Mining area road drawing method and system

Abstract

The invention discloses a mining area road drawing method and system, and relates to the technical field of image processing, and the method comprises the steps: building a prediction sample data set; training a prediction model by using the prediction sample data set; acquiring remote sensing images of two adjacent moments in the to-be-identified mining area; inputting the pre-moment remote sensing image into the trained prediction model to obtain the road updating probability of each region in the pre-moment remote sensing image; screening all areas in the remote sensing image at the later moment according to the probability; identifying an area obtained by screening the remote sensing image at the later moment by using an identification model; and drawing a mining area road map according to the road recognition result. According to the method, all pixels of the remote sensing image do not need to be identified, and meanwhile, the accuracy of an identification result is effectively ensured.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and particularly to a method and system for drawing mining area roads. Background Art

[0002] During the production process in a mining area, various products need to be transported by vehicles. Therefore, multiple roads are planned in the mining area, including both permanent roads and temporary roads. For temporary roads, as the development progress of the mining area advances, the demand for product transportation will also change. Therefore, the temporary roads will be updated regularly or irregularly, including new construction, abandonment, widening, etc. Due to the large number of vehicles in the mining area, including both internal vehicles in the mining area and external vehicles, and the internal vehicles are also divided into types such as manned driving and unmanned driving, it is necessary to draw the updated roads in a timely and accurate manner to meet the road information needs of different vehicles.

[0003] Currently, the update of road information mainly relies on manual update or third-party map data. Manual update has the defects of high cost and low accuracy, while third-party map data such as Google Maps has a long update cycle, and the drawing accuracy of this kind of map data is fixed and cannot meet the different accuracy requirements of different vehicles for the roads in the map. To address this problem, the prior art has proposed a technology for road recognition using remotely sensed images collected by drones. This technology is not limited by the update cycle of third-party map data, and the accuracy can also be defined by itself, so it is welcomed by many mines.

[0004] However, after obtaining the remotely sensed images by using drones, due to the large amount of data in the remotely sensed images, if each pixel is recognized, it will inevitably take a lot of time. And if only a part of the pixels are recognized, it will lead to the omission of road information and reduce the accuracy of the drawing result. Summary of the Invention

[0005] The embodiments of this application provide a method and system for drawing mining area roads to solve the problems existing in the prior art of recognizing each pixel in each remotely sensed image or only recognizing a part of the pixels.

[0006] On the one hand, the embodiments of this application provide a method for drawing mining area roads, including:

[0007] Establish a prediction sample data set, where the prediction sample data set includes multiple prediction samples and corresponding annotations;

[0008] Use the prediction sample data set to train a prediction model to obtain a trained prediction model;

[0009] Obtain remote sensing images of two adjacent moments in the mining area to be recognized, where the remote sensing images include the remote sensing image of the previous moment and the remote sensing image of the later moment;

[0010] Input the remote sensing image of the previous moment into the trained prediction model to obtain the probability of updated roads in each area of the remote sensing image of the previous moment;

[0011] Screen all areas in the remote sensing image of the later moment according to the probability to obtain areas with probabilities exceeding the set threshold;

[0012] Use the recognition model to recognize the areas screened from the remote sensing image of the later moment to obtain the road recognition result;

[0013] Draw a mining area road map according to the road recognition result.

[0014] On the other hand, the embodiment of the present application also provides a mining area road drawing system, including:

[0015] A dataset establishment module for establishing a prediction sample dataset, where the prediction sample dataset includes multiple prediction samples and corresponding annotations;

[0016] A model training module for training the prediction model using the prediction sample dataset to obtain the trained prediction model;

[0017] An image acquisition unit for obtaining remote sensing images of two adjacent moments in the mining area to be recognized, where the remote sensing images include the remote sensing image of the previous moment and the remote sensing image of the later moment;

[0018] A probability determination module for inputting the remote sensing image of the previous moment into the trained prediction model to obtain the probability of updated roads in each area of the remote sensing image of the previous moment;

[0019] An area screening module for screening all areas in the remote sensing image of the later moment according to the probability to obtain areas with probabilities exceeding the set threshold;

[0020] An area recognition module for using the recognition model to recognize the areas screened from the remote sensing image of the later moment to obtain the road recognition result;

[0021] A road drawing module for drawing a mining area road map according to the road recognition result.

[0022] A mining area road drawing method and system in the present application have the following advantages:

[0023] Analyze the remote sensing image of the previous moment before the road update in the mining area to determine whether there are potential areas for forming roads in the remote sensing image. For areas with such potential situations, key recognition will be carried out after the road update, without the need to recognize all pixels of the remote sensing image, but at the same time, the accuracy of the recognition result is effectively ensured. Description of the Drawings

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0025] Figure 1 It is a mileage map of a method for drawing mining area roads provided by an embodiment of the present application. Detailed Embodiments

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0027] Figure 1 It is a flowchart of a method for drawing mining area roads provided by an embodiment of the present application. An embodiment of the present application provides a method for drawing mining area roads, including:

[0028] S100. Establish a prediction sample data set, where the prediction sample data set includes multiple prediction samples and corresponding annotations.

[0029] Exemplarily, the method for establishing a prediction sample data set includes:

[0030] Obtain dual-temporal remote sensing image samples of the sample mining area;

[0031] Identify the updated roads in the dual-temporal remote sensing images;

[0032] Use the remote sensing image sample with a later time in the dual-temporal remote sensing image samples as the prediction sample, and label the endpoints of the updated roads in the prediction sample.

[0033] The dual - time remote - sensing image sample is the remote - sensing image collected twice successively in the sample mining area. This dual - time remote - sensing image can be collected by the same type of unmanned aerial vehicle at the same coordinates and altitude to ensure the consistency in scope and quality. The sample mining area is other mining areas except the mining area to be identified. Although the situations of the sample mining area and the mining area to be identified are different, when updating the roads, the same logic is followed. For example, after the mining point changes, the road needs to be extended or newly added accordingly, and the same rules also need to be followed during the road - updating process, such as the slope of the road should not be too large and there should be no obstacles on the road. Based on this same logic, processing the dual - time remote - sensing image sample of the sample mining area can clarify the underlying logic of road updating, and thus provide an accurate basis for road updating in the mining area to be identified.

[0034] Furthermore, the method for identifying the updated roads in the dual - time remote - sensing image includes:

[0035] Conduct change detection on the dual - time remote - sensing image sample to determine the corresponding change area;

[0036] Input the change area into the recognition model to determine the updated roads in the dual - time remote - sensing image.

[0037] Specifically, there will inevitably be changes in the remote - sensing image samples at different times. There are many reasons for these changes. For example, the positions of buildings and equipment in the mining area have changed, the mining point has changed, the road has changed, etc. By conducting change detection on the dual - time remote - sensing image sample, all the changed areas can be obtained. And through the recognition of the recognition model, the change of the road can be identified from all the changes to determine the road - change situation in the sample mining area.

[0038] The recognition model can adopt a model established based on a neural network, and its function is to identify whether the pixels in the change area belong to the road. It is a binary - classification model, and the output result is the recognition result of the category to which each pixel belongs. After each change area is recognized, the pixels belonging to the road in the change area can be obtained. Connecting the pixels belonging to the road in the same change area into a piece can obtain the updated road.

[0039] Furthermore, after annotating the endpoints of the updated roads in the prediction sample, determine the types of all points on the new road in the prediction sample. The types include endpoints and connection points. The endpoint is the point where the vehicle needs to turn back after completing operations in the mining area, and the connection point is the connection point between roads.

[0040] Specifically, a road is a linear structure. After determining the positions of both ends of the road, the possible range of the road can be determined. In the embodiments of the present application, the annotation of road endpoints is mainly carried out manually. During the annotation process, technicians can set annotations for each pixel of the road or the pixels in a region. After setting the annotations, an annotation set will be formed that includes the positions and categories of the pixels belonging to the road in the prediction sample. This annotation set and the corresponding prediction sample can then form a prediction sample data set.

[0041] S110. Use the prediction sample data set to train the prediction model to obtain a trained prediction model.

[0042] Exemplarily, the prediction model can also be a model established based on a neural network. The role of the prediction model is to identify whether the pixels in a remote sensing image belong to the endpoints or connection points of the road. Similar to the recognition model, the prediction model is also a binary classification model. For the input remote sensing image, its output will be the predicted category for each pixel, that is, belonging to the endpoint or connection point, or not belonging to the endpoint or connection point. It should be understood that when the prediction model predicts the category of a pixel, the output result is the probability of belonging to or not belonging to the endpoint or connection point. For pixels with a probability reaching a certain value, it will be considered that the predicted category of the pixel is acceptable.

[0043] S120. Obtain the remote sensing images of two adjacent moments in the mining area to be identified. The remote sensing images include the remote sensing image of the previous moment and the remote sensing image of the later moment.

[0044] Exemplarily, the remote sensing images of two adjacent moments can also be called dual - time remote sensing images. Before the road in the mining area to be identified is updated, the initial remote sensing image will be collected by a drone. When the road needs to be drawn later, the drone can collect the remote sensing image again at the same coordinates and height as when collecting the initial remote sensing image. Then, the two adjacent remote sensing images collected form the dual - time remote sensing images.

[0045] S130. Input the remote sensing image of the previous moment into the trained prediction model to obtain the probability of updating the road for each region in the remote sensing image of the previous moment.

[0046] Exemplarily, the remote sensing image of the previous moment is the image before the road is updated. Although it does not contain the information of the updated road, it contains potential endpoints or connection points of the updated road. By predicting the remote sensing image of the previous moment, the regions where the road may be updated can be more accurately determined, and then these regions where the road may be updated can be identified in the remote sensing image of the later moment. By adopting this method of predicting first and then identifying, it is not necessary to identify each pixel in the remote sensing image of the later moment. Only identifying a part of the pixels can ensure a sufficiently high recognition accuracy.

[0047] Further, when predicting the probability of updated roads in each area of the remote sensing image at the previous moment using the trained prediction model, each area in the remote sensing image at the previous moment is identified to determine whether there are endpoints or connection points in each area. If there are endpoints or connection points, corresponding probabilities are set for the areas between the endpoints and the connection points.

[0048] Specifically, after the prediction model is trained, the prediction accuracy for pixels belonging to endpoints or connection points in the remote sensing image has reached a sufficiently high level. Therefore, after inputting the remote sensing image at the previous moment, the probability of each pixel belonging to an endpoint or a connection point can be obtained. However, there are differences in the probabilities of each pixel belonging to an endpoint or a connection point. Only pixels with probabilities exceeding the set threshold will be considered to truly belong to endpoints or connection points. If the probability is too low, it will be considered that the pixel does not belong to an endpoint or a connection point.

[0049] The embodiments of the present application adopt the same block strategy for the remote sensing image at the previous moment and the remote sensing image at the subsequent moment, that is, when their resolutions are exactly the same, the two remote sensing images are divided into multiple areas with the same positions and sizes. After determining the category of pixels in each area, if the number of pixels considered to be endpoints or connection points in a certain area reaches a certain value, it can be considered that the prediction result of this area is reliable. At this time, it can be considered that there are endpoints or connection points in this area, and the probability of the existence of endpoints or connection points in this area can be set as the mean value of the probabilities of the pixels belonging to endpoints or connection points.

[0050] Further, after determining the areas with endpoints or connection points in the previous remote sensing image, each endpoint and each connection point are combined, and corresponding probabilities are set for the areas between the endpoints and the connection points in each combination.

[0051] Specifically, the road is the range between endpoints and connection points. After determining the endpoints and connection points, there are many possibilities for the existence of the road. To determine the possible range of the road, it is necessary to determine multiple possible areas between the endpoints and the connection points. The probabilities of these areas can be set as the mean value of the probabilities of the areas to which the corresponding endpoints and connection points belong, or the probabilities of these areas can be set to be inversely proportional to the distance from the corresponding endpoints or connection points, that is, the closer an area is to an endpoint or a connection point, the greater the probability, and vice versa, but the maximum probability cannot exceed the probability of the area where the corresponding endpoint or connection point is located. Preferably, the areas in the straight-line direction between the endpoints and the connection points can be used as the possible range where the road exists, and then probabilities are set for these areas.

[0052] Further, after obtaining multiple combinations, the combinations are filtered according to the terrain of the mining area to be identified.

[0053] Specifically, although multiple combinations are formed between endpoints and connection points, not all combinations can form roads. For example, there may be steep slopes or impassable terrains in the areas between some combinations. Obviously, roads cannot be formed in these places. Therefore, it is necessary to conduct a survey of the terrain in the mining area in advance. If there is a problem that a road cannot be formed in the area between a certain combination, then this combination needs to be deleted, and only the combinations that can theoretically form roads are retained.

[0054] S140. Screen all regions in the remote sensing image at a later time according to probability, and obtain the regions whose probability exceeds the set threshold.

[0055] Exemplarily, the remote sensing image at a later time already contains the updated road. Therefore, based on the prediction result of the remote sensing image at an earlier time, the regions that may theoretically form roads in the remote sensing image at a later time can be determined.

[0056] S150. Use the recognition model to recognize the regions screened from the remote sensing image at a later time, and obtain the road recognition result.

[0057] Exemplarily, the regions screened by S140 are rectangular regions, and not all pixels in them belong to the road. Therefore, it is still necessary to use the recognition model to further recognize the pixels in these regions to screen out the pixels that belong to the road. The pixels that belong to the road in all regions can be connected to form a road recognition result in the form of a region.

[0058] S160. Draw a mining area road map according to the road recognition result.

[0059] Exemplarily, S150 has recognized the ranges of all roads in the remote sensing image at a later time. Based on the ranges of these roads, the corresponding mining area road map can be drawn.

[0060] The embodiment of the present application also provides a mining area road drawing system, which includes:

[0061] A data set establishment module for establishing a prediction sample data set, where the prediction sample data set includes multiple prediction samples and corresponding annotations;

[0062] A model training module for training a prediction model using the prediction sample data set to obtain a trained prediction model;

[0063] An image acquisition unit for acquiring remote sensing images of two adjacent times in the mining area to be recognized, where the remote sensing images include the remote sensing image at an earlier time and the remote sensing image at a later time;

[0064] A probability determination module for inputting the remote sensing image at an earlier time into the trained prediction model to obtain the probability of updated roads for each region in the remote sensing image at an earlier time;

[0065] An area screening module, configured to screen all areas in the remote sensing image at a later time according to probabilities, and obtain areas with probabilities exceeding a set threshold;

[0066] An area recognition module, configured to use a recognition model to recognize the areas screened from the remote sensing image at a later time, and obtain a road recognition result;

[0067] A road drawing module, configured to draw a mine road map according to the road recognition result.

[0068] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0069] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. A method for drawing a mining area road, characterized in that, it includes: establishing a prediction sample data set, which includes a plurality of prediction samples and corresponding annotations; training a prediction model using the prediction sample data set to obtain the trained prediction model; acquiring remote sensing images of two adjacent moments in the mining area to be recognized, where the remote sensing images include a remote sensing image of the previous moment and a remote sensing image of the later moment; inputting the remote sensing image of the previous moment into the trained prediction model to obtain the probability of updated roads in each area of the remote sensing image of the previous moment; screening all areas in the remote sensing image of the later moment according to the probability to obtain the areas where the probability exceeds a set threshold; using an identification model to identify the areas screened from the remote sensing image of the later moment to obtain a road identification result; drawing a mining area road map according to the road identification result.

2. A method for drawing a mining area road according to claim 1, characterized in that, the method for establishing the prediction sample data set includes: acquiring a dual-time remote sensing image sample of a sample mining area; identifying the updated roads in the dual-time remote sensing image; using the remote sensing image sample with a later time in the dual-time remote sensing image sample as the prediction sample, and annotating the endpoints of the updated roads in the prediction sample.

3. A method for drawing a mining area road according to claim 2, characterized in that, the method for identifying the updated roads in the dual-time remote sensing image includes: performing change detection on the dual-time remote sensing image sample to determine the corresponding change area; inputting the change area into the identification model to determine the updated roads in the dual-time remote sensing image.

4. A method for drawing a mining area road according to claim 2, characterized in that, after annotating the endpoints of the updated roads in the prediction sample, determining the types of all points on the new roads in the prediction sample, where the types include endpoints and connection points, the endpoints are the points where the vehicle needs to turn back after completing operations in the mining area, and the connection points are the connection points between roads.

5. A method for drawing a mining area road according to claim 4, characterized in that, when predicting the probability of updated roads in each area of the remote sensing image of the previous moment using the trained prediction model, identifying each area in the remote sensing image of the previous moment to determine whether there are the endpoints or the connection points in each area. If there are the endpoints or the connection points, set the corresponding probability for the area between the endpoints and the connection points.

6. A method for drawing a mining area road according to claim 5, characterized in that, after determining the areas in the previous remote sensing image where there are the endpoints or the connection points, combining each endpoint and each connection point, and setting the corresponding probability for the area between the endpoints and the connection points in each combination.

7. A method for drawing a mining area road according to claim 6, characterized in that, after obtaining a plurality of the combinations, filtering the combinations according to the terrain of the mining area to be recognized.

8. A system applying the method for drawing a mining area road according to any one of claims 1-7, It is characterized in that including a data set establishment module for establishing a prediction sample data set, where the prediction sample data set includes multiple prediction samples and corresponding annotations; a model training module for training a prediction model using the prediction sample data set to obtain the trained prediction model; an image acquisition unit for acquiring remote sensing images of two adjacent moments in the mining area to be recognized, where the remote sensing images include a remote sensing image of the previous moment and a remote sensing image of the subsequent moment; a probability determination module for inputting the remote sensing image of the previous moment into the trained prediction model to obtain the probability of updating the road for each area in the remote sensing image of the previous moment; a region screening module for screening all regions in the remote sensing image of the subsequent moment according to the probability to obtain regions where the probability exceeds a set threshold; a region recognition module for recognizing the regions screened from the remote sensing image of the subsequent moment using a recognition model to obtain a road recognition result; a road drawing module for drawing a mining area road map according to the road recognition result.

Images