Leakage and leakage identification method, electronic equipment and computer readable storage medium

Abstract

The invention discloses a leakage identification method, electronic equipment and a computer readable storage medium, and the method comprises the steps: inputting an obtained collection image into a segmentation network model, carrying out the leakage identification of target equipment, obtaining a first identification result, and enabling the first identification result to comprise a detection result of whether the target equipment has leakage or not, the leakage and leakage position information is obtained when the detection result shows that leakage and leakage exist; in response to the first identification result representing that the target device has leakage, inputting the collected image into a large language model to obtain a second identification result output by the large language model, the second identification result comprising a detection result of whether leakage exists in the collected image; and performing verification processing on the first identification result according to the second identification result to obtain a target identification result. Therefore, the abnormal identification false alarm rate of the equipment can be effectively reduced, and the identification precision is improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method for identifying spills and leaks, an electronic device, and a computer-readable storage medium. Background Technology

[0002] Leaks and spills refer to the phenomenon of liquids or gases leaking from pipes, equipment, or other parts of the system in scenarios such as industrial production, urban water supply, and pipeline transportation. Timely detection and location of leaks and spills are of great significance for ensuring production safety, reducing resource waste, and protecting the environment.

[0003] Current methods for detecting leaks and spills mainly rely on manual inspections, which are inefficient and lack timeliness. Summary of the Invention

[0004] The main technical problem addressed by this application is to provide a method, electronic device, and computer-readable storage medium for identifying leaks, spills, and drips, which can improve the accuracy of leak detection.

[0005] To address the aforementioned technical problems, this application provides a method for identifying leaks and spills. The method includes: inputting an acquired image into a segmentation network model for leak and spill identification processing of the target device, obtaining a first identification result. The first identification result includes a detection result indicating whether the target device has leaks and spills, and location information of the leaks and spills when the detection result indicates their presence. In response to the first identification result indicating that the target device has leaks and spills, the acquired image is input into a large language model to obtain a second identification result output by the large language model. The second identification result includes a detection result indicating whether leaks and spills exist in the acquired image. The first identification result is then verified based on the second identification result to obtain a target identification result.

[0006] To solve the above-mentioned technical problems, another technical solution adopted in this application is: to provide an electronic device, including a memory and a processor, wherein the memory stores program instructions, and the processor retrieves the program instructions from the memory to execute the above-mentioned leakage identification method.

[0007] To solve the above-mentioned technical problems, another technical solution adopted in this application is to provide a computer-readable storage medium including program data, which, when executed by a processor, is used to implement the above-mentioned method for identifying leaks and spills.

[0008] The method for identifying leaks and spills in this application inputs the acquired image into a segmentation network model for leak detection processing of the target device, obtaining a first identification result. The first identification result includes the detection result of whether the target device has leaks and spills, and the location information of the leaks and spills when the detection result indicates the presence of leaks and spills. In response to the first identification result indicating the presence of leaks and spills in the target device, the acquired image is input into a large language model to obtain a second identification result output by the large language model. The second identification result includes the detection result of whether leaks and spills exist in the acquired image. The first identification result is verified based on the second identification result to obtain the target identification result. In the above scheme, the segmentation network model can quickly identify the location information of leaks and spills, while the large language model, when the segmentation network model confirms the presence of leaks and spills, verifies whether the leaks and spills actually exist through its strong generalization ability. Combining the location recognition ability of the segmentation network model and the high accuracy of the large language model, the false alarm rate of leak and spill identification can be effectively reduced. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 This is a flowchart illustrating an exemplary embodiment of the leakage detection method shown in this application; Figure 2 yes Figure 1 A flowchart illustrating an exemplary embodiment of step S120 in the method for identifying leaks and spills is shown. Figure 3 This is a flowchart illustrating another exemplary embodiment of the leakage detection method shown in this application; Figure 4 This is a flowchart illustrating another exemplary embodiment of the leakage detection method shown in this application; Figure 5 This is a schematic diagram of an exemplary embodiment of the leak detection device shown in this application; Figure 6 This is a schematic diagram of the structure of an embodiment of the electronic device provided in this application; Figure 7 This is a schematic diagram of an embodiment of the computer-readable storage medium provided in this application. Detailed Implementation

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It is understood that the specific embodiments described herein are only for explaining this application and not for limiting it. Furthermore, it should be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all structures. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0011] First, it should be noted that leakage refers to the phenomenon of gas leaks, water spills, drips, or leaks that occur during the storage and transportation of liquids or gases due to poor management and improper operation. Currently, the main method for timely detection of leakage in equipment is manual inspection. However, manual inspection is inefficient, easily affected by external interference, and difficult to standardize.

[0012] Based on this, embodiments of this application propose a method, electronic device, and computer-readable storage medium for identifying leaks and spills. A large language model is used for secondary verification to ensure the accuracy of leak and spill identification. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 This is a flowchart illustrating an exemplary embodiment of the leakage identification method shown in this application.

[0013] The execution entity of the leakage detection method can be a terminal device, a server, or other processing device. The terminal device can be a user equipment (UE), computer, mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, in-vehicle device, wearable device, etc. The execution entity of the leakage detection method can also be a leakage detection device. In some possible implementations, the leakage detection method can be implemented by a processor calling computer-readable instructions stored in memory.

[0014] Specifically, the leakage detection method in this embodiment includes the following steps: S110: Input the acquired image into the segmentation network model to perform the target device leakage detection processing to obtain the first recognition result.

[0015] The acquired image includes the target device to be detected. Exemplarily, the acquired image can be an image selected from an image set, an image acquired by an image acquisition device, or a video frame extracted from a video stream acquired by a video acquisition device. It can be any image in the image set, or the image with the best image quality in the image set; this application embodiment does not limit this. In some embodiments, after acquiring the acquired image, the leak detection device can directly perform recognition processing on the acquired image to obtain a first recognition result. In other embodiments, after acquiring the acquired image, the leak detection device can also preprocess the acquired image to improve image quality.

[0016] The segmentation network model is used to detect whether there are leaks or spills in the target device in the acquired image, and when leaks or spills are found, it segments the leaking or spilling area from the acquired image and provides the location information of the leaks or spills. For example, the segmentation network model can be BiSeNetV2 (Bilateral Segmentation Network V2), ConvNext (Convolutional Next), etc.

[0017] The target device can be any device to be detected in the current working environment. For example, the leak detection device can use a device that is currently in operation in the current working environment as the target device; it can also use a device that is in operation but not in operation in the current working environment as the target device. It should be noted that a single captured image may include one or more target devices, and a single target device may exist in multiple captured images.

[0018] The first identification result includes any abnormalities in the target device within the acquired image. Specifically, the first identification result detected by the segmentation network model can include the detection result of whether the target device is leaking or spilling, and the location information of the leak when the detection result indicates that leaking or spilling exists. The detection result of whether the target device is leaking or spilling includes whether the target device is leaking or spilling, and when leaking or spilling exists, the first identification result will also include location information about the leaking or spilling area, such as the coordinates of the leaking or spilling area. Simultaneously, the segmentation network model can also segment the pixel region with leaking or spilling in the acquired image to obtain sub-images, which are then displayed to the client. In other embodiments, the leaking or spilling identification device can use traditional image processing methods or deep learning methods to identify the target device in the acquired image to obtain the first identification result. The acquired image can be a visible light image, and the leaking or spilling identification device can determine the first identification result based on information such as color and texture of the acquired image. The acquired image can also be a thermal imaging image, and the leak detection device determines the first identification result based on the temperature gradient in the acquired image. In other embodiments, the leak detection device can also detect the target device through a sensor to obtain the first identification result.

[0019] S120: In response to the first recognition result indicating that the target device has leaks, the collected image is input into the large language model to obtain the second recognition result output by the large language model.

[0020] A large language model is a deep learning model capable of understanding and generating natural language. For example, a large language model can be trained on massive amounts of text through self-supervised learning, possessing powerful language understanding and generation capabilities. Examples of large language models include ChatGLM (Chat Generative Language Model), LLaMA (Large Language Model Meta AI), Baichuan, Qwen, ChatGPT (Chat Generative Pre-trained Transformer), BERT (Bidirectional Encoder Representations from Transformers), and Wenxin Yiyan, among others. In some embodiments, the "running away with information" (or "leaking") detection device can directly utilize the large language model to process the acquired images; in other embodiments, the device can fine-tune the large language model before using the fine-tuned model to process the acquired images.

[0021] The second recognition result is the recognition result output by the large language model. For example, the acquired image is input into the large language model, which processes the image to determine whether the target device in the image is experiencing leakage, thus obtaining the second recognition result. The second recognition result includes the detection result of whether leakage exists in the acquired image. It should be noted that while the large language model has the ability to detect leakage, it cannot accurately provide the location information of the leakage. Therefore, in this embodiment, when the segmentation network model determines that the target device has leakage, the strong generalization ability of the large language model is used to further confirm whether leakage exists. If leakage exists, it indicates that the first recognition result is accurate; if leakage does not exist, it indicates that the first recognition result is inaccurate.

[0022] The leak detection device can input the acquired image into a large language model for further recognition processing when the first recognition result indicates that the target device has leaks. Conversely, if the first recognition result indicates that the target device does not have leaks, the large language model is not used for recognition processing. In other embodiments, the leak detection device can also input the acquired image into the large language model for further recognition processing when the first recognition result indicates that the target device does not have leaks, and vice versa. In other embodiments, the leak detection device can also acquire the first recognition result of the acquired image and the second recognition result output by the large language model, and perform weighted processing on the first and second recognition results to obtain the target recognition result.

[0023] S130: Verify the first identification result based on the second identification result to obtain the target identification result.

[0024] The target identification result is a result verified by the second identification result. For example, after acquiring the second and first identification results, the leak detection device determines the consistency between them; based on this consistency, it determines the target identification result. In other embodiments, the first and second identification results can be input again into a large language model, which then determines whether the first identification result has a false alarm risk based on the second identification result. If so, the first identification result is corrected or marked as requiring further confirmation.

[0025] The target recognition results may include, but are not limited to, information such as the location of the leaking area, mask image, and verification status. Furthermore, the leak detection device can display the target recognition results to the user in a visual manner.

[0026] As can be seen, the leakage detection method in this embodiment of the application inputs the acquired image into a segmentation network model for leakage detection processing of the target device, obtaining a first detection result. The first detection result includes the detection result of whether leakage exists in the target device, and the location information of leakage when the detection result indicates the presence of leakage. In response to the first detection result indicating the presence of leakage in the target device, the acquired image is input into a large language model to obtain a second detection result output by the large language model. The second detection result includes the detection result of whether leakage exists in the acquired image. The first detection result is verified based on the second detection result to obtain the target detection result. In the above scheme, the segmentation network model can quickly identify the location information of leakage, while the large language model, when the segmentation network model confirms the presence of leakage, verifies whether leakage actually exists through its strong generalization ability. Combining the location recognition ability of the segmentation network model and the high accuracy of the large language model, the false alarm rate of leakage detection can be effectively reduced. Based on the above embodiments, the embodiments of this application adopt... Figure 2 The flowchart details how to use a large language model to generate the second recognition result of the acquired image. Please refer to [link / reference]. Figure 2 , Figure 2 yes Figure 1 The illustrated flowchart shows an exemplary embodiment of step S120 in the leak detection method. Specifically, step S120 includes the following steps: S210: Input the second recognition result and the first recognition result into the large language model for semantic analysis to obtain the recognition result output by the large language model.

[0027] After acquiring the first and second recognition results, the large language model can perform semantic analysis on them to determine the differences between them and whether the first recognition result carries a risk of misjudgment. For example, when inputting the second and first recognition results, text prompts can be added simultaneously to remind the large language model to determine whether the first recognition result is a misjudgment relative to the second recognition result. Furthermore, when processing the acquired images, text prompts can also be added to instruct the large language model to perform leak detection on the target device and obtain the second recognition result.

[0028] The first and second identification results are descriptive information for the acquired images, which may include, but are not limited to, whether the target device has any anomalies, the type of anomaly, and the area of ​​the anomaly. Since the first and second identification results are obtained using different methods, they may have different expressions. Therefore, this embodiment uses a large language model to judge the differences between the first and second identification results, determining whether the first identification result carries a risk of misjudgment relative to the second identification result, thereby obtaining the identification result of leakage from the target device. The large language model, by learning from a large amount of text data, can understand the semantic information of leakage features and has powerful natural language understanding capabilities.

[0029] In other embodiments, when the first identification result indicates that the target device has leaks, the leakage detection device inputs the acquired image and the first identification result into a large language model to obtain the identification result output by the large language model. It should be noted that this method requires prompting the large language model to first perform leakage detection on the acquired image to obtain a second identification result, and then use the second identification result to perform misjudgment analysis on the first identification result to obtain the identification result.

[0030] S220: The recognition result output by the large language model is determined as the target recognition result.

[0031] After obtaining the recognition result output by the large language model, the leakage detection device can directly determine the recognition result of the large language model as the target recognition result of the target device. In other embodiments, when the large language model determines that there is a risk of misjudgment in the first recognition result, the leakage detection device can obtain the target recognition result by manually confirming whether there is leakage in the target device.

[0032] In other embodiments, the leakage detection device can perform semantic matching processing on the first and second identification results after obtaining them to obtain a matching result; and determine the target identification result based on the matching result. A non-large language model is thus used for verification processing. When the matching result indicates that the first and second identification results do not match successfully, the first identification result is corrected to obtain the target identification result; when the matching result indicates that the first and second identification results match successfully, the first identification result is determined as the target identification result.

[0033] Specifically, in response to the second identification result indicating that the target device has leaks, the first identification result is determined as the target identification result; in response to the second identification result indicating that the target device does not have leaks, the first identification result is corrected using the second identification result to obtain the target identification result. This improves the accuracy of the target identification result.

[0034] In this embodiment, when the first identification result indicates that the target device has leaks, the acquired image is then input into the large language model to obtain the second identification result. Therefore, when the second identification result indicates that the target device has leaks, it means that the first and second identification results are consistent, and the first identification result is considered to be without misjudgment. In this case, the first identification result can be directly used to determine the target identification result. When the second identification result indicates that the target device does not have leaks, it means that the first and second identification results are inconsistent, and the first identification result is considered to be likely to be misjudged. In this case, the first identification result needs to be corrected using the second identification result to obtain the target identification result.

[0035] In other embodiments, when the second identification result indicates that the target device has leaks, the first identification result can be determined as the target identification result; when the second identification result indicates that the target device does not have leaks, the target identification result can be obtained by manually verifying whether the target device has leaks.

[0036] In some embodiments, the information in the first identification result is richer and more complete than that in the second identification result. The first identification result includes not only the conclusion of whether leakage exists, but also information such as the location of the leakage area and the mask image. For example, a domain-specific identification method can be used to identify the acquired image to obtain the first identification result. Specifically, the leakage identification device preprocesses the acquired image to obtain a preprocessed acquired image; the preprocessed acquired image is then input into a segmentation network model for identification processing to obtain the first identification result.

[0037] To improve image quality, the acquired images can be preprocessed, including but not limited to operations such as grayscale conversion and normalization. The preprocessed images are then subjected to recognition processing to obtain a first recognition result. For example, the leak detection device can improve the segmentation network model to obtain a self-attention segmentation network model; the acquired images are then input into the self-attention segmentation network model to perform leak detection on the target device, obtaining the first recognition result. In other embodiments, the leak detection device can also input the preprocessed acquired images into the self-attention segmentation network model to obtain the recognition result output by the self-attention segmentation network model; the recognition result output by the self-attention segmentation network model is then determined as the first recognition result. Therefore, compared to traditional segmentation network models, introducing an attention mechanism can better focus on the key features of the leak area, improving recognition accuracy.

[0038] A self-attention segmentation network model refers to a segmentation network model that incorporates a self-attention mechanism. This model can identify and segment overflowing / leaking areas in a captured image, outputting a segmented mask image and the locations of the overflowing / leaking areas. The self-attention segmentation network model can be an improved Swin Trasnformer network, ConvNext network, or BiSeNetV2, etc. In other embodiments, the segmentation network model can also be used to perform recognition processing on the preprocessed captured image to obtain a first recognition result.

[0039] by Figure 3 The flowchart shown further illustrates the leakage detection method of this application embodiment, as detailed below: The acquired images are input into a self-attention segmentation network model, which performs inference to obtain a first recognition result. The first recognition result includes the detection result of whether the target device has any leakage, and the location information of the leakage when the detection result indicates that leakage exists. If the first recognition result indicates that the target device does not have any leakage, the first recognition result is output. If the first recognition result indicates that the target device has any leakage, the acquired images are input into a fine-tuned large language model, which performs recognition processing on the acquired images to obtain a second recognition result. The first recognition result is then verified based on the second recognition result. If the large language model determines that there is no leakage, the second recognition result is output. If the large language model determines that there is leakage, a leakage alarm is triggered, and the target recognition result is determined based on the first recognition result and then output.

[0040] Furthermore, the training process of the segmentation network model further includes: obtaining a sample dataset, which includes at least one sample data based on the dripping and overflowing process; inputting the sample dataset into the segmentation network model to be trained to obtain the segmentation network model.

[0041] The sample dataset includes at least one sample data point, which consists of collected images related to leakage phenomena. Exemplarily, the sample data can be collected using image acquisition devices or video acquisition devices. These devices can be installed in the device's working area, using images collected over a specific time period as sample data to form the sample dataset. The leakage detection device can also obtain sample datasets related to leakage from online open-source datasets. In some embodiments, after obtaining the sample data, data cleaning and labeling processes can be performed. Data cleaning includes, but is not limited to, noise removal, error correction, and missing value imputation. Labeling processes involve adding labels to key features in the images to enable the model to learn and recognize these key features.

[0042] After obtaining the sample dataset, the sample dataset is sequentially input into the segmentation network model to be trained for model training until a trained segmentation network model is obtained. Specifically, the sample data is input into the segmentation network model to be trained to obtain the recognition result output by the segmentation network model; the loss value between the recognition result output by the segmentation network model and the real result is calculated, and the segmentation network model to be trained is trained with the goal of reducing the loss value, until the loss value meets the preset requirements, thus obtaining the segmentation network model. In some other embodiments, an improved self-attention segmentation network model can also be trained to obtain a trained self-attention segmentation network model.

[0043] Furthermore, to improve the accuracy of identifying leaks and spills, the large language model can be fine-tuned using sample data to enhance its algorithmic recognition capabilities in leak and spill scenarios. Specifically, a sample dataset is obtained, which includes at least one sample data point based on leaks and spills. This sample dataset is then input into the large language model to be trained, resulting in the large language model.

[0044] The sample dataset used for training the large language model can be the same as the sample dataset used for training the segmentation network model. Alternatively, in some embodiments, high-quality image samples can be obtained from the sample dataset of the self-segmentation network model as the sample dataset for the large language model. After obtaining the sample dataset for the large language model, the sample dataset is sequentially input into the large language model to be trained for model training until a trained large language model is obtained. Specifically, the sample data is input into the large language model to be trained to obtain the recognition result output by the large language model; the loss value between the recognition result output by the large language model and the true result is calculated, and the large language model to be trained is trained with the goal of reducing the loss value until the loss value meets the preset requirements, thus obtaining the large language model.

[0045] After model training is complete, the leak detection device deploys the two models onto the algorithm camera or edge device for practical use. In some embodiments, the final output of the two models can be directly used as the target recognition result of the device. In other embodiments, if a recognition error is found in the model during actual use, the corresponding input image can be added to the sample dataset, and the model can be retrained to improve the model's generalization ability.

[0046] Specifically, when the leakage detection device detects a discrepancy between the target recognition result and the preset true result, it acquires the labeled data of the collected image and adds the collected image and corresponding labeled data to the sample dataset, obtaining a new sample dataset. The new sample dataset is then used to train the segmentation network model and / or the large language model until the preset requirements are met. In this way, difficult examples are continuously acquired during practical applications, and the model is continuously optimized using these difficult examples to improve its recognition ability.

[0047] The preset true result can be the result of manual identification. After obtaining the target identification result, it can be manually verified again to check whether there is any leakage or spillage in the target device in the acquired image. If the target identification result is inconsistent with the manually determined preset true result, the current acquired image can be used as sample data. The acquired image is then cleaned and labeled, and the acquired image and the corresponding labeled data are added to the sample dataset to obtain a new sample dataset. The segmentation network model and / or large language model are then trained using the new sample dataset to optimize model performance. If the target identification result is consistent with the preset true result, the target identification result is determined to be valid.

[0048] For details, please refer to [link / reference]. Figure 4 , Figure 4 This is a flowchart illustrating another exemplary embodiment of the leakage detection method shown in this application. First, images of leakage are collected, which can be obtained through a data acquisition device or an online open-source dataset. The collected images are then cleaned and labeled to obtain a sample dataset. The self-attention segmentation network model to be trained is then trained sequentially using each sample data from the sample dataset. Next, the large language model to be trained is fine-tuned sequentially using each sample data from the sample dataset to improve its understanding and processing of leakage-related collected images. The trained self-attention segmentation network model and the large language model are then deployed to an algorithm camera or edge device. The two models perform on-site reasoning on the acquired images to obtain target recognition results, including whether the target device is... There is a leakage phenomenon. Specifically, the self-attention segmentation network model performs recognition processing on the acquired images to obtain the first recognition result. If the first recognition result indicates that the target device has leakage, the large language model is used to perform secondary verification on the acquired images to obtain the final target recognition result, thereby reducing the false detection rate caused by the deep learning model. Then, the target device is manually verified to see if there is leakage. If the preset true result of manual verification is consistent with the target recognition result, the process ends and the target recognition result is confirmed to be valid. If the preset true result of manual verification is inconsistent with the target recognition result, the acquired images are added to the sample dataset and returned to the data cleaning and annotation process.

[0049] Please see Figure 5 , Figure 5 This is a schematic diagram of an exemplary embodiment of the leak detection device shown in this application. The leak detection device 500 includes a first identification module 510, a second identification module 520, and a verification module 530. The first identification module 510 is used to identify the target device in the acquired image to obtain a first identification result. The second identification module 520 is used to input the acquired image into a large language model in response to the first identification result indicating that the target device has leaks, to obtain a second identification result output by the large language model. The verification module 530 is used to verify the first identification result according to the second identification result to obtain a target identification result.

[0050] In the above scheme, the leakage detection device inputs the acquired image into a segmentation network model for leakage detection processing of the target device, obtaining a first detection result. The first detection result includes the detection result of whether leakage exists in the target device, and the location information of the leakage when the detection result indicates the presence of leakage. In response to the first detection result indicating the presence of leakage in the target device, the acquired image is input into a large language model, obtaining a second detection result output by the large language model. The second detection result includes the detection result of whether leakage exists in the acquired image. The first detection result is verified based on the second detection result to obtain the target detection result. In this scheme, the segmentation network model can quickly identify the location information of leakage, while the large language model, when the segmentation network model confirms the presence of leakage, verifies whether the leakage phenomenon actually exists through its strong generalization ability. Combining the location recognition ability of the segmentation network model and the high accuracy of the large language model, the false alarm rate of leakage detection can be effectively reduced. The functions of each module can be found in the embodiment of the leakage detection method, and will not be repeated here.

[0051] To implement the leakage detection method of the above embodiments, this application proposes another electronic device, please refer to [link / reference needed]. Figure 6 , Figure 6 This is a schematic diagram of the structure of an embodiment of the electronic device provided in this application.

[0052] Electronic device 600 includes memory 610 and processor 620, wherein memory 610 and processor 620 are coupled together.

[0053] The memory 610 is used to store program data, and the processor 620 is used to execute the program data to implement the leakage identification method of the above embodiment.

[0054] In this embodiment, processor 620 can also be referred to as a CPU (Central Processing Unit). Processor 620 may be an integrated circuit chip with signal processing capabilities. Processor 620 can also be a general-purpose processor, 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, or discrete hardware components. The general-purpose processor can be a microprocessor, or processor 620 can be any conventional processor.

[0055] This application also provides a computer-readable storage medium, such as Figure 7 As shown, the computer-readable storage medium 700 is used to store program data 710, which, when executed by the processor, is used to implement the leakage identification method as described in the method embodiment of this application.

[0056] The methods involved in the leakage detection method embodiments of this application, when implemented as software functional units and sold or used as independent products, can be stored in a device, such as a computer-readable storage medium. Based on this understanding, the technical solution of this application, essentially, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0057] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method of leak detection, characterized in that, The running, leaking and dripping identification method comprises: inputting the acquired collection image into a segmentation network model for running, leaking and dripping identification processing of the target device to obtain a first identification result, the first identification result comprising a detection result of whether the target device has running, leaking and dripping, and position information of the running, leaking and dripping when the detection result represents that there is running, leaking and dripping; in response to the first identification result representing that the target device has running, leaking and dripping, inputting the collection image into a large language model to obtain a second identification result output by the large language model, the second identification result comprising a detection result of whether there is running, leaking and dripping in the collection image; performing verification processing on the first identification result according to the second identification result to obtain a target identification result.

2. The method of leak detection according to claim 1, wherein, The step of performing verification processing on the first identification result according to the second identification result to obtain a target identification result comprises: inputting the second identification result and the first identification result into the large language model for semantic analysis to obtain an identification result output by the large language model; determining the identification result output by the large language model as the target identification result.

3. The method of leak detection according to claim 1, wherein, The step of performing verification processing on the first identification result according to the second identification result to obtain a target identification result comprises: in response to the second identification result representing that the target device has running, leaking and dripping, determining the first identification result as the target identification result; in response to the second identification result representing that the target device does not have running, leaking and dripping, performing correction processing on the first identification result by using the second identification result to obtain the target identification result.

4. The method of leak detection according to claim 1, wherein, The step of inputting the acquired collection image into a segmentation network model for running, leaking and dripping identification processing of the target device to obtain a first identification result comprises: performing preprocessing on the collection image to obtain a preprocessed collection image; inputting the preprocessed collection image into the segmentation network model for identification processing to obtain the first identification result.

5. The method of claim 1, wherein, Before the step of inputting the acquired collection image into a segmentation network model for running, leaking and dripping identification processing of the target device to obtain a first identification result, the method further comprises: performing improvement processing on the segmentation network model to obtain a self-attention segmentation network model; The step of inputting the acquired collection image into a segmentation network model for running, leaking and dripping identification of the target device to obtain a first identification result comprises: inputting the acquired collection image into the self-attention segmentation network model for running, leaking and dripping identification of the target device to obtain a first identification result.

6. The method of leak detection according to claim 1, wherein, Before the step of inputting the acquired collection image into a segmentation network model for running, leaking and dripping identification processing of the target device to obtain a first identification result, the method further comprises: acquiring a sample data set, the sample data set comprising at least one sample data based on running, leaking and dripping; inputting the sample data set into a segmentation network model to be trained for model training to obtain the segmentation network model.

7. The method of leak detection according to claim 6, wherein, After the step of verifying the first recognition result according to the second recognition result to obtain a target recognition result, the method further comprises: In response to the target recognition result being different from a preset true result, obtaining labeled data of the collected image, and adding the collected image and corresponding labeled data to the sample data set to obtain a new sample data set; Performing model training on the segmentation network model and / or the large language model using the new sample data set until a preset requirement is met.

8. The method of leak detection according to claim 1, wherein, The method further comprises: Obtaining a sample data set, wherein the sample data set comprises at least one sample data based on a run-out, a leak and a drop; Inputting the sample data set into a large language model to be trained to perform model training, and obtaining the large language model.

9. An electronic device, comprising: Comprise: A memory and a processor, wherein the memory stores program instructions, and the processor retrieves the program instructions from the memory to execute the method according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, Comprise: Program data is stored, and the program data is executed by a processor to implement the method according to any one of claims 1-8.

Images