Multi-model combined infrared multispectral imaging method for insulating gas leakage

Abstract

The invention discloses a multi-model combined infrared multispectral imaging method for insulating gas leakage, belongs to the technical field of electric power safety detection, solves the problem of improving the accuracy of insulating gas leakage identification, and comprises the following steps: acquiring leaked insulating gas multispectral image data by adopting a multicomponent insulating gas multispectral imaging device; the method comprises the following steps of: classifying moving objects by using a YOLO-World model through a prompt of a user-defined application scene; the ROI of the leaked insulating gas is extracted at a semantic level by adopting an efficient SAM model, so that the gas cloud ROI is accurately extracted and rendered; training an SVM (Support Vector Machine) model to identify the leaked insulating gas; according to the method, multispectral information, semantic analysis and image processing are combined, interference of environmental change factors is more accurately and effectively removed, the false alarm rate is remarkably reduced, the gas cloud area is accurately extracted and classified, and the gas leakage recognition accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power safety detection, and relates to a multi-model combined infrared multi-spectral imaging method for insulating gas leakage. Background Art

[0002] With the continuous improvement of the voltage level of the power system and the continuous expansion of the equipment scale, the safe and stable operation of power equipment faces increasingly severe challenges. As an important insulating medium in power equipment, the state of insulating gas directly affects the operation safety of the equipment. Traditional insulating gas detection methods, such as gas chromatography, mass spectrometry, etc., although having high precision, have the disadvantages of long detection cycle, complex operation, and inability to achieve on-line monitoring, and are difficult to meet the requirements of real-time monitoring of the state of power equipment.

[0003] In recent years, infrared spectroscopy technology has been widely used in the field of gas detection due to its advantages of non-contact, fast, high sensitivity, etc. However, a single infrared spectroscopy model often has difficulty in accurately identifying and quantitatively analyzing complex mixed gases. Especially for this multi-component mixed system of insulating gas, there are the following problems:

[0004] Severe spectral overlap interference: The infrared absorption peaks of different gas molecules overlap, resulting in difficulty in distinguishing characteristic peaks and affecting the recognition accuracy.

[0005] Low concentration inversion accuracy: The relationship between gas concentration and infrared absorption intensity is affected by factors such as temperature and pressure, and a single model is difficult to accurately reflect this complex relationship.

[0006] Poor model generalization ability: A single model is highly dependent on training data and is difficult to adapt to the gas detection requirements under different environments and different equipment conditions.

[0007] Therefore, how to improve the accuracy and robustness of insulating gas identification and quantitative analysis by integrating the advantages of multiple models and provide a more reliable technical means for power equipment condition monitoring is a current research hotspot. Summary of the Invention

[0008] The technical solution of the present invention is used to solve the problem of improving the accuracy rate of insulating gas leakage identification.

[0009] The present invention solves the above technical problems through the following technical solutions:

[0010] The present invention provides a multi-model combined infrared multi-spectral imaging method for insulating gas leakage, including:

[0011] S1 Using a multi-component insulating gas multi-spectral imaging device to collect multi-spectral image data of the leaked insulating gas;

[0012] S2 classifies moving objects using the YOLO-World model through prompts for custom application scenarios;

[0013] S3 uses the efficient SAM model to extract the ROI of the leaked insulating gas at the semantic level, thereby accurately extracting and rendering the gas cloud ROI;

[0014] S4 trains an SVM model to identify the leaked insulating gas.

[0015] Furthermore, the multi-component insulating gas multi-spectral imaging device includes: a visible light camera (11), a laser ranging module (12), an infrared imaging lens (13), a multi-spectral filter wheel (14), an infrared imaging detector (15), a signal processing board (16), a lithium battery (17), and a flip-up display screen (18); the visible light camera (11) is used to collect video images of the power scene, providing a visible light scene background map for the infrared detection of insulating gas leakage on the one hand and expanding the multi-functional video monitoring function of the power scene on the other hand; the laser ranging module (12) is used to measure the distance between the device and the power equipment; the infrared imaging lens (13) is used for the field of view adjustment of the infrared imaging of insulating gas; multiple broadband filters are installed on the multi-spectral filter wheel (14) for extracting the infrared characteristics of multi-component insulating gas; the infrared imaging detector (15) is used to collect the infrared images of the power scene of multi-component insulating gas; the signal processing board (16) is used to process the data of the infrared imaging of multi-component insulating gas; the lithium battery (17) is used to supply power to the device; the flip-up display screen (18) is used to display the imaging results.

[0016] Furthermore, the method for classifying moving objects using the YOLO-World model is specifically as follows:

[0017] Input the multi-spectral image data of the leaked insulating gas into the YOLO detector to extract the multi-scale features of the image;

[0018] Input the multi-scale features of the image into the multi-scale image feature pyramid network to fuse the feature maps of different scales to form a feature pyramid;

[0019] The feature pyramid and the vocabulary are input into the vision-language path aggregation network together. At each stage of the feature pyramid, the vocabulary text is injected into the image features;

[0020] Through image perception embedding and region-text matching, the high-degree fusion of image features and text embedding is achieved, and the classification of moving objects is completed according to the similarity between the predicted bounding box and the text embedding.

[0021] Preferably, the YOLO-WORLD model is designed based on YOLOv8 and uses the deep convolutional neural network Darknet as the image encoder.

[0022] Furthermore, the method for accurately extracting and rendering the gas cloud ROI by using the efficient SAM model to extract the ROI of the leaked insulating gas at the semantic level is as follows:

[0023] Input the RBTF image of the leaked insulating gas and the infrared image of the leaked insulating gas into the efficient SAM model respectively;

[0024] Pre-train the efficient SAM model with the mask image, pre-train to learn to reconstruct features from the SAM image encoder, perform effective visual representation learning, and then use the lightweight image encoder and mask decoder pre-trained with the mask image to build the model;

[0025] Through the trained efficient SAM model, obtain the ROI of the gas in the RBTF image and the infrared image respectively.

[0026] Furthermore, the method for training the SVM model to identify the leaked insulating gas is as follows:

[0027] Extract the peak value and valley value from the multi-spectral data of the insulating gas as the features of the gas characteristics;

[0028] Use the extracted features to train the SVM model; during the training process, establish a training set and a validation set to optimize the performance of the SVM model;

[0029] Evaluate the performance of the SVM model through cross-validation, accuracy rate, and recall rate indicators;

[0030] Use the trained SVM model to classify and identify the spectral data of each pixel point within the determined image ROI, and determine the type of gas of each pixel according to the output result of the SVM model.

[0031] The present invention also provides an electronic device, including a memory and a processor. The memory is used to store a program that supports the processor to execute the above multi-model combined infrared multi-spectral imaging method for insulating gas leakage, and the processor is configured to execute the program stored in the memory.

[0032] The present invention also provides a storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of the above multi-model combined infrared multi-spectral imaging method for insulating gas leakage.

[0033] The beneficial effects of the present invention are as follows:

[0034] The method of the present invention divides data processing into two streams. One is for ratio imaging (gas filter radiation divided by background filter radiation) to extract the region of interest (ROI) of the changing area, and the other is for the ROI of moving objects. Through the prompt of the custom application scenario, the YOLO-World model is used to classify moving objects. In addition, the efficient SAM model extracts the ROI at the semantic level, enabling the precise extraction and rendering of the gas cloud ROI. Finally, a support vector machine (SVM) model is trained to classify specific types of gases present; the method of the present invention combines multi-spectral information, semantic analysis, and image processing to more precisely and effectively remove the interference of environmental change factors, significantly reduce the false alarm rate, accurately extract and classify the gas cloud area, and improve the accuracy of gas leakage recognition. Description of the Drawings

[0035] Figure 1 is the front three-dimensional view of the multi-component insulating gas multi-spectral imaging device in the first embodiment of the present invention;

[0036] Figure 2 is the rear three-dimensional view of the multi-component insulating gas multi-spectral imaging device in the first embodiment of the present invention;

[0037] Figure 3 is the internal structure three-dimensional view of the multi-component insulating gas multi-spectral imaging device in the first embodiment of the present invention;

[0038] Figure 4 is the overall flow chart of the multi-model combined infrared multi-spectral imaging method for insulating gas leakage in the second embodiment of the present invention;

[0039] Figure 5 is the method flow chart of using the YOLO-World model to classify moving objects in the multi-model combined infrared multi-spectral imaging method for insulating gas leakage in the second embodiment of the present invention;

[0040] Figure 6 is the method flow chart of using the efficient SAM model to extract the ROI of the leaking insulating gas at the semantic level in the multi-model combined infrared multi-spectral imaging method for insulating gas leakage in the second embodiment of the present invention;

[0041] Figure 7 is the method flow chart of training the SVM model to identify the leaking insulating gas in the multi-model combined infrared multi-spectral imaging method for insulating gas leakage in the second embodiment of the present invention;

[0042] Figure 8 is the infrared image of the insulating gas and the RBTF image of the gas filter during the test;

[0043] Figure 9 is the trend chart of the influence of temperature change on the change of the RBTF image during the test;

[0044] Figure 10 It updates the background of the RBTF image, extracts the foreground, classifies the gas, and generates the plume distribution map during the experiment. Specific implementation manners

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0046] The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings of the specification and specific embodiments:

[0047] Embodiment 1

[0048] An embodiment of the present invention provides a multi-component insulating gas multi-spectral imaging device, as Figures 1 to 3 shown, including: a visible light camera 11, a laser ranging module 12, an infrared imaging lens 13, a multi-spectral filter wheel 14, an infrared imaging detector 15, a signal processing board 16, a lithium battery 17, and a flip-up display screen 18; the visible light camera 11 is used to collect video images of the power scene, providing a visible light scene background map for infrared detection of insulating gas leakage on the one hand and expanding the multi-functional video monitoring function of the power scene on the other hand; the laser ranging module 12 is used to measure the distance between the device and the power equipment; the infrared imaging lens 13 is used for adjusting the infrared imaging field of view of the insulating gas; multiple broadband filters are installed on the multi-spectral filter wheel 14 for extracting the infrared characteristics of the multi-component insulating gas; the infrared imaging detector 15 is used to collect infrared images of the power scene of the multi-component insulating gas; the signal processing board 16 is used to process the data of the infrared imaging of the multi-component insulating gas; the lithium battery 17 is used to supply power to the device; the flip-up display screen 18 is used to display the imaging results.

[0049] Embodiment 2

[0050] As Figure 4 shown, an embodiment of the present invention provides a multi-model combined infrared multi-spectral imaging method for insulating gas leakage, specifically including the following steps:

[0051] Step 1: Use a multi-component insulating gas multi-spectral imaging device to collect multi-spectral image data of the leaked insulating gas.

[0052] Step 2: Classify moving objects using the YOLO-World model through prompts for custom application scenarios.

[0053] As Figure 5 shown, the method for classifying moving objects using the YOLO-World model is as follows:

[0054] 1) Input the multi-spectral image data of the leaked insulating gas into the YOLO detector to extract the multi-scale features of the image.

[0055] 2) Input the multi-scale features of the image into the multi-scale image feature pyramid network to fuse the feature maps of different scales to form a feature pyramid.

[0056] 3) Input the feature pyramid and the vocabulary into the vision-language path aggregation network. At each stage of the feature pyramid, inject the vocabulary text into the image features.

[0057] 4) Achieve a high degree of fusion of the image features and the text embedding through image perception embedding and region-text matching, and complete the classification of the moving object according to the similarity between the predicted bounding box and the text embedding.

[0058] Preferably, the YOLO-WORLD model is designed based on YOLOv8 and uses the deep convolutional neural network Darknet as the image encoder.

[0059] The YOLO-WORLD model enhances the original YOLO model through vision-language modeling and pre-training on large datasets for open-vocabulary detection. It has a reparameterizable vision-language path aggregation network (RepVL-PAN) and a region-text contrast loss, improving vision-language interaction. It is designed for real-time applications, can process the entire image at once, and improves detection speed and accuracy. The YOLO-WORLD model is good at detecting various moving objects, is seamlessly integrated with multi-spectral imaging technology, and minimizes false alarms by accurately distinguishing gas plumes from other interfering moving objects.

[0060] Step 3: Use the efficient SAM model to extract the ROI of the leaked insulating gas at the semantic level, so as to accurately extract and render the gas cloud ROI.

[0061] As Figure 6 shown, the method for using the efficient SAM model to extract the ROI of the leaked insulating gas at the semantic level, so as to accurately extract and render the gas cloud ROI is as follows:

[0062] 1) Input the RBTF image of the leaked insulating gas and the infrared image of the leaked insulating gas into the efficient SAM model respectively; among them, the RBTF image is obtained by dividing the filter image with a central wavelength of 8.17 microns by the filter image with a central wavelength of 7.49 microns;

[0063] 2) Pre-train an efficient SAM model using masked images. The pre-training learns to reconstruct features from the SAM image encoder for effective visual representation learning. Then, construct the model using the lightweight image encoder and masked decoder pre-trained with masked images.

[0064] 3) Obtain the ROIs of the gas in the RBTF image and the infrared image respectively through the trained efficient SAM model.

[0065] The efficient SAM (Segment Anything Model) model performs zero-shot segmentation, accurately segmenting objects in images without prior training on a specific dataset, improving speed and accuracy in various applications. The efficient SAM model enhances the traditional SAM model by improving segmentation speed, accuracy, and computational efficiency. It has a streamlined architecture with lightweight convolutional layers, advanced feature extraction techniques, and efficient data processing. The efficient SAM model is ideal for automatic gas detection, capable of precisely handling various object scales and types, seamlessly integrating with the YOLO-World model to provide high performance with low computational requirements, making it suitable for edge devices and industrial applications.

[0066] Step 4: Train an SVM model to identify the leaked insulating gas.

[0067] As Figure 7 shown, the method for training the SVM model to identify the leaked insulating gas is specifically as follows:

[0068] 1) Feature extraction: Extract peaks and valleys from the multi-spectral data of the insulating gas as features of the gas characteristics.

[0069] 2) SVM model training: Use the extracted features to train the SVM model; during the training process, establish a training set and a validation set to optimize the performance of the SVM model.

[0070] 3) Model evaluation and optimization: Evaluate the performance of the SVM model through cross-validation, accuracy, and recall metrics.

[0071] 4) Gas identification: Use the trained SVM model to classify and identify the spectral data of each pixel point within the determined image ROI, and determine the type of gas for each pixel according to the output result of the SVM model.

[0072] The SVM model is a binary classification model. Its core is to find a separating hyperplane such that the two types of data points are on both sides of this hyperplane, and the distance (i.e., margin) from this hyperplane to the two types of data points is maximized. For multi-classification problems, it can be solved by constructing multiple binary SVMs or using the multi-class extension algorithm of SVM. The key to the SVM algorithm lies in the selection of the kernel function, which can map the input data to a high-dimensional space to solve non-linear classification problems.

[0073] Experimental verification

[0074] The insulating gas is released at a distance of 20 meters from the measurement point. Sulfur hexafluoride is released from a cylindrical container, while perfluoroisobutyronitrile gas dissipates from a box-shaped container. This setup aims to simulate real-world conditions where environmental factors such as temperature fluctuations affect the accuracy of gas detection. The rising road temperature poses challenges to the detection system, making it difficult to distinguish gas plumes from background temperature changes, thus testing its robustness and reliability under dynamic environmental conditions.

[0075] The infrared image of the insulating gas and the relative bright temperature of the gas filter (RBTF) image are as Figure 8 shown. The RBTF image is obtained by dividing the filter image 2 (center wavelength 8.17 microns) by the filter image 1 (center wavelength 7.49 microns). In the right RBTF image, although there is significant noise in the background, the area of the insulating gas can be observed.

[0076] During the measurement, due to the combined action of sunlight and wind, the road surface temperature decreased by approximately 0.8 °C within 5 minutes. Figure 9 Shows the changes in the temperature and RBTF image at the reading intersection. It clearly shows that although the background temperature decreased, the RBTF image remained stable, indicating that the RBTF image effectively compensates for environmental temperature fluctuations.

[0077] As Figure 10 shown in the left figure below, update the background of the RBTF image and extract the foreground. Thereafter, use the SVM model to classify the multi-spectral image of the gas, as Figure 10 shown in the right figure below, where the red in the figure represents the distribution of perfluoroisobutyronitrile plumes, and the orange represents the distribution of sulfur hexafluoride plumes.

[0078] Example 3

[0079] An electronic device includes a memory and a processor. The memory is used to store a program that supports the processor to execute the multi-model combined insulating gas leakage infrared multi-spectral imaging method in Example 2, and the processor is configured to execute the program stored in the memory.

[0080] Example 4

[0081] A storage medium stores a computer program, and when the computer program is run by a processor, it executes the steps of the multi-model combined infrared multi-spectral imaging method for insulating gas leakage in the second embodiment.

[0082] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-model combined insulating gas leakage infrared multi-spectral imaging method, characterized in that: include: S1 uses a multi-component insulating gas multi-spectral imaging device to collect leaking insulating gas multi-spectral image data; S2 uses the YOLO-World model to classify moving objects based on the prompts of the custom application scenario; S3 uses an efficient SAM model to extract the ROI of leaking insulating gas at the semantic level, thereby accurately extracting and rendering the gas cloud ROI; S4 trains the SVM model to identify leaking insulating gas.

2. The multi-model combined insulating gas leakage infrared multi-spectral imaging method according to claim 1 is characterized in that: The multi-component insulating gas multi-spectral imaging device comprises: a visible light camera (11), a laser ranging module (12), an infrared imaging lens (13), a multi-spectral filter wheel (14), an infrared imaging detector (15), a signal processing board (16), a lithium battery (17), and a flip-able display screen (18); the visible light camera (11) is used to collect video images of power scenes, on the one hand to provide a visible light scene base map for insulating gas leakage infrared detection, and on the other hand to expand the multi-functional video monitoring function of the power scene; the laser ranging module (12) is used to measure the distance between the device and the power equipment; the infrared imaging lens (13) is used to adjust the insulating gas infrared imaging field of view; a plurality of broadband filters are installed on the multi-spectral filter wheel (14) to extract the infrared characteristics of the multi-component insulating gas; the infrared imaging detector (15) is used to collect infrared images of the power scene of the multi-component insulating gas; the signal processing board (16) is used to process the data of the infrared imaging of the multi-component insulating gas; the lithium battery (17) is used to power the device; and the flip-able display screen (18) is used to display the imaging results.

3. The multi-model combined insulating gas leakage infrared multi-spectral imaging method according to claim 1 is characterized in that: The method of using the YOLO-World model to classify moving objects is as follows: Input the multispectral image data of the leaked insulating gas into the YOLO detector to extract the multi-scale features of the image; The multi-scale features of the image are input into the multi-scale image feature pyramid network, and the feature maps of different scales are fused to form a feature pyramid; The feature pyramid and vocabulary are fed into the visual-linguistic path aggregation network together. At each stage of the feature pyramid, the vocabulary text is injected into the image features. Through image-aware embedding and region-text matching, a high degree of fusion of image features and text embedding is achieved, and the classification of moving objects is completed based on the similarity between the predicted bounding box and the text embedding.

4. The multi-model combined insulating gas leakage infrared multi-spectral imaging method according to claim 3 is characterized in that: The YOLO-WORLD model is designed based on YOLOv8 and uses the deep convolutional neural network Darknet as the image encoder.

5. The multi-model combined insulating gas leakage infrared multi-spectral imaging method according to claim 3 is characterized in that: The method of using an efficient SAM model to extract the ROI of leaking insulating gas at the semantic level, thereby accurately extracting and rendering the gas cloud ROI is as follows: The leaked insulating gas RBTF image and the leaked insulating gas infrared image are respectively input into the efficient SAM model; The efficient SAM model is pre-trained using mask images. The pre-training learns to reconstruct features from the SAM image encoder and conducts effective visual representation learning. Then, the model is constructed using the lightweight image encoder and mask decoder pre-trained with mask images. Through the trained efficient SAM model, the ROI of gas in RBTF image and infrared image is obtained respectively.

6. The multi-model combined insulating gas leakage infrared multi-spectral imaging method according to claim 5 is characterized in that: The method of training the SVM model to identify leaking insulating gas is as follows: Extract peaks and valleys from insulating gas multispectral data as features of gas properties; The extracted features are used to train the SVM model. During the training process, a training set and a validation set are established to optimize the performance of the SVM model. The performance of the SVM model was evaluated through cross-validation, accuracy, and recall metrics; The trained SVM model is used to classify and identify the spectral data of each pixel in the determined image ROI, and the type of gas in each pixel is determined based on the output results of the SVM model.

7. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the multi-model combined insulating gas leakage infrared multi-spectral imaging method as described in any one of claims 1 to 6, and the processor is configured to execute the program stored in the memory.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multi-model combined insulating gas leakage infrared multi-spectral imaging method described in any one of claims 1 to 6 are executed.

Images