Deep learning oil field detection method and system based on bimodal image fusion

Abstract

The invention relates to the technical field of oil field monitoring, in particular to a deep learning oil field detection method and system based on bimodal image fusion. The method comprises the steps of obtaining an oil field area image; preprocessing the acquired oil field area image; constructing an oil field oil leakage detection network model; performing feature extraction on the oil field area image by using the oil field oil leakage detection network model; carrying out feature fusion on the extracted features; and detecting the fused features based on a target detection algorithm, and outputting a target detection result. The oil leakage detection method provided by the invention has the greatest advantages that visible light and infrared bimodal images are creatively fused, and complementary information in the two images is automatically mined and integrated through a deep learning technology.

Description

Technical Field

[0001] The present invention relates to the technical field of oilfield monitoring, and in particular to a deep learning oilfield detection method and system based on dual-modal image fusion. Background Art

[0002] With the continuous advancement of technology and the deepening of international cooperation, the exploration and development capabilities of oil fields have been significantly improved, and the oil and gas production has grown steadily. In particular, the exploration and development technology of deep and ultra-deep oil and gas resources has become increasingly mature, and the development of offshore oil and gas resources has also made significant progress. At the same time, the application of intelligent and digital technologies has not only improved production efficiency, but also reduced operating costs, laying a solid foundation for the sustainable development of the oil industry.

[0003] Due to the complex geological conditions of oil fields, the harsh production environment, and human errors, oil spills still occur from time to time despite the huge amount of resources and energy invested in preventing and responding to oil spills. These incidents not only cause serious pollution to the ecological environment, but also affect the image of the oil industry. In recent years, with the continuous optimization of artificial intelligence algorithms and the continuous improvement of computing power, in terms of oil spill prevention, artificial intelligence can use big data analysis to monitor and predict various parameters in the oil field production process in real time, and identify potential risk points that may cause oil spills in advance. With the help of machine learning models, the system can continuously learn and optimize the prediction algorithm to make preventive measures more accurate and effective. At the same time, artificial intelligence can also assist in the design of a more scientific and reasonable oil field layout and production process, fundamentally reducing the incidence of oil spills. In terms of oil spill detection, artificial intelligence technology uses intelligent equipment such as drones and robots, combined with advanced technologies such as high-definition imaging and infrared thermal imaging, to achieve all-round and all-weather monitoring of oil field areas. Once an oil spill occurs, the artificial intelligence system can quickly locate the source of the oil spill, assess the scope of pollution, and automatically trigger the emergency response mechanism.

[0004] At present, although the technology of visible light image recognition of oil spills is widely used in oilfield monitoring, its technical defects are becoming increasingly prominent. Visible light images rely on good lighting conditions. Once it encounters night, rainy days or insufficient light, the image quality will drop significantly, making it difficult to clearly identify the oil spill area, thereby increasing the risk of false detection and missed detection. In addition, the interference of natural factors such as water surface reflection, shadows and floating objects also challenges the accuracy of visible light images in oil spill detection. These factors limit the effectiveness and reliability of visible light image recognition of oil spill technology. In contrast, infrared image recognition of oil spill technology can overcome the limitations of lighting conditions to a certain extent. Infrared images are based on the thermal radiation characteristics of objects and have a high sensitivity to thermal anomaly areas such as oil pollution. However, infrared image recognition of oil spill technology also has its inherent technical defects. First, when the temperature of the oil pollution is close to the ambient temperature, the thermal anomaly signal on the infrared image will become weak or even disappear, resulting in poor detection effect. Secondly, the spatial resolution of infrared images is relatively low, and it is difficult to provide detailed information comparable to visible light images, which to a certain extent limits its application scope in oil spill detection. Summary of the invention In order to solve the above-mentioned problems, the present invention provides a deep learning oilfield detection method and system based on dual-modal image fusion. The algorithm combines the advantages of visible light images and infrared images, and automatically learns and extracts key features in the two images through a deep learning algorithm. In the feature extraction stage, the deep learning model can automatically learn and extract key feature information in the two images, such as color, texture, shape, and thermal radiation characteristics, etc., to achieve effective integration of complementary information, thereby improving the robustness and accuracy of oil leak detection. In the feature fusion stage, these feature information are effectively integrated to form a more comprehensive and accurate oil leak detection model. By fusing the high-resolution texture information of visible light images and the temperature anomaly detection capability of infrared images, accurate and efficient monitoring of oil leaks in oilfield areas is achieved. The algorithm not only overcomes the limitations of a single image modality, but also significantly improves the automation level and real-time performance of oil leak detection, providing strong technical support for oilfield production safety and environmental protection.

[0005] In a first aspect, the present invention provides a deep learning oilfield detection method based on dual-modal image fusion, which adopts the following technical solution: A deep learning oilfield detection method based on dual-modal image fusion, comprising: Acquire oilfield area images; Preprocessing the acquired oilfield area images; Constructing oil field leak detection network model; The oil field leakage detection network model is used to extract features from oil field area images; Perform feature fusion on the extracted features; The fused features are detected based on the target detection algorithm and the target detection results are output.

[0006] Furthermore, the preprocessing of the acquired oilfield area image includes image normalization and image filtering to remove noise.

[0007] Furthermore, the constructing of the oilfield oil leak detection network model includes respectively constructing an RGB-attention network structure and an IR-attention network structure.

[0008] Furthermore, the use of the oilfield leak detection network model to extract features of the oilfield area image includes using an RGB-attention network structure to acquire feature information of the visible light image part of the oilfield area image; and using an IR-attention network structure to acquire feature information of the infrared light image part of the oilfield area image.

[0009] Furthermore, the feature fusion of the extracted features includes matching and fusing the depth feature information of the visible light image and the infrared light image using a multi-scale fusion mechanism to generate fused features containing multiple amounts of information and multi-scale characteristics, while retaining unique information under the two light sources.

[0010] Furthermore, the detection of the fused features based on the target detection algorithm includes detecting the fused features using a target detection model, wherein the target detection model is composed of a backbone network, a neck network and a detection head, and the detection head is used to perform target detection tasks and generate information including target category and coordinate position.

[0011] Furthermore, the use of the target detection model to detect the fused features also includes using three parallel detection heads, which correspond to the features of the IR-attention network structure, the RGB-attention network structure and the feature fusion part respectively, integrating the features passing through the RGB-attention network structure into the first detection head, integrating the features passing through the IR-attention network structure into the second detection head, and integrating the features after feature fusion into the third detection head.

[0012] In the second aspect, a deep learning oilfield detection system based on dual-modal image fusion includes: The data acquisition module is configured to acquire an image of the oilfield area; The preprocessing module is configured to preprocess the acquired oilfield area image; The model building module is configured to build an oil field oil leak detection network model; The feature extraction module is configured to extract features from the oil field area image using the oil field oil leak detection network model; The feature fusion module is configured to perform feature fusion on the extracted features; The target detection module is configured to detect the fused features based on the target detection algorithm and output the target detection result.

[0013] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, a deep learning oilfield detection method based on dual-modal image fusion.

[0014] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being suitable for being loaded by the processor and executing the deep learning oilfield detection method based on dual-modal image fusion.

[0015] In summary, the present invention has the following beneficial technical effects: The biggest advantage of the oil leak detection method proposed in this invention is that it innovatively integrates visible light and infrared dual-modal images, and automatically mines and integrates the complementary information in the two images through deep learning technology. This algorithm not only makes full use of the high-resolution texture details of visible light images, but also combines the infrared image's keen ability to capture temperature anomalies, achieving accurate identification and efficient monitoring of oil leaks in oilfield areas. This innovation not only breaks through the limitations of a single image modality in oil leak detection, but also greatly improves the degree of automation and real-time response speed of detection, building a solid technical defense line for safe production and environmental protection in oilfields. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a schematic diagram of a deep learning oilfield detection method based on dual-modal image fusion according to Embodiment 1 of the present invention; Figure 2 is a schematic diagram of the RGB-attention network structure of Example 1 of the present invention; Figure 3 It is a schematic diagram of the IR-attention network structure of Example 1 of the present invention. DETAILED DESCRIPTION

[0017] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0018] Example 1 Reference Figure 1 , a deep learning oilfield detection method based on dual-modal image fusion in this embodiment includes: Acquire oilfield area images; Preprocessing the acquired oilfield area images; Constructing oil field leak detection network model; The oil field leakage detection network model is used to extract features from oil field area images; Perform feature fusion on the extracted features; The fused features are detected based on the target detection algorithm and the target detection results are output.

[0019] Specifically, the following steps are included: S1. Obtain images of the oilfield area; install fixed cameras at key locations in the oilfield area, such as near oil wells, pumping stations, pipelines and other facilities. When selecting the installation location, factors such as wide field of view, no obstruction, and coverage of key areas should be considered. For example, a camera can be installed above the wellhead of an oil well to monitor the production of the oil well in real time; a camera can be installed at the inlet and outlet of a pumping station to monitor the operating status of the pumping station; and a camera can be installed at key nodes of a pipeline to monitor pipeline leakage.

[0020] S2. Preprocessing the acquired oilfield area image; including image normalization and image filtering to remove noise.

[0021] Among them, the image is normalized: Normalization is to map the range of image pixel values ​​to a standard interval. For images with a pixel value range of 0-255, the use of linear transformation methods can prevent certain feature extraction algorithms from being affected by large differences in pixel value ranges, thereby improving the stability and accuracy of the algorithm.

[0022] Image filtering: A mean filter is used to reduce noise in an image. The value of each pixel is replaced by the average of the values ​​of its neighboring pixels. For a 3x3 filter window, the value of the center pixel will be replaced by the average of the 9 pixel values ​​within this window. This method can effectively smooth the image. Median filtering is used to replace the value of each pixel with the median value of its neighboring pixel values. For example, for a 3x3 filter window, the 9 pixel values ​​in the window are sorted, and then the middle value is taken as the new value of the center pixel. Median filtering is very effective in removing impulse noise such as salt and pepper noise, and can preserve the edges of the image to a certain extent.

[0023] S3. Construct oil field oil leak detection network model; Among them, an oil field oil leak detection network model is constructed, including constructing RGB-attention network structure and IR-attention network structure respectively.

[0024] Constructing the RGB-attention network structure: The RGB-attention network structure is mainly composed of two 1×1 convolution kernels, one 1×3 convolution kernel, one 3×3 convolution kernel, and multiple ReLu function layers. For the input visible light image part, first use it as the parameter Mc. Then process it according to the following formula: M_(c+1)=M_c+〖F_(1×1) (F〗 (1×3) (F (1×1) (M_c))) + F_(1×1) 〖〖(F〗 (3× 3) (F〗 (1×3) 〖(F〗_(1×1) (M_c)))).

[0025] Among them, F_(1×1) represents a 1×1 convolution kernel, F_(1×3) represents a 1×3 convolution kernel, and F_(3×3) represents a 3×3 convolution kernel. Through the combination of these convolution kernels and functions, the local important information of the visible light image is gradually extracted.

[0026] Constructing the IR-attention network structure: The IR-attention network structure is similar to the RGB-attention network structure, and is also composed of two 1×1 convolution kernels, one 1×3 convolution kernel, one 3×3 convolution kernel, and multiple ReLu function layers. For the input infrared image part, it is used as the parameter M_d. Then according to the formula: M_(d+1)=M_d+〖F_(1×1) (F〗 (1×3) (F (1×1) (M_d))) + 〖F_(1×1) (F〗 (3× 3) 〖(F〗 (1×1) (M_d))) is processed.

[0027] Similarly, F_(1×1), F_(1×3), and F_(3×3) represent the corresponding convolution kernels, respectively, and these operations are used to obtain important local information of the infrared image.

[0028] S4. Use the oil field leak detection network model to extract features from the oil field area image; including using the RGB-attention network structure to obtain feature information from the visible light image part of the oil field area image; and using the IR-attention network structure to obtain feature information from the infrared light image part of the oil field area image.

[0029] The network structure is as follows: Figure 1 As shown in Figure 2, the network structure consists of two parts: feature extraction and target detection.

[0030] (1) RGB-attention The network structure of RGB-attention is as follows Figure 2 As shown in the figure, this structure is responsible for obtaining important local information of the visible light image. This part consists of two 1×1 convolution kernels, a 1×3 convolution kernel, a 3×3 convolution kernel, and multiple ReLu function layers.

[0031] The formula expression of this module is as follows: in, Indicates the parameters before inputting RGB-attention, Represents the parameters after inputting RGB-attention, represents a 1×1 convolution kernel, represents a 1×3 convolution kernel, Represents a 3×3 convolution kernel. (2) IR-attention The network structure of IR-attention is as follows Figure 3 As shown in the figure, this structure is responsible for obtaining important local information of the infrared image. This part consists of two 1×1 convolution kernels, a 1×3 convolution kernel, a 3×3 convolution kernel and multiple ReLu function layers.

[0032] The formula expression of this module is as follows: in, Indicates the parameters before inputting IR-attention. Indicates the parameters after inputting IR-attention. represents a 1×1 convolution kernel, represents a 1×3 convolution kernel, Represents a 3×3 convolution kernel.

[0033] S5. Perform feature fusion on the extracted features; including using a multi-scale fusion mechanism to match and fuse the depth feature information of the visible light image and the infrared light image, generating a fusion feature containing multiple amounts of information and multi-scale characteristics, while retaining the unique information under the two light sources.

[0034] Among them, in order to integrate the depth feature information of visible light images and infrared light images, the present invention adopts a feature fusion mechanism. The mechanism first matches and fuses the depth feature information under the two light sources to generate fusion features containing rich information and multi-scale characteristics. This process not only retains the unique information under each light source, but also enhances the comprehensiveness and robustness of the features through cross-modal complementarity. Subsequently, the multi-scale fusion features are fed back to the target detection part. In this process, the fusion features serve as high-level guidance signals to guide the target detection part to focus more accurately on the key information area and further optimize its extraction strategy. This feedback mechanism not only improves the efficiency and accuracy of target detection, but also enables the entire system to adaptively adjust to adapt to changes in different scenarios and needs.

[0035] The parameters of the two features are H×W×C, where H represents the image height, W represents the image width, and C represents the number of image channels. The two features are fused through channels, and the parameters of the spliced ​​features are H×W×2C. The spliced ​​features are then input into a convolution layer with a convolution kernel size of 5×5 for convolution operation. The feature map parameters become H×W×C. The features that have undergone channel dimension reduction are passed through the Sigmoid activation function to obtain the fused features.

[0036] The fusion process is shown in the following formula: in is the Sigmoid function. It is a 5×5 convolution. Perform channel fusion for two features. Represents the fused features.

[0037] S6. Detect the fused features based on the target detection algorithm and output the target detection results.

[0038] Among them, in the target prediction part of the present invention, multiple detection head modules operating in parallel are used, and these modules are closely connected with the feature extraction part of the present invention to realize the conversion from features to prediction information. Each detection head module receives the corresponding features from the feature extraction part and uses these features to perform the target detection task. After completing the detection, the detection head module will generate information including the target category and coordinate position, and output the final target detection result accordingly.

[0039] The target detection model is mainly composed of a backbone network, a neck network, and a detection head. The backbone network is mainly composed of a CSP-Darknet53+SPP network structure. The neck network adopts an FPN+PAN structure. There are three detection heads and they are set in parallel, corresponding to the features of the IR-attention part, the RGB-attention part, and the feature fusion part. The features that have passed the RGB-attention module are integrated into the first detection head, the features that have passed the IR-attention module are integrated into the second detection head, and the features that have passed the feature fusion are integrated into the third detection head. The detection head is used to obtain prediction information at different scales from the features, and generate the coordinate position of the target category set based on the prediction information at different scales to obtain the target detection result.

[0040] Example 2 This embodiment provides a deep learning oilfield detection system based on dual-modal image fusion, including: A deep learning oilfield detection system based on dual-modal image fusion, comprising: The data acquisition module is configured to acquire an image of the oilfield area; The preprocessing module is configured to preprocess the acquired oilfield area image; The model building module is configured to build an oil field oil leak detection network model; The feature extraction module is configured to extract features from the oil field area image using the oil field oil leak detection network model; The feature fusion module is configured to perform feature fusion on the extracted features; The target detection module is configured to detect the fused features based on the target detection algorithm and output the target detection result.

[0041] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for a deep learning oilfield detection method based on dual-modal image fusion.

[0042] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor to implement a deep learning oilfield detection method based on dual-modal image fusion.

[0043] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A deep learning oilfield detection method based on dual-modal image fusion, characterized in that: include: Acquire oilfield area images; Preprocessing the acquired oilfield area images; Constructing oil field leak detection network model; The oil field leakage detection network model is used to extract features from oil field area images; Perform feature fusion on the extracted features; The fused features are detected based on the target detection algorithm and the target detection results are output.

2. The deep learning oilfield detection method based on dual-modal image fusion according to claim 1 is characterized in that: The preprocessing of the acquired oilfield area image includes image normalization and image filtering to remove noise.

3. The deep learning oilfield detection method based on dual-modal image fusion according to claim 2 is characterized in that: The construction of the oilfield oil leakage detection network model includes respectively constructing an RGB-attention network structure and an IR-attention network structure.

4. The deep learning oilfield detection method based on dual-modal image fusion according to claim 3 is characterized in that: The oilfield oil leakage detection network model is used to extract features from oilfield area images, including using an RGB-attention network structure to obtain feature information from the visible light image part of the oilfield area image; and using an IR-attention network structure to obtain feature information from the infrared light image part of the oilfield area image.

5. The deep learning oilfield detection method based on dual-modal image fusion according to claim 4 is characterized in that: The feature fusion of the extracted features includes matching and fusing the depth feature information of the visible light image and the infrared light image using a multi-scale fusion mechanism to generate fused features containing multiple amounts of information and multi-scale characteristics, while retaining the unique information under the two light sources.

6. The deep learning oilfield detection method based on dual-modal image fusion according to claim 5 is characterized in that: The target detection algorithm is used to detect the fused features, including using a target detection model to detect the fused features, wherein the target detection model is composed of a backbone network, a neck network and a detection head, and the detection head is used to perform target detection tasks and generate information including target categories and coordinate positions.

7. The deep learning oilfield detection method based on dual-modal image fusion according to claim 6 is characterized in that: The method of using the target detection model to detect the fused features also includes using three parallel detection heads, which correspond to the features of the IR-attention network structure, the RGB-attention network structure and the feature fusion part respectively, integrating the features passing through the RGB-attention network structure into the first detection head, integrating the features passing through the IR-attention network structure into the second detection head, and integrating the features fused into the third detection head.

8. A deep learning oilfield detection system based on dual-modal image fusion, characterized in that: include: The data acquisition module is configured to acquire an image of the oilfield area; The preprocessing module is configured to preprocess the acquired oilfield area image; The model building module is configured to build an oil field oil leak detection network model; The feature extraction module is configured to extract features from the oil field area image using the oil field oil leak detection network model; The feature fusion module is configured to perform feature fusion on the extracted features; The target detection module is configured to detect the fused features based on the target detection algorithm and output the target detection result.

9. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the method according to claim 1 .

10. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the method as claimed in claim 1 .

Images