optimal cutting size for field plots.

[0062] Several random pixels are randomly selected from the original orthophoto of Daejeon. Using these random pixels as centers, square images with side lengths of the following values ​​are successively cropped to obtain multiple cropped images: S = {1024, 2048, 3072, 4096, 5120, 6144, 7168, 8192, 9216, 10240}. S represents the set of cropped side lengths.

[0063] The cropped images were uniformly downsampled to 1024×1024, and the Mask R-CNN deep learning model for field cell recognition was used to automatically identify field cell instances in each cropped image, obtaining the recognition results. Then, the boundary coordinates of the field cell instances in the recognition results were extracted. Combining the Ground Sample Distance (GSD) and the number of pixels occupied by the field cell boundary, the maximum horizontal and vertical dimensions of each field cell instance were calculated. Finally, the recognition results of all cropped images were integrated to obtain the maximum horizontal dimension array W. max and the vertical maximum size array H max Next, based on the horizontal maximum size array W... max and the vertical maximum size array H max Create a histogram and calculate the maximum horizontal dimension W of the most frequent occurrence. max,est The highest frequency of the maximum longitudinal dimension W max,est The final calculated optimal cutting size S crop Enough to make the value |S–max{W max,est W max,est The minimum value is determined by substituting any side length from the set of cutting side lengths, checking the calculated value for each side length, and then determining the side length corresponding to the minimum calculated value as the optimal cutting size.

[0064] 1.2 Cropping of the original orthophoto of Daejeon.

[0065] Because the file size of the uncompressed Daejeon raw orthophoto is large, the memory of a typical computer cannot meet the computational processing requirements. Therefore, the Daejeon raw orthophoto is cropped to an optimal cropping size S with both the long and wide sides fixed in pixels. crop Square cropped orthophotos of Daejeon. Cropping is performed along both the x and y axes of the original Daejeon orthophoto, with each pair of adjacent square cropped orthophotos having a consistent pixel overlap. The number of cropping operations (N) along the x and y axes. x and N y It is calculated by the following formula:

[0066]

[0067]

[0068] Among them, W Orthomosaic and H Orthomosaic The length and width of the original orthophoto of Daejeon are read from the image's accompanying information; S crop O and O represent the side length (i.e., the optimal cropping size) and overlap of the square field cropped orthophoto image, respectively.

[0069] In the coordinate system of the original Daejeon orthophoto, the cropping ranges of the square Daejeon cropped orthophoto along the x and y axes are as follows:

[0070] X min,i =(n x,i -1) / (S crop -O*S crop ).

[0071] X max,i =min{n x,i ×(S crop –O)+O,W Orthomosaic}

[0072] Y min,j =(n y,j -1) / (S crop -O*S crop ).

[0073] Y max,j =min{n y,j ×(S crop –O)+O,H Orthomosaic}

[0074] Where, n x,i and n y,j These represent the i-th and j-th overlapping regions on the x-axis and y-axis, respectively, of the square cropped orthophoto image. The parameter selection of O is adjusted according to different implementation methods, generally set to the optimal cropping size S. crop 25%.

[0075] The range of the i-th cropped orthophoto on the X-axis is X. min,i To X max,i , that is, X min,i It is the minimum value of the i-th cropped orthophoto on the X-axis.

[0076] 1.3 Field Plot Instance Recognition

[0077] The Mask R-CNN deep learning model for field cell identification, which has already been built, is used to identify field cell instances in each square cropped orthophoto image of a large field. A binary prediction image is generated for each field cell instance in the image. To facilitate data storage and subsequent processing, the findcontours function of the cv2 module is used to extract the boundary coordinates of the predicted field cell instance corresponding to each square cropped orthophoto image of a large field.

[0078] 1.4 Merging of Field Block Instances

[0079] Based on the boundary coordinates of the field plot instances in each square field cropped orthophoto, the coordinates are transformed from the cropped image coordinate system to the original field original orthophoto coordinate system, and the field plot instances are merged.

[0080] Step 1: Extract the boundary coordinates of all field plot instances from a single large square cropped orthophoto image. Calculate the overlap area between each pair of field plot instances. If the following conditions are met, merge the two field plot instances:

[0081] (A m ∩A n ) / A m >α|(A m ∩A n ) / A n >α.

[0082] Among them, A m and A n These represent the m-th plot instance region and the n-th plot predicted instance region, respectively. α is the overlap rate hyperparameter, with a default value of 0.9; | indicates OR.

[0083] Step 2: Based on the optimal cutting size S crop With an overlap of 0, the overlapping cross region of the adjacent 2x2 square cropped orthophotos of the field is extracted; the coordinates of the boundary points of the overlapping cross region are: [(X min,i Y min,j+1 ), (X min,i+1 Y min,j+1 ), (X min,i+1， Y min,j ), (X max,i Y min,j ), (X max,i Y min,j+1 ), (X max,i+1 Y min,j+1 ), (X max,i+1 Y max,j ), (X max,i Y max,j ), (X max,i Y max,j+1 ), (X min,i Y max,j+1 ), (X min,i+1 Y max,j ), (X min,i+1 Y max,j ), (X min,i Y min,j+1 )).

[0084] Among them, X min,i and X max,iThe coordinate range of the i-th overlapping region on the x-axis represents the square Daejeon cropped orthophoto image. min,j and Y max,j The coordinate range of the j-th overlapping region on the y-axis represents the square Daejeon cropped orthophoto image.

[0085] Step 3: Merge the field plot instances using the overlapping cross regions of the orthophoto image cropped from each 2x2 square field as units. Filter field plot instances that overlap with the overlapping cross regions, calculate the overlap area between each field plot instance and the overlapping cross regions, as well as the pairwise overlap area. If the following conditions are met, merge the two field plot instances:

[0086] (A p ∩C i,j )∩(A q ∩C i,j ) / (A p ∩C i,j )>β|(A p ∩C i,j )∩(A q ∩C i,j ) / (A q ∩C i,j )>β.

[0087] Among them, A p and A q These represent the instance region of the p-th field and the predicted instance region of the q-th field, respectively. C i,j represents the i-th and j-th overlapping cross regions on the x and y axes of a square cropped orthophoto image, β is the overlap rate hyperparameter, which defaults to 0.6; | represents OR.

[0088] After the above operations, a list of consolidated and merged field plot instances is obtained.

[0089] 2. Filtering the list of field plot instances

[0090] For field plots that have identified false positives, filters are needed to remove them. Filtering methods include, but are not limited to, the appropriate filter can be selected for different fields.

[0091] 2.1 Based on the size of the plot area: Since the area of ​​plots within the same large field area is generally consistent, plots that are too large or too small can be removed. Specifically, Grubbs' Test is used to detect and filter outliers in the area of ​​all plots.

[0092] G = max|S i -S mean | / σ.

[0093] Where G represents the Grubbs value of the area of ​​the plot instance, S i Let S be the area of ​​the i-th plot instance. mean σ and σ' are the mean and standard deviation of the area of ​​all field plot instances, respectively. Outliers are removed if the following formula is satisfied, and the field plot instance corresponding to that outlier is then deleted.

[0094]

[0095] Where N is the sample size, and ∝ is the confidence level. For N-2 degrees of freedom, the significance level is the critical value of the t-distribution with ∝ / 2N.

[0096] 2.2 Based on the distance between field plot instances: When designing large field areas, fields are usually arranged closely together. Therefore, if the distance between a field plot instance and its adjacent field plot instance is significantly different from other field plot instances, it can be removed.

[0097] The distance D between the i-th and j-th plot instances is... i,j It is derived from the following formula.

[0098] D i,j =sqrt((x i,center -x j,center ) 2 +(y i,center -y j,center ) 2 ).

[0099] Where, x i,center and y i,center Let x be the x-coordinate and y-coordinate of the center point of the i-th field plot instance. j,center and y j,center Let x and y be the center coordinates of the j-th field plot instance.

[0100] Sort the distances in the distance list (the distances between each plot of land instance and every other plot of land instance) in ascending order, then select the k smallest distances and average them to obtain M. i The default value for k is 5.

[0101] M i =mean([D i,min1 D i,min2 D i,min3 D i,min4 ,…D i,mink ]).

[0102] Use Grubbs' Test on M iThe formula for performing outlier detection and filtering, and then deleting the field instance corresponding to the outlier, has been given above.

[0103] 2.3 Based on the distribution of long and short sides of field cell instances: The bounding box width and length of each field cell instance are calculated. Regardless of the design shape of the field cell, the side lengths of its bounding box in the horizontal and vertical directions in the image, i.e., the bounding box width and length, generally conform to a normal distribution (using the ShapiroWilk test). The bounding box width and length are then input as two variables into the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering algorithm to remove noise points in sparse regions that do not belong to any cluster.

[0104] 3. Complete the list of examples of plots in the field.

[0105] Calculate the center point coordinates (x, y) of each field plot instance. center ,y center Then, the distance between the center points of each pair of field plot instances is calculated; the center distance between the i-th field plot instance and the j-th field plot instance is:

[0106] D i,j =sqrt((x i,center -x j,center ) 2 +(y i,center -y j,center ) 2 ).

[0107] Sort to find the minimum center distance D i,j We obtain the pairing of field cell instances with the smallest center distance, denoted as i and k. We then calculate the angle between the centers of the two field cell instances:

[0108] Making β i Histogram and calculate the median β of the interval with the highest frequency. mean Rotate the coordinates of all field plot instances by β. mean The angle is adjusted to ensure that the field plot instances are arranged as horizontally and vertically as possible. After the completion step, the coordinates of all field plot instances are rotated by -β. mean Angle, restore the original coordinates.

[0109] Calculate the bounding box of each plot instance and obtain the median width w of the plot instance. median and length median h median .

[0110] 3.1 Based on the bounding box coordinates of the field plot instance.

[0111] Read the field design table and obtain the number of rows n. row With the number of fields n col .

[0112] The interval I of the field plot instances on the x and y axes x With interval I y The calculation formula is:

[0113] I x =(x max -x min -W median ) / n col .

[0114] I y =(y max -y min -H median ) / n row .

[0115] Where, x min ,x max ,y min ,y max The extreme values ​​of all field plot instances in the original orthophoto coordinate system of the field are given.

[0116] Therefore, the center point (X) of the field plot instance in the i-th column and j-th row is calculated. i Y j Complete the list of missing locations for field plot instances:

[0117]

[0118]

[0119] 3.2 Gridding of the center point of the field plot example.

[0120] Read the field design table and obtain the number of rows n. row With the number of fields n col .

[0121] Based on the center coordinates of the field plot instances, the x-coordinate distribution array {x} is obtained. center} and the number of y-coordinate distributions {y center}

[0122] Based on Euclidean distance calculation, Ward's minimum variance method is used to analyze the x-coordinate distribution array {x}. center Perform cluster analysis and divide into n col Classes, each denoted as Col according to its x-coordinate in ascending order. i , i∈[1,n colSimilarly, the distribution number of the y-coordinate {y center} divided into n row Classes, each denoted as Row according to its y-coordinate in ascending order. j j∈[1,n row Therefore, each plot instance possesses (Col). i Row j )Directional attribute.

[0123] For missing field cell instances (Col) m Row n ), calculate its coordinates (X) one by one according to the following method. i ,Y j First, filter out all instances belonging to Row from the list of field plot instances. n A collection of instances of the class {N}.

[0124] Solve the linear regression equation under least squares estimation:

[0125]

[0126]

[0127] Where x i and y i Let the x and y coordinates of the i-th field plot instance in the set {N} be given, and the equation of the straight line is obtained as follows:

[0128] Similarly, in the list of field instances, all instances belonging to Col are filtered out. m Instance {M} of class, repeat the above steps, and obtain the equation of the straight line as follows:

[0129] Solve the system of equations: Find the coordinates (X) of the missing field cell instance. i ,Y j Complete the supplementary list of instances for missing locations in the field plots.

[0130] 3.3 Manual revision.

[0131] After obtaining the preliminary completion information according to the above steps, print the complete orthophoto and all field plot instances in the interactive interface. If misidentified instances are found, the user should remove them from the list of field plot instances. If missed instances are found, the user should click on the location of the missed field plot instance in the complete orthophoto to create a new instance and obtain the center point coordinates (X) of the missed field plot instance. i ,Y j Complete the supplementary list of instances for missing locations in the field plots.

[0132] 4. Matching of field plot orthophotos, orientation attributes, and information attributes.

[0133] 4.1 Add field location attributes.

[0134] Calculate the bounding box pixel coordinates of the field cell instance.

[0135]

[0136] Based on the bounding box pixel coordinates of the field plot instances and the latitude and longitude range of the original orthophoto of the field, the bounding box latitude and longitude coordinates of the field plot instances are converted, and the center point pixel coordinates, bounding box pixel coordinates, and latitude and longitude coordinates of each field plot instance are packaged and arranged in row and column order.

[0137] 4.2 Field image cropping.

[0138] Based on the latitude and longitude coordinates of the bounding boxes of the above field plot examples, the original orthophoto of the field is cropped to obtain the orthophoto of each field plot and arranged in row and column order.

[0139] 4.3 Add field information attributes.

[0140] Since the field information table and the field design table have the same rows and columns, a direct association and matching between the two tables is established to create an index. Simultaneously, the row and column information of the field design table, along with the information attributes of the field plots, such as variety, planting time, experimental treatment gradient, water and fertilizer environment, is further indexed with the field information table, enabling the addition of information attributes to the field plot image data. The field information table is obtained by packaging the center point pixel coordinates, bounding box pixel coordinates, bounding box latitude and longitude coordinates, and orthophoto of each field plot instance and arranging them in row and column order.

[0141] Example 3

[0142] In order to implement the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, a field plot identification and arrangement system in original field orthophotos is provided below.

[0143] like Figure 6 As shown in the figure, an embodiment of the present invention provides a field plot identification and arrangement system in original field orthophotos, comprising:

[0144] The field plot instance identification module 601 is used to identify field plot instances in the original orthophoto of the field using a deep learning algorithm, and obtain a list of field plot instances.

[0145] The filtering operation module 602 is used to perform filtering operations on the list of field plot instances.

[0146] The completion module 603 is used to complete the list of field plot instances after the filtering operation.

[0147] The fusion information determination module 604 is used to determine the orthophoto, orientation attribute and information attribute of the field plot instance based on the list of field plot instances after the completion operation, and to compile and fuse the orthophoto, orientation attribute and information attribute to obtain the fusion information of the field plot instance.

[0148] Example 4

[0149] This invention provides an electronic device including a memory and a processor. The memory stores a computer program, and the processor runs the computer program to enable the electronic device to perform the field cell identification and arrangement method in the original orthophoto of a field as described in Embodiment 1.

[0150] Alternatively, the aforementioned electronic device may be a server.

[0151] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for identifying and arranging field plots in the original orthophotos of a field as described in Embodiment 1.

[0152] Compared with existing technologies, this invention can significantly reduce the time required to collect image data of target crops in fields, improve the efficiency of integrating and archiving massive amounts of field images and data, and provide a systematic data analysis process for crop varieties in both time and space dimensions.

[0153] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0154] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for identifying and listing field plot small areas in a field raw orthographic image, characterized in that, The application comprises the following steps: According to the deep learning algorithm, the field block plot instance in the original orthographic image of the field is identified to obtain the field block plot instance list; According to the deep learning algorithm, the field block plot instance in the original orthographic image of the field is identified to obtain the field block plot instance list, specifically comprising: adopting the optimal cutting size, cutting in the x-axis and y-axis two dimensions of the original orthographic image of the field to obtain a plurality of square field cutting orthographic images; wherein each two adjacent square field cutting orthographic images have a consistent pixel size overlap; using the Mask R-CNN field block plot identification deep learning model which has been constructed to identify the field block plot instance in each square field cutting orthographic image to obtain the boundary coordinates of the field block plot instance corresponding to each square field cutting orthographic image; based on the boundary coordinates of the field block plot instance in each square field cutting orthographic image, the field block plot instance is merged to obtain the field block plot instance list; the determination process of the optimal cutting size is: randomly selecting a plurality of random pixel points in the original orthographic image of the field, and cutting a square image with any random value of side length as the center of the plurality of random pixel points in turn to obtain a plurality of cutting pictures; wherein the length of each two adjacent cutting pictures is different; using the Mask R-CNN field block plot identification deep learning model which has been constructed to identify the field block plot instance in each cutting picture to obtain the field block plot instance identification result; extracting the boundary coordinates of the field block plot instance in the field block plot instance identification result, and calculating the horizontal maximum size and the vertical maximum size of each field block plot instance according to the boundary coordinates of the field block plot instance; according to the horizontal maximum size and the vertical maximum size of all field block plot instances, the highest frequency horizontal maximum size and the highest frequency vertical maximum size are determined; according to the highest frequency horizontal maximum size and the highest frequency vertical maximum size, the optimal cutting size is determined; Filtering operation is performed on the field block plot instance list; Completing operation is performed on the field block plot instance list after filtering operation; According to the field block plot instance list after completing operation, the orthographic image, orientation attribute and information attribute of the field block plot instance are determined, and the orthographic image, orientation attribute and information attribute are listed and fused to obtain the fusion information of the field block plot instance.

2. The method for identifying and arranging field plots in original field orthophotos according to claim 1, characterized in that, Based on the boundary coordinates of the field block plot instance in each square field cutting orthographic image, the field block plot instance is merged to obtain the field block plot instance list, specifically comprising: extracting the boundary coordinates of all field block plot instances in each large square field cutting orthographic image, and calculating the overlapping area between each two field block plot instances in any large square field cutting orthographic image based on the boundary coordinates, then based on the overlapping area, two field block plot instances meeting the first condition are merged to obtain the merged field block plot instance in each large square field cutting orthographic image; Based on the optimal cutting size and the degree of overlap, an overlapping cross region of a 2x2 target square field cutting orthographic image adjacent to the overlapping cross region is extracted, and the field plot parcel instances are merged in units of the overlapping cross region to obtain a field plot parcel instance list; the target square field cutting orthographic image is a square field cutting orthographic image after the field plot parcel instances are merged; The first condition is: (A m ∩A n ) / A m > α | (A m ∩A n ) / A n > α; wherein A m and A n are the mth field plot instance region and the nth field prediction instance region, respectively, and α is an overlap rate hyperparameter, and | indicates or.

3. The method for identifying and arranging field plots in original orthophotos of a field as described in claim 2, characterized in that, The field plot parcel instances are merged in units of the overlapping cross region to obtain a field plot parcel instance list, specifically including: Determining a target field plot parcel instance; the target field plot parcel instance is a field plot parcel instance that overlaps with the overlapping cross region; Merging the target field plot parcel instances that meet the second condition to obtain a field plot parcel instance list; The second condition is: (A p ∩C i , j )∩(A q ∩C i , j ) / (A p ∩C i , j ) > β | (A p ∩C i , j )∩(A q ∩C i , j ) / (A q ∩C i , j )> β; wherein A p and A q are the pth field plot instance region and the qth field prediction instance region, respectively; C i , j is the ith overlapping cross region and jth overlapping cross region of the target square field cut orthographic image on the x-axis and y-axis, and β is an overlap rate hyperparameter.

4. The method for identifying and arranging field plots in original orthophotos of a field as described in claim 1, characterized in that, Filtering the field plot parcel instance list, specifically including: Filtering the field plot parcel instance list according to the field plot parcel instance area size, the distance between the field plot parcel instances, or the long and short side distribution of the field plot parcel instances.

5. The method of claim 1, wherein the field original orthoimage is divided into a plurality of small areas, and the small areas are identified and listed according to the field block. Completing the field plot parcel instance list after the filtering operation, specifically including: Completing the missing positions of the field plot parcel instances in the field plot parcel instance list after the filtering operation according to the field plot parcel instance bounding box coordinates and the field plot parcel instance center point grid operation.

6. The method of claim 1, wherein the field original orthoimage is divided into a plurality of small areas, and the small areas are identified and listed according to the field block. According to the field plot parcel instance list after the completion operation, determining the orthographic image, orientation attribute and information attribute of the field plot parcel instance, specifically including: According to the field plot parcel instance list after the completion operation, determining the bounding box pixel coordinates and center point pixel coordinates of each field plot parcel instance; According to the bounding box pixel coordinates of the field plot parcel instance and the latitude and longitude range of the field plot original orthographic image, calculating the bounding box latitude and longitude coordinates of each field plot parcel instance; According to the bounding box latitude and longitude coordinates of the field plot parcel instance and the field plot original orthographic image, determining the orthographic image of each field plot parcel; Packing the center point pixel coordinates, bounding box pixel coordinates, bounding box latitude and longitude coordinates and orthographic image of each field plot parcel instance, and listing them in row and column order to obtain a field plot information table; the orientation attribute includes center point pixel coordinates, bounding box pixel coordinates and bounding box latitude and longitude coordinates; Associating and matching the field plot information table with the field plot design table to determine the information attribute of each field plot parcel.

7. A system for identifying and arranging field plots in original orthophotos of a field, characterized in that, Including: A field plot parcel instance identification module is configured to identify field plot parcel instances in a field plot original orthographic image according to a deep learning algorithm to obtain a field plot parcel instance list; According to the deep learning algorithm, the field plot instance in the original orthographic image of the field is recognized to obtain a field plot instance list, specifically including: adopting the optimal cutting size, cutting in the x-axis and y-axis two dimensions of the original orthographic image of the field to obtain a plurality of square field cutting orthographic images; wherein, each two adjacent square field cutting orthographic images have a consistent pixel size overlap; using the Mask R-CNN field plot recognition deep learning model which has been constructed to recognize the field plot instance in each square field cutting orthographic image to obtain the boundary coordinates of the field plot instance corresponding to each square field cutting orthographic image; based on the boundary coordinates of the field plot instance in each square field cutting orthographic image, the field plot instance is merged to obtain the field plot instance list; the determination process of the optimal cutting size is: randomly selecting a plurality of random pixel points in the original orthographic image of the field, and cutting a square image with any random value of side length as the center of the plurality of random pixel points in turn to obtain a plurality of cutting pictures; wherein, the side length of each two adjacent cutting pictures is different; using the Mask R-CNN field plot recognition deep learning model which has been constructed to recognize the field plot instance of each cutting picture to obtain the field plot instance recognition result; extracting the boundary coordinates of the field plot instance in the field plot instance recognition result, and calculating the horizontal maximum size and the vertical maximum size of each field plot instance according to the boundary coordinates of the field plot instance; according to the horizontal maximum size and the vertical maximum size of all field plot instances, the highest frequency horizontal maximum size and the highest frequency vertical maximum size are determined; according to the highest frequency horizontal maximum size and the highest frequency vertical maximum size, the optimal cutting size is determined; The filtering operation module is used for filtering the field plot instance list; The completion operation module is used for completing the field plot instance list after filtering operation; The fusion information determination module is used for determining the orthographic image, orientation attribute and information attribute of the field plot instance according to the field plot instance list after completion operation, and compiling and fusing the orthographic image, orientation attribute and information attribute to obtain the fusion information of the field plot instance.

8. An electronic device, comprising: The electronic device includes a memory and a processor, the memory is used for storing a computer program, and the processor runs the computer program to make the electronic device execute the field plot recognition and compilation method in the original orthographic image of the field according to any one of claims 1-6.

Images