A remote sensing image intelligent auxiliary labeling method and device

Abstract

This application discloses an intelligent assisted annotation method and apparatus for remote sensing images, relating to the field of image annotation technology. The method includes: constructing a total cost function based on a preprocessed remote sensing image; treating each pixel of the remote sensing image as a node in a connected weighted graph, and determining the cost value of weighted edges between adjacent pixels based on the total cost function; wherein the total cost function is configured such that the cost value of weighted edges between pixels located on the edges of features in the remote sensing image is less than a preset cost value; and drawing the shortest path between adjacent marker points based on marker points sequentially input by the operator along the edges of features in the remote sensing image, forming a preliminary outline. This method can improve the annotation efficiency of annotators, save manpower and resources, and effectively solve the problem of decreased annotation accuracy due to annotator fatigue.

Description

Technical Field

[0001] This application relates to the field of image annotation technology, and in particular to a method and apparatus for intelligent auxiliary annotation of remote sensing images. Background Technology

[0002] Image annotation is fundamental to the research and development of computer vision products. Whether it's facial recognition, drone reconnaissance, medical diagnosis, or autonomous driving, image annotation plays an irreplaceable role. Image annotation is the process of attaching labels to images. In image annotation, a single label can be assigned to the entire image, or multiple labels can be assigned to each group of pixels. These labels are predetermined by engineers and selected to provide information about the objects displayed in the image for the computer vision model. In the field of geographic information, after obtaining remote sensing images, they need to be segmented to separate the various regions present in the remote sensing image and then labeled accordingly. The annotation results provide a reference for remote sensing mapping, improving its accuracy. This process requires a large amount of labeled data.

[0003] Currently, image segmentation data annotation methods on the market include manual annotation and intelligent annotation. Manual annotation requires annotators to read and recognize the image, then use a mouse to continuously click along the target in the image to outline the edge contour of the object, ultimately making a correct annotation. This traditional manual annotation method is inefficient and requires a large amount of manpower, severely impacting work efficiency. Furthermore, after prolonged annotation work, annotators may experience fatigue, leading to a decrease in annotation accuracy. While intelligent annotation overcomes the shortcomings of manual annotation, it still requires a large amount of labeled data for training before being implemented, relying on manual annotation to provide training samples. The quality of these training samples directly affects the accuracy of intelligent annotation. Summary of the Invention

[0004] This application provides an intelligent auxiliary annotation method for remote sensing images, which solves the problems of low efficiency and high manpower consumption in manual annotation in the prior art, which seriously affects work efficiency. Furthermore, after a long period of annotation work, the annotation accuracy will decrease due to fatigue. This method provides an auxiliary method for annotation of remote sensing images, which improves the work efficiency of annotation personnel and ensures the accuracy of annotation.

[0005] In a first aspect, embodiments of this application provide an intelligent assisted annotation method for remote sensing images, comprising: constructing a total cost function based on a preprocessed remote sensing image; treating each pixel of the remote sensing image as a node in a connected weighted graph, and determining the cost value of the weighted edge between adjacent pixels based on the total cost function; wherein the total cost function is configured to make the cost value of the weighted edge between pixels located on the edge of a feature in the remote sensing image less than a preset cost value; and drawing the shortest path between adjacent markers based on markers sequentially input by an operator along the edge of a feature in the remote sensing image, thereby forming a preliminary outline.

[0006] In conjunction with the first aspect, in a first possible implementation, constructing the total cost function based on the preprocessed remote sensing image includes: obtaining the strong edge point values ​​of the remote sensing image; calculating the pixel gradient magnitude and pixel gradient direction cost of the remote sensing image; and performing a weighted summation of the strong edge point values, the pixel gradient magnitude, and the pixel gradient direction cost to form the total cost function.

[0007] In conjunction with the first possible implementation of the first aspect, in the second possible implementation, obtaining the strong edge point values ​​of the remote sensing image includes: performing ground feature edge detection on the remote sensing image; performing grayscale processing on the remote sensing image, and determining the strong edge point values ​​based on the pixel values ​​of the grayscale image of the remote sensing image; the formula for the strong edge point values ​​is as follows:

[0008]

[0009] In the formula, I(p) is the pixel value of pixel p representing the edge of a ground feature in the remote sensing image after edge detection and grayscale conversion, Z represents a strong edge point, and f Z (p) represents the value of a strong edge point.

[0010] In conjunction with the first aspect, in a third possible implementation, drawing the shortest path between adjacent marker points includes: iteratively calculating the first-generation value between adjacent marker points and taking the marker point with the smallest first-generation value as the parent node; traversing all the parent nodes on the edge of the land feature in the remote sensing image and connecting adjacent parent nodes to obtain the shortest path between the marker points.

[0011] In conjunction with the third possible implementation of the first aspect, in the fourth possible implementation, the step of sequentially calculating the first-generation value between adjacent marker points includes: traversing the marker points and adding them to a node queue; sequentially using each marker point as a starting point and setting the initial generation value of the starting point to 0; traversing the neighboring nodes within the neighborhood of the starting point and calculating the first-generation value from the neighboring nodes added to the node queue to the starting point; wherein, if the neighboring node is on the diagonal of the starting point, the first-generation value is multiplied by...

[0012] In conjunction with the fourth possible implementation of the first aspect, in the fifth possible implementation, the step of taking the marker point with the smallest value in the first generation as the parent node includes: if the sum of the value of the first generation and the initial cost of the starting point is less than the generation value of the neighboring node, then the value of the first generation is assigned to the neighboring node; the starting point at this time is defined as the parent node of the neighboring node, and the starting point is removed from the node queue.

[0013] In conjunction with the first aspect, in the sixth possible implementation, the formation of the preliminary outline further includes: converting the preliminary outline into a binary image; searching for the outer outline of the preliminary outline based on the binary image to obtain the coordinates of the outline point set of the outer outline; converting the outline point set from pixel coordinates to geographic coordinates according to the geographic coordinate information of the remote sensing image and generating a geographic coordinate file; and refining the geographic coordinate file according to the corresponding manual operation until the preliminary outline meets the accuracy requirements.

[0014] Secondly, embodiments of this application provide an intelligent auxiliary annotation device for remote sensing images, characterized in that it includes: a total cost function module, used to construct a total cost function based on a preprocessed remote sensing image; a connected weighted graph module, used to treat each pixel of the remote sensing image as a node in a connected weighted graph, and to determine the cost value of the weighted edge between adjacent pixels based on the total cost function; wherein the total cost function is configured to make the cost value of the weighted edge between pixels located on the edge of a feature in the remote sensing image less than a preset cost value; and a preliminary contour module, used to draw the shortest path between adjacent marker points based on the marker points sequentially input by the operator along the edge of the feature in the remote sensing image, forming a preliminary contour.

[0015] In conjunction with the second aspect, in the first possible implementation, the step of constructing the total cost function based on the preprocessed remote sensing image includes: obtaining the strong edge point values ​​of the remote sensing image; calculating the pixel gradient magnitude and pixel gradient direction cost of the remote sensing image; and performing a weighted summation of the strong edge point values, the pixel gradient magnitude, and the pixel gradient direction cost to form the total cost function.

[0016] In conjunction with the first possible implementation of the second aspect, in the second possible implementation, obtaining the strong edge point values ​​of the remote sensing image includes: performing ground feature edge detection on the remote sensing image; performing grayscale processing on the remote sensing image, and determining the strong edge point values ​​based on the pixel values ​​of the grayscale image of the remote sensing image; the formula for the strong edge point values ​​is as follows:

[0017]

[0018] In the formula, I(p) is the pixel value of pixel p representing the edge of a ground feature in the remote sensing image after edge detection and grayscale conversion, Z represents a strong edge point, and f Z (p) represents the value of a strong edge point.

[0019] In conjunction with the second aspect, in a third possible implementation, drawing the shortest path between adjacent marker points includes: iteratively calculating the first-generation value between adjacent marker points and taking the marker point with the smallest first-generation value as the parent node; traversing all the parent nodes on the edge of the land feature in the remote sensing image and connecting adjacent parent nodes to obtain the shortest path between the marker points.

[0020] In conjunction with the third possible implementation of the second aspect, in the fourth possible implementation, the step of sequentially calculating the first-generation value between adjacent marker points includes: traversing the marker points and adding them to the node queue; sequentially using the marker points as starting points and setting the initial generation value of the starting points to 0; traversing the neighboring nodes in the neighborhood of the starting point and calculating the first-generation value from the neighboring nodes added to the node queue to the starting point; wherein, if the neighboring node is on the diagonal of the starting point, the first-generation value is multiplied by...

[0021] In conjunction with the fourth possible implementation of the second aspect, in the fifth possible implementation, the step of taking the marker point with the smallest value in the first generation as the parent node includes: if the sum of the value of the first generation and the initial cost of the starting point is less than the generation value of the neighboring node, then the value of the first generation is assigned to the neighboring node; the starting point at this time is defined as the parent node of the neighboring node, and the starting point is removed from the node queue.

[0022] In conjunction with the second aspect, in the sixth possible implementation, the formation of the preliminary outline further includes: converting the preliminary outline into a binary image; searching for the outer contour of the preliminary outline based on the binary image to obtain the coordinates of the outline point set of the outer contour; converting the outline point set from pixel coordinates to geographic coordinates according to the geographic coordinate information of the remote sensing image and generating a geographic coordinate file; and refining the geographic coordinate file according to the corresponding manual operation until the preliminary outline meets the accuracy requirements.

[0023] Thirdly, embodiments of this application provide an apparatus comprising: a processor; a memory for storing processor-executable instructions; wherein, when the processor executes the executable instructions, it implements the method as described in the first aspect or any possible implementation of the first aspect.

[0024] Fourthly, embodiments of this application provide a non-volatile computer-readable storage medium, the non-volatile computer-readable storage medium including storage for storing a computer program or instructions that, when executed, cause the method described in the first aspect or any possible implementation of the first aspect to be implemented.

[0025] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0026] This application's embodiments construct a cost function and establish a connected weighted graph using pixels of the remote sensing image as nodes. This allows for the acquisition of feature edges in the remote sensing image. Then, the shortest distance between adjacent points on the feature edge contour is calculated to obtain a preliminary feature edge contour. This effectively solves the problem of existing technologies requiring annotators to concentrate for extended periods on point-by-point annotation. It provides an assisted annotation method where annotators only need to roughly outline the feature edges on the remote sensing image to form a relatively accurate preliminary contour. After manual annotation, a labeled image meeting accuracy requirements can be obtained. This avoids lengthy point-by-point annotation by annotators, thereby improving annotation efficiency and saving manpower and resources. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A flowchart of the intelligent auxiliary annotation method for remote sensing images provided in the embodiments of this application;

[0029] Figure 2A flowchart for constructing the total cost function provided in this application embodiment;

[0030] Figure 3 A flowchart for drawing the shortest path between adjacent marker points provided in this application embodiment;

[0031] Figure 4 A schematic diagram of the intelligent auxiliary annotation device for remote sensing images provided in the embodiments of this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0033] Figure 1 This is a flowchart of the intelligent assisted annotation method for remote sensing images provided in this application embodiment, including steps 101 to 103. Wherein, Figure 1 This is merely one execution order shown in the embodiments of this application and does not represent the only execution order of the intelligent assisted annotation method for remote sensing images. Where the final result can be achieved, Figure 1 The steps shown can be performed in parallel or in reverse order.

[0034] Step 101: Construct the total cost function based on the preprocessed remote sensing image. In this embodiment, the acquired remote sensing image needs to be preprocessed before constructing the total cost function. A preprocessing method is shown below. Those skilled in the art should recognize that various changes and modifications can be made to the preprocessing embodiments described herein without departing from the scope and spirit of this application. Specifically, this includes filtering the acquired remote sensing image to highlight the edges of ground features in the image. In this step, bilateral filtering is used exemplarily in this application; those skilled in the art can also use other methods to achieve the effect of edge highlighting. Furthermore, those skilled in the art can add other denoising and smoothing operations before this step according to the actual situation.

[0035] The specific implementation method of step 101 is as follows: Figure 2 As shown, steps 201 to 203 are included, specifically:

[0036] Step 201: Obtain the strong edge point values ​​of the remote sensing image. Specifically, perform ground object edge detection on the remote sensing image. To obtain relatively accurate ground object edges, the HED edge detection algorithm is used here as an example to detect the ground object edges of the remote sensing image. The Canny algorithm can also be used, but its accuracy is slightly lower than that of the HED edge detection algorithm. The remote sensing image after edge detection is converted to grayscale, and the strong edge point values ​​are determined based on the pixel values ​​of the grayscale image; the formula for the strong edge point value is as follows:

[0037]

[0038] In the formula, I(p) is the pixel value of pixel p representing the edge of a ground feature in the remote sensing image after edge detection and grayscale conversion, Z represents the strong edge point, and f Z (p) represents the strong edge point value. The pixel value of the grayscale remote sensing image ranges from 0 to 255. In this embodiment, pixels with a pixel value of 200 or higher are considered strong edge points. The strong edge point value represents edge characteristics; pixels with higher edge characteristics have lower values. Therefore, the strong edge point value for pixels with a pixel value of 200 or higher is 0, and the strong edge point value for pixels with a pixel value of 200 or lower is 1. In this embodiment, after grayscale processing, Gaussian filtering can also be performed to eliminate Gaussian noise in the image.

[0039] Step 202: Calculate the pixel gradient magnitude and pixel gradient direction cost of the remote sensing image. In this embodiment, the pixel gradient magnitude f G (p), the formula is as follows:

[0040]

[0041] In the formula, f G (p) represents the gradient magnitude of a pixel, G represents the pixel gradient, p represents the pixel, and sx and sy represent the gradients in the x and y directions of the remote sensing image, respectively. A larger pixel gradient indicates more pronounced edge features but a lower computational value.

[0042] Pixel gradient direction cost f D (p,q), the formula is as follows:

[0043]

[0044] In the formula, f D(p,q) represents the gradient direction cost of a pixel, where p and q are pixels, D is the gradient direction of the pixel, dp is the gradient direction of pixel p plus the vector product of the gradient of pixels p and q, and dq is the gradient direction of pixel q plus the vector product of the gradient of pixels p and q. The pixel gradient direction cost can play a role in edge smoothing; therefore, a higher cost is set at locations with drastic boundary changes, and a lower cost is set at locations with gentle boundary changes.

[0045] Step 203: The costs of strong edge points, pixel gradient magnitudes, and pixel gradient directions are weighted and summed to form the total cost function. In this embodiment, the total cost function is as follows:

[0046] L (p,q) =w1f Z (p)+w2f G (p)+w3f D (p,q)

[0047] In the formula, f Z (p) represents the value of a strong edge point, f G (p) represents the gradient magnitude at a pixel, f D (p,q) represents the gradient direction cost of a pixel, where p and q are two pixels, and w1, w2, and w3 are weights. In the embodiments of this application, w1 + w2 + w3 = 1, and 0.7 ≤ w1 ≤ 0.83, 0.03 ≤ w2 ≤ 0.05, and 0.12 ≤ w3 ≤ 0.25; for example, w1 = 0.83, w2 = 0.03, and w3 = 0.14 can achieve better results in this application.

[0048] Step 102: Treat each pixel of the remote sensing image as a node in a connected weighted graph, and determine the cost of the weighted edge between adjacent pixels based on the total cost function. The total cost function is configured such that the cost of the weighted edge between pixels located on the edges of features in the remote sensing image is less than a preset cost. Specifically, a connected weighted graph is constructed using each pixel of the remote sensing image as a node, and each node in the connected weighted graph is assigned a different cost. Adjacent pixels are connected by a weighted edge, the cost of which is calculated using the defined total cost function.

[0049] Furthermore, the purpose of image annotation is to find the edge contours of ground features and generate labels for training neural network models. Therefore, the cost function of the shortest path is closely related to edge features; the more significant the edge features, the lower the cost between pixels, ensuring that the shortest path is as close as possible to the edge of the ground feature. To ensure that the edges of ground features in the found remote sensing images are as close as possible to the ground features, the cost function must be defined such that the cost of the weighted edge between pixels on the edge of the ground feature is less than a preset cost. This preset cost is set by those skilled in the art based on experience or the results of multiple experiments.

[0050] Step 103: Based on the marker points sequentially input by the operator along the edges of features in the remote sensing image, draw the shortest path between adjacent marker points to form a preliminary outline. In this embodiment, the mouse is moved along the edge of the feature; the mouse position corresponds to a marker point, and the shortest path between adjacent marker points is drawn. The specific steps for drawing the shortest path between adjacent marker points are as follows: Figure 3 As shown, steps 301 to 302 are as follows:

[0051] Step 301: Iteratively calculate the first-generation value between adjacent markers and take the marker with the smallest first-generation value as the parent node. Specifically, iterate through the markers and add them to the node queue; sequentially use each marker as a starting point and set its initial generation value to 0; traverse the neighboring nodes within the starting point's neighborhood and calculate the first-generation value from the added neighboring nodes to the starting point; if a neighboring node is on the diagonal of the starting point, multiply the first-generation value by... In this embodiment of the application, the neighborhood node is eight pixels in eight directions around the starting neighborhood: up, down, left, right, and four diagonals.

[0052] The marker with the lowest value in the first generation is designated as the parent node. Specifically, if the sum of the first-generation value and the initial cost of the starting point is less than the cost of the neighboring nodes, the first-generation value is assigned to the neighboring node; the starting point at this time is defined as the parent node of the neighboring nodes, and the starting point is removed from the node queue. In this embodiment, the sum of the first-generation value and the initial cost of the starting point is compared with the cost of the neighboring nodes in the neighborhood. Since the initial cost of the starting point is set to 0, this actually compares the first-generation value with the cost of the neighboring nodes. If the first-generation value is less than the initial cost of the nodes in the neighborhood, the first-generation value is assigned to the neighborhood node, the starting point at this time is defined as the parent node of the neighboring nodes, and the starting point is removed from the node queue. All markers are processed sequentially until all markers are iteratively calculated, i.e., the node queue is empty in this step.

[0053] Step 302: Traverse all parent nodes on the edges of features in the remote sensing image and connect adjacent parent nodes to obtain the shortest path between marker points. In this embodiment, the mouse is used to backtrack along the edge of the feature, continuously searching for the parent node of the current edge node. This continues until all parent nodes are found, and adjacent parent nodes are connected to obtain the shortest path between adjacent marker points. The connecting edges between all adjacent parent nodes constitute the preliminary outline of the feature.

[0054] In this embodiment, after obtaining the preliminary outline of the ground features, the preliminary outline is converted into a binary image; based on the binary image, the outer contour of the preliminary outline is searched to obtain the coordinates of the outline point set; according to the geographic coordinate information of the remote sensing image, the outline point set is converted from pixel coordinates to geographic coordinates and a geographic coordinate file is generated; the geographic coordinate file is annotated according to the corresponding manual operation until the preliminary outline meets the accuracy requirements. For example, the geographic coordinate file can be an SHP file, JSON file, GML file, etc. The resulting preliminary outline is already relatively accurate; at this point, only manual annotation of the preliminary outline is needed to adjust the parts and obtain a labeled file that meets the standard accuracy requirements.

[0055] While this application provides method operation steps as shown in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in this embodiment is merely one possible execution order among many and does not represent the only execution order. In actual device or client product execution, the method can be executed sequentially according to this embodiment or the accompanying drawings, or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0056] like Figure 4 As shown in the figure, this application embodiment also provides a remote sensing image intelligent auxiliary annotation device 400. The device includes: a total cost function module 401, a connected weighted graph module 402, and a preliminary contour module 403.

[0057] The total cost function module 401 is used to construct a total cost function based on the preprocessed remote sensing image. Specifically, the total cost function module 401 is used to: obtain the values ​​of strong edge points in the remote sensing image; calculate the pixel gradient magnitude and pixel gradient direction cost of the remote sensing image; and perform a weighted summation of the strong edge point values, pixel gradient magnitude, and pixel gradient direction cost to form the total cost function.

[0058] The connected weighted graph module 402 is used to treat each pixel of the remote sensing image as a node in the connected weighted graph, and to determine the cost of the weighted edge between adjacent pixels based on the total cost function. The total cost function is configured such that the cost of the weighted edge between pixels located on the edge of a feature in the remote sensing image is less than a preset cost. Specifically, the connected weighted graph module 402 is used for:

[0059] The preliminary contour module 403 is used to draw the shortest path between adjacent marker points based on the marker points sequentially input by the operator along the edge of the remote sensing image, forming a preliminary contour. Specifically, the preliminary contour module 403 is used to: iteratively calculate the first-generation value between adjacent marker points and take the marker point with the smallest first-generation value as its parent node; traverse all parent nodes on the edge of the remote sensing image and connect adjacent parent nodes to obtain the shortest path between marker points. The iterative calculation of the first-generation value between adjacent marker points specifically includes: traversing the marker points and adding them to the node queue; sequentially taking each marker point as a starting point and setting its initial generation value to 0; traversing the neighboring nodes within the starting point's neighborhood and calculating the first-generation value from the added neighboring nodes to the starting point; wherein, if a neighboring node is on the diagonal of the starting point, the first-generation value is multiplied by... The marker point with the lowest value in the first generation is taken as the parent node. Specifically, this includes: if the sum of the first-generation value and the initial cost of the starting point is less than the cost of the neighboring nodes, then the first-generation value is assigned to the neighboring node; the starting point at this time is defined as the parent node of the neighboring nodes, and the starting point is removed from the node queue. Forming the preliminary contour also includes: converting the preliminary contour into a binary image; searching for the outer contour of the preliminary contour based on the binary image to obtain the coordinates of the contour point set; converting the contour point set from pixel coordinates to geographic coordinates based on the geographic coordinate information of the remote sensing image and generating a geographic coordinate file; and refining the geographic coordinate file according to the corresponding manual operation until the preliminary contour meets the accuracy requirements.

[0060] Some modules in the apparatus described in this application can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0061] The apparatus or module described in the above embodiments can be implemented by a computer chip or physical entity, or by a product with a certain function. For ease of description, the above apparatus is described by dividing it into various modules according to their functions. When implementing the embodiments of this application, the functions of each module can be implemented in one or more software and / or hardware. Of course, a module that implements a certain function can also be implemented by combining multiple sub-modules or sub-units.

[0062] The methods, apparatus, or modules described in this application can be implemented in a computer-readable program code manner. The controller can be implemented in any suitable manner, such as a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of a memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code manner, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included within it for implementing various functions can also be considered as structures within the hardware component. Alternatively, the device used to implement various functions can be viewed as either a software module that implements the method or a structure within a hardware component.

[0063] This application also provides an apparatus, the apparatus comprising: a processor; a memory for storing processor-executable instructions; wherein, when the processor executes the executable instructions, it implements the method described in this application.

[0064] This application also provides a non-volatile computer-readable storage medium storing a computer program or instructions thereon, which, when executed, enables the method described in this application embodiment to be implemented.

[0065] Furthermore, in the various embodiments of the present invention, each functional module can be integrated into a processing module, or each module can exist independently, or two or more modules can be integrated into a single module.

[0066] The aforementioned storage media include, but are not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), Cache, Hard Disk Drive (HDD), or Memory Card. The memory can be used to store computer program instructions.

[0067] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary hardware. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, or it can be embodied in the process of data migration. The computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, mobile terminal, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0068] The various embodiments described in this specification are presented in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. All or part of this application can be used in numerous general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, mobile communication terminals, multiprocessor systems, microprocessor-based systems, programmable electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0069] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of this application.

Claims

1. A method for intelligent auxiliary annotation of remote sensing images, characterized in that, include: A total cost function is constructed based on the preprocessed remote sensing images; wherein the total cost function is as follows: ； In the formula, Values ​​for strong edge points. The gradient magnitude of a pixel. The cost is the gradient direction of the pixel. and For two pixels, , , For weights; The method for obtaining strong edge point values ​​includes: performing ground feature edge detection on the remote sensing image; performing grayscale processing on the remote sensing image; and determining strong edge point values ​​based on the pixel values ​​of the grayscale image; the formula for the strong edge point values ​​is as follows: ； In the formula, The pixels representing the edges of ground features in the remotely sensed image after edge detection and grayscale conversion. The pixel value, Z represents a strong edge point. Values ​​for strong edge points; Each pixel of the remote sensing image is used as a node in a connected weighted graph, and the weighted edge value of adjacent pixels is determined according to the total cost function. The total cost function is configured such that the cost of the weighted edge between pixels on the edge of a ground feature in the remote sensing image is less than a preset cost. Based on the marker points entered sequentially by the operator along the edges of the features in the remote sensing image, the shortest path between adjacent marker points is drawn to form a preliminary outline.

2. The method according to claim 1, characterized in that, The construction of the total cost function based on the preprocessed remote sensing image includes: Obtain the values ​​of strong edge points in the remote sensing image; Calculate the pixel gradient magnitude and pixel gradient direction cost of the remote sensing image; The total cost function is formed by weighted summation of the strong edge point value, the pixel gradient magnitude, and the pixel gradient direction cost.

3. The method according to claim 1, characterized in that, Drawing the shortest path between adjacent marker points includes: Iteratively calculate the first-generation value between adjacent marker points, and take the marker point with the smallest first-generation value as the parent node; Traverse all the parent nodes on the edges of the features in the remote sensing image and connect adjacent parent nodes to obtain the shortest path between the marker points.

4. The method according to claim 3, characterized in that, The step of sequentially calculating the first-generation value between adjacent marker points includes: Traverse the marked points and add them to the node queue; The marked points are used as starting points in sequence, and the initial cost of the starting points is set to 0. Traverse the neighboring nodes within the neighborhood of the starting point, and calculate the first-generation value of the neighboring nodes added to the node queue up to the starting point; wherein, if the neighboring node is on the diagonal of the starting point, multiply the first-generation value by [missing information]. .

5. The method according to claim 4, characterized in that, The step of taking the marker point with the lowest value in the first generation as the parent node includes: If the sum of the first generation value and the initial cost of the starting point is less than the generation value of the neighboring node, then the first generation value is assigned to the neighboring node. Define the starting point at this time as the parent node of the neighboring node, and remove the starting point from the node queue.

6. The method according to claim 1, characterized in that, The process of forming a preliminary outline also includes: The preliminary contour is converted into a binary image; The outer contour of the preliminary contour is found based on the binary image, and the coordinates of the contour point set of the outer contour are obtained. Based on the geographic coordinate information of the remote sensing image, the contour point set is converted from pixel coordinates to geographic coordinates and a geographic coordinate file is generated; The geographic coordinate file is revised according to the response of manual operation until the preliminary outline meets the accuracy requirements.

7. A remote sensing image intelligent auxiliary annotation device, characterized in that, include: The total cost function module is used to construct a total cost function based on the preprocessed remote sensing image; wherein the total cost function is as follows: ； In the formula, Values ​​for strong edge points. The gradient magnitude of a pixel. The cost is the gradient direction of the pixel. and For two pixels, , , For weights; The method for obtaining strong edge point values ​​includes: performing ground feature edge detection on the remote sensing image; performing grayscale processing on the remote sensing image; and determining strong edge point values ​​based on the pixel values ​​of the grayscale image; the formula for the strong edge point values ​​is as follows: ； In the formula, The pixels representing the edges of ground features in the remotely sensed image after edge detection and grayscale conversion. The pixel value, Z represents a strong edge point. Values ​​for strong edge points; The connected weighted graph module is used to treat each pixel of the remote sensing image as a node of the connected weighted graph, and to determine the cost of the weighted edge between adjacent pixels according to the total cost function; wherein, the total cost function is configured to make the cost of the weighted edge between pixels located on the edge of the ground feature in the remote sensing image less than a preset cost. The preliminary contour module is used to draw the shortest path between adjacent marker points based on the marker points sequentially input by the operator along the edges of the features in the remote sensing image, thus forming a preliminary contour.

8. An apparatus for performing an intelligent assisted annotation method for remote sensing images, characterized in that, include: processor; Memory used to store processor-executable instructions; When the processor executes the executable instructions, it implements the method as described in any one of claims 1 to 6.

9. A non-volatile computer-readable storage medium, characterized in that, Includes storage of computer programs or instructions that, when executed, cause the method as described in any one of claims 1 to 6 to be implemented.

Images