Gas security check management system and method based on image recognition

Abstract

The invention relates to the technical field of fuel gas security inspection, in particular to a fuel gas security inspection management system and method based on image recognition. The server comprises an image acquisition module which is used for acquiring corresponding to-be-identified gas security check image data from a security check business system; the object recognition module is used for recognizing a corresponding object in the to-be-recognized fuel gas security check image data based on a preset object recognition model according to the collected to-be-recognized fuel gas security check image data, and generating corresponding object data; the hidden danger identification module is used for identifying the hidden danger corresponding to the object data in the to-be-identified gas security check image data based on a preset object hidden danger identification model according to the identified object data and the corresponding to-be-identified gas security check image data, and outputting the hidden danger information corresponding to the object data; and the early warning module is used for sending early warning information to a user according to the risk hidden danger information corresponding to the output object data.

Description

Technical Field

[0001] The present invention relates to the technical field of gas safety inspection, and in particular to a gas safety inspection management system and method based on image recognition. Background Art

[0002] Gas safety inspection plays a vital role in ensuring public safety, preventing accidents, protecting the environment and maintaining social stability.

[0003] Currently, gas safety inspection mainly relies on two methods: manual inspection and detection using existing image hazard recognition models;

[0004] Among them, manual inspection requires security personnel to inspect gas equipment and pipelines on site and evaluate safety hazards based on experience and professional knowledge. Although this method is intuitive and flexible, it has disadvantages such as low efficiency, strong subjectivity, and susceptibility to environmental factors.

[0005] The direct image hidden danger identification model uses a single deep learning model to directly identify potential safety hazards from the collected images. This method attempts to improve efficiency through automation, but faces many challenges in practical applications:

[0006] Insufficient accuracy: Since the two stages of object detection and hidden danger identification are not distinguished, the model may find it difficult to accurately capture the complex object features and their state changes, resulting in false positives or negative negatives.

[0007] Based on this, there is an urgent need for a gas safety inspection management system and method based on image recognition, which can realize separate detection of object detection and hidden danger identification, reduce recognition interference, improve detection accuracy, and reduce the possibility of false alarms or missed alarms. Summary of the invention

[0008] One of the purposes of the present invention is to provide a gas safety inspection management system and method based on image recognition, which can realize separate detection of object detection and hidden danger identification, reduce identification interference, improve detection accuracy, and reduce the possibility of false alarm or missed alarm.

[0009] In order to achieve the above-mentioned purpose, a gas safety inspection management system based on image recognition is provided, including a service end;

[0010] The server includes:

[0011] An image acquisition module is used to acquire corresponding gas safety inspection image data to be identified from the safety inspection business system;

[0012] The object recognition module is used to recognize the corresponding object in the gas safety inspection image data to be recognized based on the collected gas safety inspection image data to be recognized and based on the preset object recognition model, and generate the corresponding object data;

[0013] A hidden danger identification module is used to identify the risk hidden dangers corresponding to the object data in the gas safety inspection image data to be identified based on the identified object data and the corresponding gas safety inspection image data to be identified, based on a preset object hidden danger identification model, and output the risk hidden danger information of the corresponding object data;

[0014] The early warning module is used to send early warning information to users based on the risk information corresponding to the output object data.

[0015] Technical principles and effects of this solution: In this solution, the gas safety inspection image data to be identified is first automatically obtained from the gas safety inspection business system, which ensures the stability and timeliness of the data source and provides a high-quality data foundation for subsequent processing.

[0016] Afterwards, the collected gas safety inspection image data is combined with the preset object recognition model to perform object detection and classification on the collected images to form the corresponding object data.

[0017] Then, based on the object recognition results and the original image, combined with the preset object hidden danger recognition model, it is determined whether there are potential safety hazards and outputs detailed risk hidden danger information. Finally, once a risk hidden danger is detected, an early warning notification is immediately sent to the relevant personnel or departments to ensure that timely measures are taken.

[0018] In this solution, the object recognition module and the hidden danger recognition module are linked to perform object detection on the collected gas safety inspection image data to be identified, and the key components in the image (such as gas meters, valves, pipeline connection points, etc.) are accurately located and classified. This step ensures the basic accuracy of subsequent hidden danger identification. Based on the identified objects and their location information, the status and relationship of these objects are further analyzed in combination with the original image, so as to more accurately determine whether there are risk hazards. For example, it can more accurately identify whether there are signs of leakage in the gas pipeline or whether the equipment is in an abnormal working state. Through the preliminary object detection, background noise and other non-related elements can be filtered out, so that hidden danger identification is more focused on specific target objects, reducing the possibility of false alarms. Even in poor lighting conditions, large angle changes or occlusion, the hidden danger recognition module can still work effectively because the object detection provides preliminary screening and positioning, which enhances the system's ability to cope with complex environments. That is, it can realize separate detection of object detection and hidden danger identification, reduce identification interference, improve detection accuracy, and reduce the possibility of false alarms or missed alarms.

[0019] Furthermore, the server also includes:

[0020] The model building and training module is used to build and train the object recognition model and the object hidden danger recognition model respectively according to the historical gas safety inspection image data in the historical database.

[0021] Beneficial effects: In this solution, the object recognition model and the object hidden danger recognition model are constructed and trained separately, which reduces the interference of recognition and greatly improves the accuracy of recognition.

[0022] Furthermore, the model building training module includes:

[0023] A data retrieval module is used to retrieve historical gas safety inspection image data from a historical database;

[0024] The annotation module is used to annotate the retrieved historical gas safety inspection image data, and the corresponding annotation content is the historical object data corresponding to the historical gas safety inspection image data, as well as the historical hidden danger risk data;

[0025] A partitioning module is used to partition the annotated historical gas safety inspection image data into a training set and a validation set according to a preset partitioning ratio;

[0026] A first building module, used to build an object recognition model;

[0027] A random module is used to determine all model parameters corresponding to the object recognition model according to the constructed object recognition model, and randomly generate several model parameter groups;

[0028] The first training module is used to input the annotated historical gas safety inspection image data in the training set as input data into the object recognition model corresponding to each constructed model parameter group, and output the training prediction data corresponding to each model parameter group;

[0029] An evaluation module is used to calculate the model prediction accuracy corresponding to each model parameter group based on the training prediction data corresponding to each model parameter group and the corresponding annotation content based on a preset model prediction accuracy calculation formula;

[0030] A selection module is used to arrange the model prediction accuracies corresponding to the model parameter groups in descending order within a preset first number of iterations, and select them in descending order based on a preset selection ratio until the selection ratio is met;

[0031] After exceeding the preset first number of iterations, according to the model prediction accuracy corresponding to each model parameter group, the corresponding model parameter group is selected based on the preset model parameter group selection strategy;

[0032] A replacement module is used to randomly combine the selected model parameter groups in pairs to form corresponding replacement groups, and in the corresponding replacement groups, generate a random number for each model parameter of the two model parameter groups in the replacement group, and determine whether the corresponding two model parameter groups are to be replaced based on a preset random number judgment strategy. If so, randomly select a replacement segment to replace, otherwise, the model parameter group in the replacement group is not replaced;

[0033] A loop module, used for re-executing the first training module according to each replaced model parameter group after completing the replacement of all replacement groups, until a preset number of iterations is met;

[0034] An output module is used to select each model parameter group with a preset ranking after a preset number of iterations is met, and randomly select a model parameter group as the model parameter group corresponding to the object recognition model after the corresponding training is completed, and form an object recognition model corresponding to the model parameter group;

[0035] The verification module is used to input the historical gas safety inspection image data in the verification set into the corresponding object recognition model according to the object recognition model corresponding to the formed model parameter group, determine the verification accuracy corresponding to the corresponding object recognition model, and when the verification accuracy is greater than the preset verification threshold, it is judged that the corresponding object recognition model has completed training, otherwise the random module is re-executed.

[0036] Beneficial effects: Gas safety inspection image data is retrieved from the historical database. This ensures the stability and diversity of the data source, provides a rich sample basis for subsequent model training, and improves the model's generalization ability. The retrieved historical image data is labeled, including object data and hidden danger risk data. This provides high-quality supervision signals, enabling the model to learn accurate object features and hidden danger patterns, and enhancing the quality and guidance of training data.

[0037] When selecting the model parameter group for each iteration, first sort the model parameter groups from large to small according to their prediction accuracy, and select them according to the preset ratio. Model parameter groups with better performance are retained first, which speeds up the convergence to the global optimal solution. Although a high-accuracy model is selected, a certain number of suboptimal solutions are still maintained for subsequent exploration. Then, after exceeding the preset number of iterations, different model parameter group selection strategies are adopted, such as comprehensive considerations based on accuracy, complexity, and diversity. No longer relying solely on accuracy, other factors are comprehensively considered to ensure that the selected model not only has good performance, but is also easy to deploy and maintain. Avoid missing out on potential better solutions due to premature stopping, extend the exploration time, and improve the quality of the final model.

[0038] In the process of changing model parameters, random numbers are set to control the change of model parameters. By introducing randomness, the system can explore more possibilities in the parameter space and avoid falling into the local optimal solution. Increasing the exploration scope in the model optimization process helps to discover a wider and more effective combination of parameters, thereby improving the performance of the final model.

[0039] Furthermore, the preset model parameter group selection strategy is:

[0040] S10, after exceeding the preset first number of iterations, according to the model prediction accuracy corresponding to each model parameter group, selecting the model parameter group with the maximum model prediction accuracy corresponding to the corresponding number of iterations;

[0041] S20, determining the correlation between the selected model parameter group and other model parameter groups based on the difference between the selected model parameter group with the highest model prediction accuracy and the parameter values ​​corresponding to each model parameter in other model parameter groups;

[0042] S30, determining the increase coefficients corresponding to other model parameter groups according to the correlation between the selected model parameter group with the greatest model prediction accuracy and other model parameter groups;

[0043] S40. According to the increase coefficient corresponding to each other model parameter group and based on the model prediction accuracy corresponding to each model parameter group, the new model prediction accuracy corresponding to each other model parameter group is calculated, and the model parameter group with the largest new model prediction accuracy is selected, and S20 is re-executed. At this time, the remaining other model parameter groups are used as the other model parameter groups corresponding to this cycle until the number of selected model parameter groups reaches a preset threshold.

[0044] Beneficial effects: In this scheme, by selecting the model parameter group with the maximum preset accuracy corresponding to each cycle in turn, and by preliminarily selecting the model parameter group with the maximum accuracy, the optimization direction is quickly locked in, reducing unnecessary exploration time.

[0045] Then the correlation between the model parameter group and the remaining other model parameter groups is calculated, and the corresponding increase coefficient of other model parameter groups is determined, so as to update the model prediction accuracy corresponding to each other model parameter group. Even the model parameter groups with low correlation retain a certain opportunity, preventing premature convergence and ensuring the diversity and robustness of the model. As the number of iterations increases, the correlation and increase coefficient are continuously updated, so that the system can flexibly respond to the needs of different stages and gradually approach the global optimal solution.

[0046] Then, according to the prediction accuracy of the updated model, the largest one is selected, and S20 is executed again until the number of selected model parameter groups reaches the preset threshold. A cyclic optimization process is formed, gradually approaching the global optimal solution. Each cycle is adjusted based on the latest correlation and increase coefficient, which speeds up the improvement of model performance.

[0047] The method of increasing the coefficient by calculating the correlation degree and updating the model prediction accuracy by increasing the coefficient has played an important role in the model optimization process of the gas safety inspection management system. This strategy not only improves the directionality and accuracy of model optimization, but also enhances the stability and robustness of the model, accelerates the convergence speed, promotes comprehensive evaluation, and improves resource utilization efficiency.

[0048] Furthermore, the preset random number judgment strategy is:

[0049] According to the random numbers corresponding to the model parameters corresponding to the two model parameter groups in the replacement group, based on a preset random difference degree calculation formula, the random difference degrees corresponding to the two model parameter groups are calculated;

[0050] The preset random gap calculation formula is:

[0051]

[0052] In the formula, S is the random gap degree, P1 i is the random number corresponding to the i-th model parameter in one of the model parameter groups in the replacement group, P2 i is the random number corresponding to the i-th model parameter in another model parameter group in the replacement group; n is the number of model parameters in the model parameter group;

[0053] If the random difference degree is greater than the preset difference threshold, the two model parameter groups in the replacement group will not undergo parameter replacement. Otherwise, it is determined that the two model parameter groups in the replacement group need to undergo parameter replacement, and based on the preset parameter replacement number and the random difference between the corresponding model parameters in the two model parameter groups, the model parameters that need to be replaced in the two model parameter groups in the replacement group are selected, and the corresponding parameter replacement is performed.

[0054] Beneficial effects: In this solution, by calculating the random gap and setting the threshold, the system can identify the model parameter groups with large differences and avoid unnecessary parameter changes. This reduces the possibility of blind exploration and improves the optimization efficiency. Even when changing parameters, a certain proportion of suboptimal solutions is still retained to prevent premature convergence to the local optimal solution. At the same time, it not only relies on a single accuracy indicator, but also combines the random gap and other factors to ensure that the selected model not only has good performance, but also is stable and reliable.

[0055] Furthermore, the model building training module also includes:

[0056] The second building module is used to build an object hidden danger identification model;

[0057] The second training module is used to input the annotated historical gas safety inspection image data and the corresponding historical object data in the training set as input data into the constructed object hidden danger recognition model for iterative training;

[0058] After each iterative training is completed, the trained object hidden danger recognition model is output. The training effect of the trained object hidden danger recognition model is verified by inputting the historical gas safety inspection image data and the corresponding historical object data in the verification set to judge whether the training effect meets the standard. If not, continue training, otherwise stop training, and test the corresponding object hidden danger recognition model through the historical gas safety inspection image data and historical object data in the verification set. The test results output are judged on the feasibility of the object hidden danger recognition model based on the current test results and the previous test results. If it is not feasible, continue training, otherwise, output the corresponding object hidden danger recognition model.

[0059] Beneficial effects: By introducing a special object hidden danger identification model construction and training module, this solution not only solves the problems existing in the existing technology, but also significantly improves the accuracy of hidden danger identification and the robustness of the system.

[0060] The present invention also provides a gas safety inspection management method based on image recognition, using the above-mentioned gas safety inspection management system based on image recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a logic block diagram of a gas safety inspection and management system based on image recognition in Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0062] The following is further described in detail through specific implementation methods:

[0063] Embodiment 1

[0064] The gas safety inspection management system based on image recognition is basically Figure 1 As shown, including the server;

[0065] The server includes:

[0066] An image acquisition module is used to acquire corresponding gas safety inspection image data to be identified from the safety inspection business system;

[0067] The object recognition module is used to identify the objects corresponding to the gas safety inspection image data to be identified based on the collected gas safety inspection image data and a preset object recognition model, and generate corresponding object data; in this embodiment, the object recognition model can be used to identify corresponding object data such as water heaters, gas meters, stoves, under-stove hoses, front-of-meter pipes, alarms, security inspection forms, etc.

[0068] A hidden danger identification module is used to identify the risk hidden dangers corresponding to the object data in the gas safety inspection image data to be identified based on the identified object data and the corresponding gas safety inspection image data to be identified, based on a preset object hidden danger identification model, and output the risk hidden danger information of the corresponding object data;

[0069] The early warning module is used to send early warning information to users based on the risk information corresponding to the output object data.

[0070] The server also includes:

[0071] The model building and training module is used to build and train the object recognition model and the object hidden danger recognition model respectively according to the historical gas safety inspection image data in the historical database.

[0072] The model building training module includes:

[0073] A data retrieval module is used to retrieve historical gas safety inspection image data from a historical database;

[0074] The annotation module is used to annotate the retrieved historical gas safety inspection image data, and the corresponding annotation content is the historical object data corresponding to the historical gas safety inspection image data, as well as the historical hidden danger risk data;

[0075] A partitioning module is used to partition the annotated historical gas safety inspection image data into a training set and a validation set according to a preset partitioning ratio;

[0076] A first building module, used to build an object recognition model;

[0077] A random module is used to determine all model parameters corresponding to the object recognition model according to the constructed object recognition model, and randomly generate several model parameter groups;

[0078] The first training module is used to input the annotated historical gas safety inspection image data in the training set as input data into the object recognition model corresponding to each constructed model parameter group, and output the training prediction data corresponding to each model parameter group;

[0079] An evaluation module is used to calculate the model prediction accuracy corresponding to each model parameter group based on the training prediction data corresponding to each model parameter group and the corresponding annotation content based on a preset model prediction accuracy calculation formula;

[0080] The preset model prediction accuracy calculation formula is:

[0081]

[0082] In the formula, Y is the prediction accuracy of the model, m is the number of samples in the training set, is the training prediction data corresponding to the j-th historical gas safety inspection image data, y j is the annotation content corresponding to the j-th historical gas safety inspection image data, is an exponential function, when , the corresponding value is 1, otherwise it is 0.

[0083] A selection module is used to arrange the model prediction accuracies corresponding to the model parameter groups in descending order within a preset first number of iterations, and select them in descending order based on a preset selection ratio until the selection ratio is met;

[0084] After exceeding the preset first number of iterations, according to the model prediction accuracy corresponding to each model parameter group, the corresponding model parameter group is selected based on the preset model parameter group selection strategy;

[0085] The preset model parameter group selection strategy is:

[0086] S10, after exceeding the preset first number of iterations, according to the model prediction accuracy corresponding to each model parameter group, selecting the model parameter group with the maximum model prediction accuracy corresponding to the corresponding number of iterations;

[0087] S20, determining the correlation between the selected model parameter group and other model parameter groups based on the difference between the selected model parameter group with the highest model prediction accuracy and the parameter values ​​corresponding to each model parameter in other model parameter groups;

[0088] S30, determining the increase coefficients corresponding to other model parameter groups according to the correlation between the selected model parameter group with the greatest model prediction accuracy and other model parameter groups;

[0089] S40, according to the increase coefficient corresponding to each other model parameter group, based on the model prediction accuracy corresponding to each model parameter group, calculate the new model prediction accuracy corresponding to each other model parameter group, and select the model parameter group with the largest new model prediction accuracy, and re-execute S20, at which time the remaining other model parameter groups are used as the other model parameter groups corresponding to this cycle, until the number of selected model parameter groups reaches a preset threshold. For example, the initial model parameter group: [M_1, M_2, M_3, M_4, M_5];

[0090] Model prediction accuracy (hypothetical value):

[0091] (M 1 ):85%,(M 2 ):90%,(M 3 ):88%,(M 4 ):92%,(M_5):87%

[0092] The model with the highest prediction accuracy is (M_4) (92%), so (M_4) is selected as the current optimal solution. The correlation is determined based on the difference between the model parameter group ((M_4)) with the highest model prediction accuracy and the parameter values ​​corresponding to each model parameter in other model parameter groups. Based on the correlation, the increase coefficients corresponding to other model parameter groups are determined.

[0093] A replacement module is used to randomly combine the selected model parameter groups in pairs to form corresponding replacement groups, and in the corresponding replacement groups, generate a random number for each model parameter of the two model parameter groups in the replacement group, and determine whether the corresponding two model parameter groups are to be replaced based on a preset random number judgment strategy. If so, randomly select a replacement segment to replace, otherwise, the model parameter group in the replacement group is not replaced;

[0094] The preset random number judgment strategy is:

[0095] According to the random numbers corresponding to the model parameters corresponding to the two model parameter groups in the replacement group, based on a preset random difference degree calculation formula, the random difference degrees corresponding to the two model parameter groups are calculated;

[0096] The preset random gap calculation formula is:

[0097]

[0098] In the formula, S is the random gap degree, P1 i is the random number corresponding to the i-th model parameter in one of the model parameter groups in the replacement group, P2 iis the random number corresponding to the i-th model parameter in another model parameter group in the replacement group; n is the number of model parameters in the model parameter group;

[0099] If the random difference degree is greater than the preset difference threshold, the two model parameter groups in the replacement group will not undergo parameter replacement. Otherwise, it is determined that the two model parameter groups in the replacement group need to undergo parameter replacement, and based on the preset parameter replacement number and the random difference between the corresponding model parameters in the two model parameter groups, the model parameters that need to be replaced in the two model parameter groups in the replacement group are selected, and the corresponding parameter replacement is performed.

[0100] A loop module, used for re-executing the first training module according to each replaced model parameter group after completing the replacement of all replacement groups, until a preset number of iterations is met;

[0101] An output module is used to select each model parameter group with a preset ranking after a preset number of iterations is met, and randomly select a model parameter group as the model parameter group corresponding to the object recognition model after the corresponding training is completed, and form an object recognition model corresponding to the model parameter group;

[0102] The verification module is used to input the historical gas safety inspection image data in the verification set into the corresponding object recognition model according to the object recognition model corresponding to the formed model parameter group, determine the verification accuracy corresponding to the corresponding object recognition model, and when the verification accuracy is greater than the preset verification threshold, it is judged that the corresponding object recognition model has completed training, otherwise the random module is re-executed.

[0103] The model building training module also includes:

[0104] The second building module is used to build an object hidden danger identification model;

[0105] The second training module is used to input the annotated historical gas safety inspection image data and the corresponding historical object data in the training set as input data into the constructed object hidden danger recognition model for iterative training;

[0106] After each iterative training is completed, the trained object hidden danger recognition model is output. The training effect of the trained object hidden danger recognition model is verified by inputting the historical gas safety inspection image data and the corresponding historical object data in the verification set to judge whether the training effect meets the standard. If not, continue training, otherwise stop training, and test the corresponding object hidden danger recognition model through the historical gas safety inspection image data and historical object data in the verification set. The test results output are judged on the feasibility of the object hidden danger recognition model based on the current test results and the previous test results. If it is not feasible, continue training, otherwise, output the corresponding object hidden danger recognition model.

[0107] This embodiment also discloses a gas safety inspection management method based on image recognition, using the above-mentioned gas safety inspection management system based on image recognition.

[0108] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is described too much here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field of the invention before the application date or priority date, can know all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. Gas safety inspection and management system based on image recognition, characterized by: Including the server side; The server includes: An image acquisition module is used to acquire corresponding gas safety inspection image data to be identified from the safety inspection business system; The object recognition module is used to recognize the corresponding object in the gas safety inspection image data to be recognized based on the collected gas safety inspection image data to be recognized and based on the preset object recognition model, and generate corresponding object data; A hidden danger identification module is used to identify the risk hidden dangers corresponding to the object data in the gas safety inspection image data to be identified based on the identified object data and the corresponding gas safety inspection image data to be identified, based on a preset object hidden danger identification model, and output the risk hidden danger information of the corresponding object data; The early warning module is used to send early warning information to users based on the risk information corresponding to the output object data.

2. The gas safety inspection and management system based on image recognition according to claim 1 is characterized in that: The server also includes: The model building and training module is used to build and train the object recognition model and the object hidden danger recognition model respectively according to the historical gas safety inspection image data in the historical database.

3. The gas safety inspection and management system based on image recognition according to claim 2 is characterized in that: The model building training module includes: A data retrieval module is used to retrieve historical gas safety inspection image data from a historical database; The annotation module is used to annotate the retrieved historical gas safety inspection image data, and the corresponding annotation content is the historical object data corresponding to the historical gas safety inspection image data, as well as the historical hidden danger risk data; A partitioning module is used to partition the annotated historical gas safety inspection image data into a training set and a validation set according to a preset partitioning ratio; A first building module, used to build an object recognition model; A random module is used to determine all model parameters corresponding to the object recognition model according to the constructed object recognition model, and randomly generate several model parameter groups; The first training module is used to input the annotated historical gas safety inspection image data in the training set as input data into the object recognition model corresponding to each constructed model parameter group, and output the training prediction data corresponding to each model parameter group; An evaluation module is used to calculate the model prediction accuracy corresponding to each model parameter group based on the training prediction data corresponding to each model parameter group and the corresponding annotation content based on a preset model prediction accuracy calculation formula; A selection module is used to arrange the model prediction accuracies corresponding to the model parameter groups in descending order within a preset first number of iterations, and select them in descending order based on a preset selection ratio until the selection ratio is met; After exceeding the preset first number of iterations, according to the model prediction accuracy corresponding to each model parameter group, the corresponding model parameter group is selected based on the preset model parameter group selection strategy; A replacement module is used to randomly combine the selected model parameter groups in pairs to form corresponding replacement groups, and in the corresponding replacement groups, generate a random number for each model parameter of the two model parameter groups in the replacement group, and determine whether the corresponding two model parameter groups are to be replaced based on a preset random number judgment strategy. If so, randomly select a replacement segment to replace, otherwise, the model parameter group in the replacement group is not replaced; A loop module, used for re-executing the first training module according to each replaced model parameter group after completing the replacement of all replacement groups, until a preset number of iterations is met; An output module is used to select each model parameter group with a preset ranking after a preset number of iterations is met, and randomly select a model parameter group as the model parameter group corresponding to the object recognition model after the corresponding training is completed, and form an object recognition model corresponding to the model parameter group; The verification module is used to input the historical gas safety inspection image data in the verification set into the corresponding object recognition model according to the object recognition model corresponding to the formed model parameter group, determine the verification accuracy corresponding to the corresponding object recognition model, and when the verification accuracy is greater than the preset verification threshold, it is judged that the corresponding object recognition model has completed training, otherwise the random module is re-executed.

4. The gas safety inspection and management system based on image recognition according to claim 3 is characterized in that: The preset model parameter group selection strategy is: S10, after exceeding the preset first number of iterations, according to the model prediction accuracy corresponding to each model parameter group, selecting the model parameter group with the maximum model prediction accuracy corresponding to the corresponding number of iterations; S20, determining the correlation between the selected model parameter group and other model parameter groups based on the difference between the selected model parameter group with the highest model prediction accuracy and the parameter values ​​corresponding to each model parameter in other model parameter groups; S30, determining the increase coefficients corresponding to other model parameter groups according to the correlation between the selected model parameter group with the greatest model prediction accuracy and other model parameter groups; S40. According to the increase coefficient corresponding to each other model parameter group and based on the model prediction accuracy corresponding to each model parameter group, the new model prediction accuracy corresponding to each other model parameter group is calculated, and the model parameter group with the largest new model prediction accuracy is selected, and S20 is re-executed. At this time, the remaining other model parameter groups are used as the other model parameter groups corresponding to this cycle until the number of selected model parameter groups reaches a preset threshold.

5. The gas safety inspection and management system based on image recognition according to claim 4 is characterized in that: The preset random number judgment strategy is: According to the random numbers corresponding to the model parameters corresponding to the two model parameter groups in the replacement group, based on a preset random difference degree calculation formula, the random difference degrees corresponding to the two model parameter groups are calculated; The preset random gap calculation formula is: Where S is the random gap degree, P1 i is the random number corresponding to the i-th model parameter in one of the model parameter groups in the replacement group, P2 i is the random number corresponding to the i-th model parameter in another model parameter group in the replacement group; n is the number of model parameters in the model parameter group; If the random difference degree is greater than the preset difference threshold, the two model parameter groups in the replacement group will not undergo parameter replacement. Otherwise, it is determined that the two model parameter groups in the replacement group need to undergo parameter replacement, and based on the preset parameter replacement number and the random difference between the corresponding model parameters in the two model parameter groups, the model parameters that need to be replaced in the two model parameter groups in the replacement group are selected, and the corresponding parameter replacement is performed.

6. The gas safety inspection and management system based on image recognition according to claim 5 is characterized in that: The model building training module also includes: The second building module is used to build an object hidden danger identification model; The second training module is used to input the annotated historical gas safety inspection image data and the corresponding historical object data in the training set as input data into the constructed object hidden danger recognition model for iterative training; After each iterative training is completed, the trained object hidden danger recognition model is output. The training effect of the trained object hidden danger recognition model is verified by inputting the historical gas safety inspection image data and the corresponding historical object data in the verification set to judge whether the training effect meets the standard. If not, continue training, otherwise stop training, and test the corresponding object hidden danger recognition model through the historical gas safety inspection image data and historical object data in the verification set. The test results output are judged on the feasibility of the object hidden danger recognition model based on the current test results and the previous test results. If it is not feasible, continue training, otherwise, output the corresponding object hidden danger recognition model.

7. A gas safety inspection management method based on image recognition, characterized in that: A gas safety inspection and management system based on image recognition using any one of claims 1 to 6 above.

Images