Oil field discharge facility identification method and device based on small sample library driving

Abstract

The invention discloses an oil field discharge facility identification method and device based on small sample library driving. Comprising the following steps: acquiring different types of satellite remote sensing images acquired from a target oil field, and fusing the satellite remote sensing images to obtain a fused image; inputting the fused image into a pre-trained emission facility identification model, and detecting the fused image according to a preset sliding window through the emission facility identification model to obtain a plurality of prediction frames; determining whether the same oil field methane emission facility exists in the overlapping region of the adjacent prediction frames, and if yes, fusing the adjacent prediction frames to obtain a target prediction frame; and determining position information of the oil field discharge facility based on the target prediction frame. The method can be adapted to super-large remote sensing images, the recognition of the oil field emission facilities is expanded to the level of hundred million-level pixel images, the oil field methane emission facilities are recognized on the scale of multiplying ten thousand-level pixels by ten thousand-level pixels, and the recognition rate of the tiny emission facilities is improved.

Description

Technical Field

[0001] The present application relates to the field of oilfield emission facility detection, and in particular to a method and device for identifying oilfield emission facilities based on a small sample library. Background Art

[0002] Currently, identifying methane emission facilities in oil fields faces numerous challenges. First, oil fields are often geographically dispersed and contain a large number of facilities and equipment. These facilities may be scattered across vast areas, requiring large-scale remote sensing imagery, or they may be interspersed with dense urban buildings, resulting in little distinction between the target and the background, making identification extremely complex. Second, oil fields contain a wide variety of equipment, including drilling platforms, oil storage tanks, sewage tanks, and other structures. These devices vary in appearance and shape, and the methane emission characteristics of different devices may also vary, making the identification process difficult.

[0003] Traditional image processing methods for identifying oilfield methane emission facilities generally involve template matching based on the structural characteristics of the emission facilities. This method requires high professional knowledge, is time-consuming and labor-intensive, and cannot achieve rapid identification. Summary of the Invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the present application provides an oilfield emission facility identification method and device driven by a small sample library.

[0005] According to one aspect of an embodiment of the present application, a method for identifying oilfield discharge facilities based on a small sample library is provided, comprising: Acquiring different types of satellite remote sensing images acquired from the target oil field, and fusing the satellite remote sensing images to obtain a fused image; Inputting the fused image into a pre-trained emission facility recognition model, detecting the fused image according to a preset sliding window using the emission facility recognition method model to obtain multiple prediction frames; Determine whether there is a methane emission facility in the same oil field in the overlapping area of ​​adjacent prediction frames. If there is a methane emission facility in the same oil field, fuse the adjacent prediction frames to obtain the target prediction frame. The location information of the oilfield discharge facility is determined based on the target prediction frame.

[0006] Furthermore, the acquiring of different types of satellite remote sensing images acquired from the target oil field and fusing the satellite remote sensing images to obtain a fused image includes: Acquire panchromatic and multispectral images of the target oil field; The panchromatic image and the multispectral image are fused to obtain the fused image.

[0007] Furthermore, the fused image is input into a pre-trained emission facility recognition model, and the emission facility recognition method model detects the fused image according to a preset sliding window to obtain multiple prediction frames, including: Adjusting the impact size of the fused image to a target size, wherein the target size is the size of a preset sliding window used by the emission facility identification model; Determining the overlap corresponding to the preset sliding window; The emission facility recognition model is used to identify the fused image according to the preset sliding window and the overlap degree to obtain a plurality of prediction frames.

[0008] Furthermore, the training process of the emission facility identification model includes: Acquire a first fused image sample and annotation information corresponding to the first fused image sample, wherein the first fused image sample includes a plurality of oilfield discharge facilities, and the annotation information is used to annotate actual prediction boxes corresponding to the oilfield discharge facilities; Inputting the first fused image sample and the annotation information into an initial detection model, and identifying an initial prediction box corresponding to the oilfield discharge facility in the first fused image sample through the initial detection model; The training loss of the initial detection model is determined based on the initial prediction frame and the actual prediction frame, and the initial detection model is optimized based on the training loss until the training loss of the optimized initial detection model meets a preset condition.

[0009] Furthermore, the initial detection model includes: a backbone network, a connection network and a detection network, wherein the backbone network includes multiple parallel polarized self-attention modules, and the connection network includes multiple attention modules of different types.

[0010] Furthermore, the first fused image sample and the annotation information are input into an initial detection model, and an initial prediction frame corresponding to the oilfield discharge facility in the first fused image sample is identified by the initial detection model, including: Inputting the fused image into the backbone network; Identifying the fused image through the parallel polarization self-attention module of the backbone network, and transmitting the extracted initial feature information of the fused image to the connection network; Refining the initial feature information through a plurality of different types of attention modules in the connection network to obtain target feature information, and transmitting the target feature information to the detection network; The detection network outputs an initial prediction frame of the oilfield discharge facility based on the target feature information.

[0011] Furthermore, the method further comprises: acquiring a plurality of second fused image samples; Testing the initial detection model using the second fused image sample to obtain a prediction box corresponding to the oilfield discharge facility in the second fused image sample; When the confidence of the prediction box corresponding to the oilfield discharge facility in the second fused image sample is greater than the preset confidence, the second fused image sample is filled into the first fused image sample, and the prediction box corresponding to the oilfield discharge facility in the second fused image sample is used as annotation information.

[0012] According to another aspect of the embodiment of the present application, there is also provided an oilfield discharge facility identification device driven by a small sample library, comprising: An acquisition module is used to acquire different types of satellite remote sensing images acquired from the target oil field, and fuse the satellite remote sensing images to obtain a fused image; An input module, configured to input the fused image into a pre-trained emission facility recognition model, detect the fused image according to a preset sliding window using the emission facility recognition method model, and obtain a plurality of prediction frames; The detection module is used to determine whether there is a methane emission facility in the same oil field in the overlapping area of ​​adjacent prediction frames. If there is a methane emission facility in the same oil field, the adjacent prediction frames are fused to obtain the target prediction frame. A determination module is used to determine the location information of the oilfield discharge facility based on the target prediction frame.

[0013] According to another aspect of an embodiment of the present application, a storage medium is further provided, which includes a stored program, and the above steps are executed when the program is run.

[0014] According to another aspect of an embodiment of the present application, an electronic device is also provided, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; wherein: the memory is used to store computer programs; the processor is used to execute the steps in the above method by running the program stored in the memory.

[0015] An embodiment of the present application also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the steps in the above method.

[0016] The above-mentioned technical solution provided by the embodiment of the present application has the following advantages compared with the existing technology: the method provided by the embodiment of the present application can be adapted to ultra-large-scale remote sensing images, and expands the recognition of oilfield emission facilities to the level of billion-pixel images. It can identify oilfield methane emission facilities at the scale of ten thousand pixels by ten thousand pixels and improve the recognition rate of tiny emission facilities. BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0018] Figure 1 A flowchart of a method for identifying oilfield discharge facilities based on a small sample library provided in an embodiment of the present application; Figure 2 A schematic diagram of the detection results of an oilfield discharge facility provided in another embodiment of the present application; Figure 3 A schematic diagram of the overall network structure provided in another embodiment of the present application; Figure 4 A block diagram of an oilfield emission facility identification device driven by a small sample library provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another similar entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0021] This application provides a method and apparatus for identifying oilfield discharge facilities based on a small sample library. The method provided by this embodiment can be applied to any desired electronic device, such as a server or terminal. For ease of description, this is not specifically limited and will be referred to as an electronic device.

[0022] Example 1 According to one aspect of the embodiments of the present application, a method embodiment of an oilfield discharge facility identification method based on a small sample library is provided. Figure 1 A flowchart of an oilfield emission facility identification method based on a small sample library driven by an embodiment of the present application is provided. Figure 1 As shown, the method includes: Step S11 : acquiring different types of satellite remote sensing images acquired from the target oil field, and fusing the satellite remote sensing images to obtain a fused image.

[0023] In an embodiment of the present application, different types of satellite remote sensing images collected from a target oil field are obtained, and the satellite remote sensing images are fused to obtain a fused image, including: obtaining a panchromatic image and a multispectral image collected from the target oil field; and fusing the panchromatic image and the multispectral image to obtain a fused image.

[0024] It should be noted that satellite remote sensing images are obtained by capturing oilfields on the ground using satellite remote sensing systems. Such systems, such as the Gaofen-7 satellite, can acquire high-spatial-resolution remote sensing images. These images, including panchromatic and multispectral images, are used for monitoring and analyzing the Earth's surface. In oilfield applications, the fusion of panchromatic and multispectral images can improve the identification and monitoring of oilfield features.

[0025] The panchromatic and multispectral images used in this embodiment each have a size of 40,000 x 46,270 pixels. By fusing the panchromatic and multispectral images, a fused image with both high spatial and spectral resolution is obtained, preserving both high spatial resolution and the color information of the ground features. Experimental measurements have shown that this method can more accurately identify oilfield drainage facilities.

[0026] In step S12, the fused image is input into a pre-trained emission facility recognition model, and the fused image is detected by the emission facility recognition method model according to a preset sliding window to obtain multiple prediction frames.

[0027] In an embodiment of the present application, the fused image is input into a pre-trained emission facility recognition model, and the fused image is detected by the emission facility recognition method model according to a preset sliding window to obtain multiple prediction frames, including: adjusting the impact size of the fused image to the target size, wherein the target size is the size of the preset sliding window adopted by the emission facility recognition model; determining the overlap corresponding to the preset sliding window; and identifying the fused image according to the preset sliding window and overlap by the emission facility recognition model to obtain multiple prediction frames.

[0028] Specifically, the fused image is slid according to a preset sliding window size and overlap. Within each sliding window, the emission facility recognition model is used to predict the image within the window, resulting in a prediction frame within that window. After sliding the entire fused image, multiple prediction frames are obtained, representing areas at different locations where the model believes emission facilities may exist. This process allows the model to perform local target recognition on the entire fused image. By setting the sliding window and overlap, the entire image can be covered, improving the accuracy and comprehensiveness of emission facility detection. The resulting multiple prediction frames can be used for subsequent target fusion and processing to obtain the final recognition result.

[0029] It should be noted that this step selects the standard image size of the YOLOv5 input network as the sliding window size. To meet the detection network structure, images of any size will be first adjusted to 640 by 640 when entering the YOLOv5 model. Therefore, in order to improve the computing speed and accuracy of YOLOv5, the sliding window size is also set to 640 by 640, and oilfield methane emission facilities are extracted on each sliding window image, thereby improving the detection effect on ultra-large-scale remote sensing images.

[0030] Step S13: determine whether the overlapping area of ​​adjacent prediction frames contains the same oilfield methane emission facility. If so, the adjacent prediction frames are fused to obtain a target prediction frame.

[0031] In an embodiment of the present application, since the overlapping sliding window detection method will cause a certain degree of overlap between adjacent sliding window images, if there are oilfield methane emission facilities in the overlapping area, then these emission facilities will be identified and generated into prediction frames within these sliding windows. In response to this situation, the embodiment of the present application sets the overlap of the prediction frames, and uses the non-maximum suppression method to fuse the prediction frames with overlapping prediction frames, thereby increasing the accuracy of the prediction frames and obtaining the final extraction results.

[0032] It should be noted that when determining whether the overlapping regions of adjacent prediction boxes contain the same oilfield methane emission facility, "the same oilfield methane emission facility" refers to the target entities predicted within different sliding windows being the same actual object or target. When the prediction boxes within two adjacent sliding windows overlap and the predicted target class is the same, they can be considered to be different parts or perspectives of the same object or target.

[0033] Fusion of adjacent prediction boxes involves combining overlapping prediction boxes into a single, more accurate target prediction box, eliminating redundant and repeated predictions and improving the accuracy and consistency of prediction results. This fusion is typically achieved through methods such as non-maximum suppression (NMS). The NMS algorithm retains the prediction box with the highest confidence and removes any other prediction boxes that overlap with it. This fusion operation reduces duplicate predictions and retains only the most accurate target prediction box, thereby improving the accuracy and reliability of the object detection model.

[0034] Step S14: determining the location information of the oilfield discharge facility based on the target prediction frame.

[0035] In the examples of this application, the test results are as follows: Figure 2 As shown, the coordinates of the target prediction box are then used to determine the location of the oilfield discharge facility. Specifically, the coordinates of the target prediction box are converted from the image coordinate system to a geographic coordinate system. This typically requires the use of geolocation technologies such as the Global Positioning System (GPS) or Geographic Reference System (GRS). Through this conversion, the image location of the target prediction box can be mapped to actual longitude and latitude coordinates.

[0036] This embodiment of the present application combines Gaofen-7 satellite imagery data with an improved YOLOv5 model to accurately extract and locate oilfield methane emission facilities within the imagery, determining the coordinates of the emission facilities. The identified oilfield methane emission facilities are uniformly output in a 640x640 format, with the coordinates of each emission facility output as the horizontal and vertical coordinates of the upper left corner and the horizontal and vertical coordinates of the lower right corner. By importing the output images and coordinates of oilfield methane emission facilities of the same size into a database, an oilfield emission facility database can be constructed and continuously updated.

[0037] Example 2 The present application also provides a method for training an emission facility identification model, which includes: Step A1: Obtain a first fused image sample and annotation information corresponding to the first fused image sample, wherein the first fused image sample includes a plurality of oilfield discharge facilities, and the annotation information is used to annotate actual prediction boxes corresponding to the oilfield discharge facilities.

[0038] In the embodiment of the present application, this step uses images of oilfield methane emission facilities from various data sets such as AID, NWPU-RESISC45, RSOD-Dataset, DIOR, and LEVIR. After screening out suitable images, the LabelImg tool in Anaconda is used to perform label annotation to generate a training set. The training set is then preprocessed using Mosaic data enhancement and geometric distortion methods to obtain the final first fused image sample. Subsequently, by passing the image into the detection model for learning, the background of the detected object can be greatly enriched. The geometric distortion method performs random scaling, cropping, flipping, rotation, and other operations on the training images, thereby increasing the complexity of the training samples. The significance of data enhancement is mainly to expand the oilfield emission facility dataset so that the model has higher robustness to emission facility images obtained under different environments.

[0039] Step A2: inputting the first fused image sample and the annotation information into an initial detection model, and identifying an initial prediction frame corresponding to the oilfield discharge facility in the first fused image sample through the initial detection model; In the embodiment of the present application, the initial detection model includes: Figure 3 As shown, a backbone network, a connection network and a detection network, wherein the backbone network includes multiple parallel polarization self-attention modules, and the connection network includes multiple attention modules of different types.

[0040] In the embodiment of the present application, the initial detection model is the YOLOv5x model. In the backbone network of the YOLOv5x model, a ParallelPolarized self-attention module is constructed (see Figure 3 ), the ParallelPolarized self-attention module replaces the original C3 module in the second, fourth, sixth and eighth layers.

[0041] In the connection network (Neck) of the YOLOv5x model, three attention modules with SimAM attention model as the core and three attention modules with Dynamic Conv as the core are constructed (see Figure 3The SimAM module consists of a Conv layer, an upsampling layer, a Concat layer, three C3 layers, and a SimAM layer. The Dynamic Conv module consists of a DynamicConv layer, three C3 layers, and a SimAM layer. In this model, the SimAM attention module occupies layers 10 to 14, 15 to 19, and 20 to 24, while the Dynamic Conv attention module occupies layers 25 to 29, 30 to 34, and 35 to 39.

[0042] In the detection network (Head) of the YOLOv5x model, a new detection head is added, and layers 24, 29, 34, and 39 of the model are used as the output layers of the model.

[0043] The present embodiment improves the YOLOv5 model by integrating the Dynamic Conv module into YOLOv5, making the model more adaptable, with higher representation capabilities and accuracy. The SimAM and ParallelPolarized attention mechanism modules are integrated into YOLOv5, helping the network find regions of interest in images with large area coverage. Furthermore, a new prediction head is added to the original three prediction heads to detect objects at a wider range of scales.

[0044] In step A3, the training loss of the initial detection model is determined based on the initial prediction box and the actual prediction box, and the initial detection model is optimized based on the training loss until the training loss of the optimized initial detection model meets the preset conditions.

[0045] In an embodiment of the present application, a first fused image sample and annotation information are input into an initial detection model, and an initial prediction frame corresponding to the oilfield discharge facility in the first fused image sample is identified through the initial detection model, including: inputting the fused image into the backbone network; identifying the fused image through the parallel polarized self-attention module (ParallelPolarized) of the backbone network, and passing the extracted initial feature information of the fused image to the connection network; refining the initial feature information through multiple different types of attention modules in the connection network (an attention module with the SimAM attention model as the core and an attention module with Dynamic Conv as the core) to obtain target feature information, and passing the target feature information to the detection network; outputting the initial prediction frame of the oilfield discharge facility based on the target feature information through the detection network.

[0046] The fused image is fed into a backbone network, where its parallel polarization self-attention module performs feature extraction and recognition. The initial feature information extracted from the fused image is then passed to a connection network, where different types of attention modules refine this initial feature information to obtain target feature information. Finally, this target feature information is passed to a detection network, which outputs an initial prediction box for oilfield discharge facilities based on this target feature information. This method can improve the recognition and positioning accuracy of oilfield discharge facilities.

[0047] It's important to note that SimAM, as a simple, efficient, and lightweight module, can be integrated into most layers of the YOLOv5 network. Unlike existing channel and spatial attention modules, SimAM infers 3D attention weights for feature maps in a layer without adding parameters to the YOLOv5 network, adaptively refining features and significantly improving model performance. Gaofen-7 images cover a vast area and contain complex geographic features, making it challenging to distinguish between objects and background. The SimAM attention module extracts attention regions, helping YOLOv5 to withstand the clutter of geographic information and focus on the target object.

[0048] The Dynamic Conv module adaptively adjusts convolution parameters based on the input image, increasing the model's expressive power without increasing the network's depth or width. In YOLOv5, the Dynamic Conv module makes YOLOv5 more adaptable, adaptively generating convolution kernels, making it more suitable for processing large-scale inputs such as Gaofen-7 images. Furthermore, the Dynamic Conv module enhances YOLOv5's representational capabilities, enabling it to better extract features from input data. Combining these two features, the Dynamic Conv module enables YOLOv5 to achieve higher accuracy and better adapt to varying inputs, helping to reduce YOLOv5 model parameters and training time.

[0049] In addition, a new prediction head is added which is generated by low-level and high-resolution feature maps (see Figure 3 The smallest detector head in the bottom layer (in the image above) is more sensitive to tiny targets within oilfield methane emission facilities and can accurately locate them in dense scenes. Adding additional detector heads significantly improves detection performance for tiny oilfield methane emission facilities, despite the increased computational and memory costs.

[0050] In an embodiment of the present application, the method further includes: acquiring a plurality of second fused image samples; The initial detection model is tested using the second fused image sample to obtain the prediction box corresponding to the oilfield discharge facilities in the second fused image sample; when the confidence of the prediction box corresponding to the oilfield discharge facilities in the second fused image sample is greater than the preset confidence, the second fused image sample is filled into the first fused image sample, and the prediction box corresponding to the oilfield discharge facilities in the second fused image sample is used as annotation information.

[0051] It should be noted that after the model trains prediction frames for the second fused image sample used for testing, the correct prediction frames are screened and added to the sample library along with their corresponding label information as new samples. With each update of the sample library, the model is optimized. This step is repeated until the sample size is appropriately expanded and the imagery of the model can be ignored, resulting in the optimal YOLOv5 model. This step can be used as a valuable method for expanding dataset capacity, effectively addressing the limited sample size of oilfield methane emission facilities.

[0052] The embodiment of the present application utilizes the YOLOv5 model as the basic architecture, integrates the Dynamic Convolution, SimAM and ParallelPolarized attention modules, and adds a prediction head for detecting small targets. In terms of detection methods, the present invention also integrates a series of effective means such as data enhancement, real target box retraining, overlapping sliding window detection and prediction box fusion, etc., so that it can realize the accurate identification and positioning of oilfield methane emission facilities on the Gaofen-7 ultra-large image based on a small sample library. The present invention can improve the quantity and quality of the oilfield emission facility sample library and accurately, quickly and cost-effectively identify oilfield methane emission facilities in high-resolution remote sensing images, thereby generating an oilfield methane emission facility database, providing key support for the compilation of an oilfield methane emission inventory.

[0053] Example 3 According to one aspect of the embodiments of the present application, a method embodiment of an oilfield discharge facility identification method based on a small sample library is provided, specifically comprising: Step S1: Collect and annotate a database of existing oilfield methane emission facilities worldwide. Preprocess the annotated sample library using a geometric distortion method. After the sample image enters the YOLOv5 network, perform mosaic data enhancement on it. Step S2: In the Backbone part of the YOLOv5x model, the original C3 module is replaced with the ParallelPolarized self-attention module in the second, fourth, sixth, and eighth layers; In step S3, three attention modules based on the SimAM attention model and one based on the DynamicConv attention model are constructed in the Neck section of the YOLOv5x model. The SimAM module contains a Conv layer, an upsampling layer, a Concat layer, three C3 layers, and a SimAM layer. The DynamicConv module contains a DynamicConv layer, three C3 layers, and a SimAM layer. In step S4, a new detection head is added to the Head part of the YOLOv5x model, and layers 24, 29, 34, and 39 of the model are used as the output layers of the model; Step S5: The real target frames and label information obtained from the training set and test set after model training are used as training samples and then re-entered into the model for training. This can increase the sample capacity to a certain extent and improve the training effect. Step S6, performing image fusion processing on the GF-7 image, fusing the panchromatic image and the multispectral image to obtain a fused image with high spatial resolution and high spectral resolution; Step S7: When the GF-7 fused image enters the model, an overlapping sliding window detection strategy is adopted, a sliding window of 640 x 640 is selected, and an overlap degree of 20% is used to perform sliding detection on the GF-7 image; Step S8: Perform a non-maximum suppression method on the image generated by the sliding window detection to fuse the prediction boxes of the same target on different sliding windows in the overlapping area to obtain the final extraction result; Step S9: Output the coordinates and slices of the oilfield methane emission facilities identified by the Gaofen-7 image to construct an oilfield methane emission facility database.

[0054] Figure 4 This is a block diagram of an oilfield emission facility identification device based on a small sample library provided in an embodiment of the present application. The device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. Figure 4 As shown, the device includes: An acquisition module 41 is used to acquire different types of satellite remote sensing images acquired from the target oil field and fuse the satellite remote sensing images to obtain a fused image; An input module 42 is used to input the fused image into a pre-trained emission facility recognition model, and detect the fused image according to a preset sliding window using the emission facility recognition method model to obtain multiple prediction boxes; A detection module 43 is used to determine whether the overlapping area of ​​adjacent prediction frames contains the same oilfield methane emission facility. If so, the adjacent prediction frames are merged to obtain a target prediction frame. The determination module 44 is configured to determine the location information of the oilfield discharge facility based on the target prediction frame.

[0055] In the embodiment of the present application, the acquisition module 41 is used to acquire a panchromatic image and a multispectral image acquired from a target oil field; and fuse the panchromatic image and the multispectral image to obtain a fused image.

[0056] In an embodiment of the present application, the input module 42 is used to adjust the impact size of the fused image to the target size, where the target size is the size of the preset sliding window adopted by the emission facility recognition model; determine the overlap corresponding to the preset sliding window; and identify the fused image according to the preset sliding window and overlap by the emission facility recognition model to obtain multiple prediction frames.

[0057] In an embodiment of the present application, the device also includes: a training module, used to obtain a first fused image sample and annotation information corresponding to the first fused image sample, wherein the first fused image sample includes multiple oilfield discharge facilities, and the annotation information is used to annotate the actual prediction frame corresponding to the oilfield discharge facility; input the first fused image sample and the annotation information into the initial detection model, and identify the initial prediction frame corresponding to the oilfield discharge facility in the first fused image sample through the initial detection model; determine the training loss of the initial detection model based on the initial prediction frame and the actual prediction frame, and optimize the initial detection model based on the training loss until the training loss of the optimized initial detection model meets the preset conditions.

[0058] In an embodiment of the present application, the initial detection model includes: a backbone network, a connection network and a detection network, wherein the backbone network includes multiple parallel polarized self-attention modules, and the connection network includes multiple attention modules of different types.

[0059] In the embodiment of the present application, the training module is used to input the fused image into the backbone network; The fused image is recognized through the parallel polarization self-attention module of the backbone network, and the initial feature information of the extracted fused image is passed to the connection network; the initial feature information is refined by multiple different types of attention modules in the connection network to obtain target feature information, and the target feature information is passed to the detection network; the detection network outputs the initial prediction frame of the oilfield emission facilities based on the target feature information.

[0060] In an embodiment of the present application, the device also includes: a sample expansion module for obtaining multiple second fused image samples; using the second fused image samples to test the initial detection model to obtain a prediction box corresponding to the oilfield discharge facilities in the second fused image sample; when the confidence of the prediction box corresponding to the oilfield discharge facilities in the second fused image sample is greater than a preset confidence, the second fused image sample is filled into the first fused image sample, and the prediction box corresponding to the oilfield discharge facilities in the second fused image sample is used as annotation information.

[0061] The present application also provides an electronic device, such as Figure 5 As shown, the electronic device may include: a processor 1501 , a communication interface 1502 , a memory 1503 and a communication bus 1504 , wherein the processor 1501 , the communication interface 1502 , and the memory 1503 communicate with each other via the communication bus 1504 .

[0062] Memory 1503, used for storing computer programs; The processor 1501 is configured to implement the steps of the above embodiment when executing the computer program stored in the memory 1503 .

[0063] The communication bus mentioned in the terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or only one type of bus.

[0064] The communication interface is used for communication between the above terminal and other devices.

[0065] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage. Alternatively, the memory may be at least one storage device located away from the processor.

[0066] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0067] In another embodiment provided by the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute any of the methods described in the above embodiments.

[0068] In another embodiment provided by the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute any of the methods described in the above embodiments.

[0069] In the above embodiments, all or part of the embodiments can be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, hard disk, tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive).

[0070] The above description is only a preferred embodiment of the present application and is not intended to limit the scope of protection of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application are included in the scope of protection of the present application.

[0071] The foregoing is merely a list of specific embodiments of the present application, intended to enable those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the broadest scope consistent with the principles and novel features of the present application.

Claims

1. A method for identifying oilfield discharge facilities based on a small sample library, characterized in that: include: Acquiring different types of satellite remote sensing images acquired from the target oil field, and fusing the satellite remote sensing images to obtain a fused image; Inputting the fused image into a pre-trained emission facility recognition model, detecting the fused image according to a preset sliding window using the emission facility recognition method model to obtain multiple prediction frames; Determine whether there is a methane emission facility in the same oil field in the overlapping area of ​​adjacent prediction frames. If there is a methane emission facility in the same oil field, fuse the adjacent prediction frames to obtain the target prediction frame. The location information of the oilfield discharge facility is determined based on the target prediction frame.

2. The method according to claim 1, characterized in that The acquiring of different types of satellite remote sensing images acquired from the target oil field and fusing the satellite remote sensing images to obtain a fused image includes: Acquire panchromatic and multispectral images of the target oil field; The panchromatic image and the multispectral image are fused to obtain the fused image.

3. The method according to claim 1, characterized in that The fused image is input into a pre-trained emission facility recognition model, and the emission facility recognition method model detects the fused image according to a preset sliding window to obtain multiple prediction frames, including: Adjusting the impact size of the fused image to a target size, wherein the target size is the size of a preset sliding window used by the emission facility identification model; Determining the overlap corresponding to the preset sliding window; The emission facility recognition model is used to identify the fused image according to the preset sliding window and the overlap degree to obtain a plurality of prediction frames.

4. The method according to claim 3, characterized in that The training process of the emission facility identification model includes: Acquire a first fused image sample and annotation information corresponding to the first fused image sample, wherein the first fused image sample includes a plurality of oilfield discharge facilities, and the annotation information is used to annotate actual prediction boxes corresponding to the oilfield discharge facilities; Inputting the first fused image sample and the annotation information into an initial detection model, and identifying an initial prediction box corresponding to the oilfield discharge facility in the first fused image sample through the initial detection model; The training loss of the initial detection model is determined based on the initial prediction frame and the actual prediction frame, and the initial detection model is optimized based on the training loss until the training loss of the optimized initial detection model meets a preset condition.

5. The method according to claim 4, characterized in that The initial detection model includes: a backbone network, a connection network and a detection network, wherein the backbone network includes multiple parallel polarization self-attention modules, and the connection network includes multiple attention modules of different types.

6. The method according to claim 5, characterized in that Inputting the first fused image sample and the annotation information into an initial detection model, and identifying an initial prediction frame corresponding to the oilfield discharge facility in the first fused image sample by the initial detection model, including: Inputting the fused image into the backbone network; Identifying the fused image through the parallel polarization self-attention module of the backbone network, and transmitting the extracted initial feature information of the fused image to the connection network; Refining the initial feature information through a plurality of different types of attention modules in the connection network to obtain target feature information, and transmitting the target feature information to the detection network; The detection network outputs an initial prediction frame of the oilfield discharge facility based on the target feature information.

7. The method according to claim 4, characterized in that The method further comprises: acquiring a plurality of second fused image samples; Testing the initial detection model using the second fused image sample to obtain a prediction box corresponding to the oilfield discharge facility in the second fused image sample; When the confidence of the prediction box corresponding to the oilfield discharge facility in the second fused image sample is greater than the preset confidence, the second fused image sample is filled into the first fused image sample, and the prediction box corresponding to the oilfield discharge facility in the second fused image sample is used as annotation information.

8. An oilfield discharge facility identification device driven by a small sample library, characterized in that: include: An acquisition module is used to acquire different types of satellite remote sensing images acquired from the target oil field, and fuse the satellite remote sensing images to obtain a fused image; An input module, configured to input the fused image into a pre-trained emission facility recognition model, detect the fused image according to a preset sliding window using the emission facility recognition method model, and obtain a plurality of prediction frames; The detection module is used to determine whether there is a methane emission facility in the same oil field in the overlapping area of ​​adjacent prediction frames. If there is a methane emission facility in the same oil field, the adjacent prediction frames are fused to obtain the target prediction frame. A determination module is used to determine the location information of the oilfield discharge facility based on the target prediction frame.

9. A storage medium, characterized in that: The storage medium includes a stored program, wherein the program executes the method steps of any one of claims 1 to 7 when running.

10. An electronic device, characterized in that: The system comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; wherein: Memory for storing computer programs; A processor, configured to execute the method steps according to any one of claims 1 to 7 by running a program stored in a memory.

Images