An oil well identification extraction method and device

Abstract

The application provides an oil well identification extraction method and device. The method comprises the following steps: obtaining a to-be-identified remote sensing image; inputting the to-be-identified remote sensing image into an oil well station extraction model, and outputting an oil well station classification result; wherein the oil well station extraction model adopts a Mask R-CNN model introducing a D-LinkNet network. According to the scheme, the D-LinkNet network is used to realize feature extraction of detailed information, and the newly-added semantic segmentation branch is used to analyze and utilize the potential connection between road information and oil well stations, so that the effect of oil well station identification extraction is improved.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing technology, and specifically relates to a method and apparatus for oil well identification and extraction. Background Technology

[0002] Oil and gas production is crucial for economic development. However, oil and gas extraction inevitably has adverse environmental impacts. The most common technology for oil and gas production involves constructing drilling sites, along with a series of well sites and other facilities. The extraction of numerous well sites and resource roads inevitably leads to landscape destruction or fragmentation. Earth observation or remote sensing technologies provide an effective means to extract information about surface disturbances, supporting environmental assessments and risk analyses related to oil extraction activities.

[0003] Currently, visual interpretation based on remote sensing images is the conventional method for oil well location extraction. However, this method is time-consuming and heavily reliant on specialized knowledge. In recent years, Deep Learning (DL) has been applied to oil industry facility identification due to its powerful feature extraction capabilities and end-to-end data processing, enabling it to solve complex problems with higher accuracy by training on large amounts of data. However, mask images of well sites are crucial for the scientific analysis of land disturbance. Therefore, in automated oil well site detection and extraction, in addition to locating the oil well site, the extraction of mask information should also be considered.

[0004] In existing technologies, the accuracy of oil well site identification is relatively low. Summary of the Invention

[0005] The purpose of the embodiments in this specification is to provide an oil well identification and extraction method and apparatus that can improve the identification accuracy of mask information at oil well sites.

[0006] To solve the above-mentioned technical problems, the embodiments of this application are implemented in the following ways:

[0007] In a first aspect, this application provides a method for identifying and extracting oil wells, the method comprising:

[0008] Acquire the remote sensing image to be identified;

[0009] The remote sensing image to be identified is input into the oil well site extraction model, and the oil well site classification results are output; the oil well site extraction model adopts the Mask R-CNN model with D-LinkNet network.

[0010] In one embodiment, the D-LinkNet network includes a semantic segmentation branch for extracting information about oil well sites and their connected roads.

[0011] In one embodiment, the loss function of the well site extraction model includes the intersection-union ratio loss function of the semantic segmentation branch.

[0012] In one embodiment, the training dataset used to train the oil well site extraction model is a multi-source remote sensing image dataset.

[0013] In one embodiment, the multi-source remote sensing image dataset includes remote sensing images with different spatial and spectral resolutions, wherein the remote sensing images with different spatial resolutions include low spatial resolution remote sensing images and high resolution remote sensing images.

[0014] In one embodiment, constructing a training dataset includes:

[0015] Examine the spectral band information of all the remote sensing images and select all common spectral bands to maximize the use of the collected remote sensing images and make the spectral resolution of all the remote sensing images consistent.

[0016] The residual channel attention network is used to perform super-resolution on the low spatial resolution remote sensing image in the remote sensing image after the spectral resolution is consistent, so as to obtain the super-resolution remote sensing image. The resolution of the super-resolution remote sensing image is consistent with the resolution of the high resolution remote sensing image.

[0017] Super-resolution remote sensing images and high-resolution remote sensing images are used to form a training dataset.

[0018] In one embodiment, the oil well site classification results are divided into those including oil well sites and those excluding oil well sites;

[0019] If the classification result of the oil well site includes oil well sites, the oil well site extraction model also outputs mask images of all oil well sites.

[0020] Secondly, this application provides an oil well identification and extraction device, the device comprising:

[0021] The acquisition module is used to acquire the remote sensing image to be identified;

[0022] The processing module is used to input remote sensing images into the oil well site extraction model and output oil well site classification results and mask images; the oil well site extraction model adopts a Mask R-CNN network with D-LinkNet network.

[0023] As can be seen from the technical solutions provided in the embodiments of this specification above, this solution: uses a Mask R-CNN model incorporating a D-LinkNet network as the oil well site extraction model, which can not only identify the location of oil well sites but also provide mask images of the oil well sites. The D-LinkNet network is used to achieve perception and recognition of oil wells and their connected road information, and the potential connections between road information and oil well sites can be analyzed and utilized to improve the effectiveness of oil well site identification and extraction. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A schematic diagram illustrating the principle of the oil well identification and extraction method provided in this application;

[0026] Figure 2 A flowchart illustrating the oil well identification and extraction method provided in this application;

[0027] Figure 3 This is a schematic diagram of the structure of the oil well identification and extraction device provided in this application. Detailed Implementation

[0028] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0029] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0030] Various modifications and variations can be made to the specific embodiments described in this application without departing from the scope or spirit of this application, as will be apparent to those skilled in the art. Other embodiments derived from this application will be obvious to those skilled in the art. This application specification and embodiments are merely exemplary.

[0031] The terms “include,” “including,” “have,” “contain,” etc., used in this article are all open-ended terms, meaning that they include but are not limited to.

[0032] In automated oil well site detection and extraction, in addition to identifying the oil well site, it is also necessary to extract the mask information of the oil well site. Furthermore, oil well site and road information often appear in combination, and road information is also a key feature for oil well site extraction. Therefore, oil well site identification and extraction mainly face the following technical bottlenecks: 1) How to develop an instance segmentation-based method to quickly and efficiently extract oil well site mask information; 2) How to use road information from oil wells and analyze and utilize the special combination relationship between oil wells and road information to improve the effectiveness of oil well site mask information identification and extraction; 3) How to use multi-source remote sensing images with different spatial resolutions to jointly train the model and improve the model's generalization ability.

[0033] To address the aforementioned technical bottlenecks, this application provides an oil well identification and extraction method that can not only improve the generalization ability of the model but also increase the accuracy of identifying oil well sites.

[0034] like Figure 1 The diagram illustrates the principle of the oil well identification and extraction method provided in this application. This method uses a Mask R-CNN (Mask Region-based Convolutional Neural Network) model to construct an oil well site extraction model for identification and extraction. Specifically, the Mask R-CNN model incorporates a D-LinkNet network (a high-resolution satellite image road extraction semantic segmentation network based on a pre-trained encoder and dilated convolution) to replace the original CNN (Convolutional Neural Networks). Simultaneously, a new semantic segmentation branch is added to the Mask R-CNN based on the D-LinkNet network. The oil well identification and extraction method performs super-resolution processing (using a residual channel attention network pre-trained with RCAN in the diagram) on the acquired remote sensing image, and then inputs the resulting image into the oil well site extraction model (i.e.,...). Figure 1In the OWS (Mask R-CNN) oil well site extraction model, the model sequentially passes through a D-LinkNet network, a Feature Pyramid Network (FPN), a Region Proposal, a Region of Interest Align (RoI) operation, and fully connected operations (FC Layers and FC) to obtain the oil well site classification result. If the oil well site classification result includes oil well sites, the model then sequentially passes through a D-LinkNet network, a Feature Pyramid Network, a Region Proposal, a Region of Interest Alignment operation, and a Fully Connected Convolutional Network (FCN) to obtain the mask images of all oil well sites.

[0035] Continue to refer to Figure 1 The D-LinkNet network includes a semantic segmentation branch, which is trained using mask images of real-world oil well sites and their connected linear roads.

[0036] Since the D-LinkNet network can achieve refined perception and recognition of linear features, such as road information, and can analyze and utilize the potential connection between road information and oil well sites, the oil well identification and extraction method provided in this application can improve the identification accuracy of oil well sites by using the Mask R-CNN model with D-LinkNet network to identify oil well sites in the remote sensing image to be identified.

[0037] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0038] Reference Figure 2 It shows a flowchart of the oil well identification and extraction method applicable to the embodiments of this application.

[0039] like Figure 2 As shown, the oil well identification and extraction method may include:

[0040] S210. Acquire the remote sensing image to be identified.

[0041] Specifically, the remote sensing images to be identified can be remote sensing images of different resolutions, such as RapidEye satellite sensor images and WorldView-3 satellite remote sensing images.

[0042] Understandably, since the acquired remote sensing images to be identified cover a wide range of spatial and spectral resolutions, super-resolution and band selection can be performed on these images to improve their consistency. The following remote sensing images to be identified are those after super-resolution and band selection.

[0043] S220. Input the remote sensing image to be identified into the oil well site extraction model, and output the oil well site classification results. The oil well site classification results are divided into those including oil well sites and those excluding oil well sites. If the oil well site classification result includes oil well sites, the oil well site extraction model also outputs mask images of all oil well sites.

[0044] Among them, the oil well site extraction model adopts the Mask R-CNN model with the introduction of D-LinkNet network, which can improve the network model's ability to identify oil well sites, such as road information recognition, by utilizing relevant information about oil wells and roads.

[0045] Specifically, the Mask R-CNN model that introduces the D-LinkNet network replaces the original CNN (Convolutional Neural Network) in the Mask R-CNN model with a D-LinkNet network. The D-LinkNet network uses an encoder-decoder structure, dilated convolutions, and a pre-trained encoder for the road extraction task.

[0046] Optionally, the D-LinkNet network includes a semantic segmentation branch, which can extract information about oil well sites and their connected roads. This enables the entire oil well site extraction model to perceive the potential relationship between roads (linear objects) and oil well sites, allowing road information around the well locations to participate in the recognition process and improving the model's recognition capability. Specifically, during the training phase of the oil well site extraction model, the semantic segmentation branch uses real ground data containing both oil well site mask information and road mask information.

[0047] The newly added semantic segmentation branch is for segmenting information about oil well sites and their connected roads. Introducing a soft jaccard loss function into this segmentation task can improve the oil well site extraction performance. Specifically, the loss function of the oil well site extraction model includes the cross-over ratio loss function from the semantic segmentation branch; that is, the loss function of the oil well site extraction model is the same as the loss function of the Mask R-CNN model, with the cross-over ratio loss function from the semantic segmentation branch added.

[0048] The loss function for the oil well site extraction model is:

[0049] L = L cls +L box +L mask +L sem

[0050] The first three terms are the multi-class cross-entropy loss function, the bounding box regression parameter loss function, and the mask information prediction loss function (defined as the sum of the binary classification losses for each pixel), which are consistent with the loss function of the Mask R-CNN model. sem is the intersection-union ratio loss function for semantic segmentation branches.

[0051] In this embodiment, a Mask R-CNN model incorporating a D-LinkNet network is used as the oil well site extraction model. This model can not only identify the location of oil well sites but also provide mask images of them. The D-LinkNet network is used to perceive and identify linear features such as roads, and the potential connection between road information and oil well sites can be analyzed and utilized to improve the effectiveness of oil well site identification and extraction.

[0052] The oil well identification and extraction method provided in this application provides an effective means to extract information about surface disturbances, providing support for environmental assessment and risk analysis related to oil extraction activities.

[0053] In one embodiment, the training dataset used to train the oil well site extraction model is a multi-source remote sensing image dataset. This multi-source remote sensing image dataset includes remote sensing images with different spatial and spectral resolutions, wherein the remote sensing images with different spatial resolutions include low-spatial-resolution remote sensing images and high-resolution remote sensing images.

[0054] Understandably, multi-source remote sensing image datasets can be image data from multiple sensors, such as RapidEye 2 / 3 satellite sensor images and WorldView-3 satellite remote sensing images. Therefore, by selecting common bands to identify applicable spectral bands in the remote sensing images, and applying a pre-trained Residual Channel Attention Network (RCAN) to a super-resolution task with images of different spatial resolutions, the resolution of images with different resolutions is unified, resulting in the training dataset for the oil well site extraction model.

[0055] The construction of the training dataset includes:

[0056] Examine the spectral band information of all the remote sensing images and select all common spectral bands to maximize the use of the collected remote sensing images and make the spectral resolution of all the remote sensing images consistent.

[0057] The residual channel attention network is used to perform super-resolution on the low spatial resolution remote sensing image in the remote sensing image after the spectral resolution is consistent, so as to obtain the super-resolution remote sensing image. The resolution of the super-resolution remote sensing image is consistent with the resolution of the high resolution remote sensing image.

[0058] Super-resolution remote sensing images and high-resolution remote sensing images are used to form a training dataset.

[0059] Specifically, a pre-trained residual channel attention network is used to perform super-resolution on low spatial resolution remote sensing images to make them consistent with high-resolution remote sensing images. Then, the super-resolution remote sensing images and high-resolution remote sensing images are combined to form a multi-source remote sensing image dataset. Finally, the multi-source remote sensing image dataset is cropped to a size of 512*512 to form the training dataset for the model.

[0060] In this embodiment, the training dataset used to train the oil well site extraction model is a multi-source remote sensing image dataset. This enables the identification of remote sensing images with different spatial resolutions when using the oil well site extraction model to identify and extract oil well sites, thereby realizing the comprehensive utilization of multi-source remote sensing image data in oil well identification and improving the generalization ability of the oil well site extraction model.

[0061] Reference Figure 3 The diagram shows a structural schematic of an oil well identification and extraction device according to an embodiment of this application.

[0062] like Figure 3 As shown, the oil well identification and extraction device 300 may include:

[0063] The acquisition module 310 is used to acquire the remote sensing image to be identified;

[0064] The processing module 320 is used to input remote sensing images into the oil well site extraction model and output oil well site classification results and mask images; wherein, the oil well site extraction model adopts a Mask R-CNN network with D-LinkNet network.

[0065] Optionally, the D-LinkNet network includes a semantic segmentation branch, which is used to extract information about oil well sites and their connected roads.

[0066] Optionally, the loss function for the well site extraction model may include the intersection-union ratio loss function of the semantic segmentation branch.

[0067] Optionally, the training dataset used to train the oil well site extraction model is a multi-source remote sensing image dataset.

[0068] Optionally, the multi-source remote sensing image dataset contains remote sensing images with different spatial and spectral resolutions, wherein the remote sensing images with different spatial resolutions include low spatial resolution remote sensing images and high resolution remote sensing images.

[0069] Optionally, the oil well identification and extraction device further includes: a construction module for: constructing a training dataset, including:

[0070] Examine the spectral band information of all the remote sensing images and select all common spectral bands to maximize the use of the collected remote sensing images and make the spectral resolution of all the remote sensing images consistent.

[0071] The residual channel attention network is used to perform super-resolution on the low spatial resolution remote sensing image in the remote sensing image after the spectral resolution is consistent, so as to obtain the super-resolution remote sensing image. The resolution of the super-resolution remote sensing image is consistent with the resolution of the high resolution remote sensing image.

[0072] Super-resolution remote sensing images and high-resolution remote sensing images are used to form a training dataset.

[0073] Optionally, the oil well site classification results can be divided into those including oil well sites and those excluding oil well sites;

[0074] If the classification result of the oil well site includes oil well sites, the oil well site extraction model also outputs mask images of all oil well sites.

[0075] The oil well identification and extraction device provided in this embodiment can perform the above-described method, and its implementation principle and technical effect are similar, so they will not be described again here.

[0076] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0077] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

Claims

1. An oil well identification extraction method characterized by, The method comprises: acquiring a to-be-identified remote sensing image, the to-be-identified remote sensing image being an image after super-resolution and band selection; inputting the to-be-identified remote sensing image into an oil well site extraction model to output an oil well site classification result, the oil well site classification result being classified into an oil well site and no oil well site; if the oil well site classification result is the oil well site, the oil well site extraction model further outputs a mask image of all oil well sites; wherein the oil well site extraction model adopts a Mask R-CNN model introducing a D-LinkNet network, the D-LinkNet network adopts an encoder-decoder structure, a hollow convolution and a pre-trained encoder for a road extraction task; the D-LinkNet network comprises a semantic segmentation branch, and the semantic segmentation branch is used for extracting oil well sites and road information connected with the oil well sites; a loss function of the oil well site extraction model comprises a IoU loss function of the semantic segmentation branch, and the loss function is: Among them, the first three are multi-classification cross-entropy loss function, bounding box regression parameter loss function, and mask information prediction loss function, is the intersection over union loss function of the semantic segmentation branch.

2. The method of claim 1, wherein, training data sets used for training the oil well site extraction model are multi-source remote sensing image data sets.

3. The method of claim 2, wherein, The multi-source remote sensing image data sets comprise remote sensing images with different spatial and spectral resolutions, wherein the remote sensing images with different spatial resolutions comprise low spatial resolution remote sensing images and high resolution remote sensing images.

4. The method of claim 3, wherein, The training data sets are constructed, comprising: checking spectral band information of all the remote sensing images and selecting all common spectral bands to maximize utilization of collected remote sensing images, so that spectral resolutions of all the remote sensing images are consistent; performing super-resolution on the low spatial resolution remote sensing images in the remote sensing images with consistent spectral resolutions by using a residual channel attention network to obtain super-resolution remote sensing images, the super-resolution remote sensing images having consistent resolutions with the high resolution remote sensing images; the super-resolution remote sensing images and the high resolution remote sensing images form the training data sets.

5. An oil well identification extraction apparatus characterized by comprising: The device comprises: an acquisition module configured to acquire a to-be-identified remote sensing image, the to-be-identified remote sensing image being an image after super-resolution and band selection; a processing module configured to input the remote sensing image into an oil well site extraction model to output an oil well site classification result and a mask image; wherein the oil well site extraction model adopts a Mask R-CNN network introducing a D-LinkNet network, the D-LinkNet network adopts an encoder-decoder structure, a hollow convolution and a pre-trained encoder for a road extraction task; the D-LinkNet network comprises a semantic segmentation branch, and the semantic segmentation branch is used for extracting oil well sites and road information connected with the oil well sites; a loss function of the oil well site extraction model comprises a IoU loss function of the semantic segmentation branch, and the loss function is: Among them, the first three are multi-classification cross-entropy loss function, bounding box regression parameter loss function, and loss function of mask information prediction, The intersection over union loss function for the semantic segmentation branch.

Images