KR20220061071A - Method for detecting fire, and learning neural network model for fire detection based on block chain system - Google Patents

Abstract

The present invention relates to a learning method of a fire monitoring model. The blockchain-based learning method of a fire monitoring model comprises the steps of: releasing a fire monitoring model which performs at least one convolution operation on an image captured from a camera so that at least one fire monitoring module corresponding to each node of a blockchain can use; receiving, by each node, update information of the fire monitoring model generated according to a fire monitoring result of the released fire monitoring model; and verifying, by each node, the update information and updating the fire monitoring model using the verified update information. The present invention can improve the performance of the fire monitoring model.

Description

{Method for detecting fire, and learning neural network model for fire detection based on block chain system}

The present invention relates to a fire monitoring method based on a neural network model, and more particularly, to a learning method of a fire monitoring model using block chain technology.

Since fires cause extensive human and material damage when they occur, sensing technologies are being developed for these areas that require prevention and early detection, and products that determine the occurrence of fires are being commercialized in various forms.

However, in the case of a conventional sensor that judges heat, smoke, or flame, it is possible to judge only in a local area, and a detection error can easily occur depending on the surrounding environment, so there is a limit in detecting the occurrence of a fire in a wider area.

Recently, with the development of artificial intelligence technology, artificial intelligence technologies are being applied to various fields. In particular, artificial intelligence technology is mainly applied to the field of detection and tracking of objects in images, and has the advantage of being able to detect and track objects in a shorter time than conventional image processing.

Therefore, by applying this artificial intelligence technology to CCTV images, an intelligent monitoring system that has little influence on the surrounding environment and monitors a large open area is being operated, and a technology to detect a fire using this is being developed.

The artificial neural network model used in artificial intelligence is trained to detect fire-related flames or smoke as an object in the image, and has the advantage of being able to detect fire in a shorter time than conventional image processing.

However, these disasters such as fire and disaster situation data have little regularity and it is difficult to obtain learning data than data used for general object recognition. there is.

Therefore, it is necessary to train the neural network model by using the actual monitored image as learning data, and it is necessary to efficiently manage the monitoring results of a plurality of images and to continuously update the neural network model distributed in various regions.

An object of the present invention is to propose an efficient learning and management method of a neural network model for fire monitoring in order to solve the above technical problem.

Another object of the present invention is to propose a method for a fire monitoring model to learn based on various images captured locally.

Another object of the present invention is to propose a method for efficiently managing a fire monitoring model using a block chain system.

A block chain-based learning method of a fire monitoring model according to the present invention for solving the above technical problem is a fire monitoring model that performs at least one convolution operation on an image taken from a camera to each node of the block chain. distributing (Release) the corresponding at least one fire monitoring module to be available; receiving, by each node, update information of the fire monitoring model generated according to a fire monitoring result of the distributed fire monitoring model; and verifying the updated information by each node and updating the fire monitoring model using the verified updated information.

Preferably, the update information is generated using a user response to a fire monitoring result provided through the fire monitoring module.

In the receiving step, it is preferable to compare the output value of the fire monitoring model with a user response and receive, as update information, weight change information of a layer performing the composite product operation.

In the updating step, it is preferable to verify the received update information from each node based on the block chain system, and update the fire monitoring model distributed to each node using the verified update information.

The block chain system further includes the step of managing the version of the updated fire monitoring model and redistributing the updated fire monitoring model to each fire monitoring module.

Preferably, the fire monitoring model generates one-time update information by inputting a captured image.

A block chain-based learning system of a fire monitoring model according to the present invention for solving the above technical problem is a fire monitoring model that performs at least one convolution operation on an image taken from a camera to each node of the block chain. a fire monitoring model providing unit for distributing (Release) the corresponding at least one fire monitoring module to be available; an update information receiving unit for each node to receive update information of the fire monitoring model generated according to a fire monitoring result of the distributed fire monitoring model; and a fire monitoring model update unit that verifies the updated information by each node and updates the fire monitoring model using the verified updated information.

Preferably, the update information is generated using a user response to the identification result provided through the user terminal.

Preferably, the update unit verifies the received update information from each node based on the block chain system, and updates the fire monitoring model distributed to each node using the verified update information.

A block chain-based monitoring method of a fire monitoring model according to the present invention for solving the above technical problem comprises the steps of receiving a monitoring image; outputting a fire monitoring result using a fire monitoring model that performs at least one composite product operation on the input monitoring image; learning the fire monitoring model using the fire monitoring result and generating update information for the composite product operation; and transmitting the updated information to a node in a blockchain system that manages the fire monitoring model.

According to the present invention, an image that is actually taken can be used for learning as learning data, and thus the performance of the fire monitoring model can be improved.

In addition, according to the present invention, it is possible to update the fire monitoring model by efficiently verifying the update information performed and generated in each monitoring area based on the block chain system.

In addition, according to the present invention, it is possible to prevent the leakage of unnecessary information when the fire monitoring model is updated using the block chain system.

1 is a diagram illustrating a system configuration in which a block chain-based fire monitoring model operates according to an embodiment of the present invention.
2 is a diagram illustrating the operation of a block chain-based fire monitoring module according to an embodiment of the present invention.
3 is a diagram illustrating the configuration of a block chain system according to an embodiment of the present invention.
4 is a diagram illustrating an operation of a node constituting a block chain system according to an embodiment of the present invention.
5 is a diagram illustrating a method of updating a fire monitoring model through a block chain system according to an embodiment of the present invention.
6 is a diagram illustrating a node configuration of a block chain system according to an embodiment of the present invention.
7 is a diagram illustrating the operation of a block chain-based fire monitoring module according to an embodiment of the present invention.

The following is merely illustrative of the principles of the invention. Therefore, those skilled in the art can devise various devices that, although not explicitly described or shown herein, embody the principles of the invention and are included in the spirit and scope of the invention. In addition, it should be understood that all conditional terms and examples listed herein are, in principle, expressly intended only for the purpose of understanding the inventive concept, and are not limited to the specifically enumerated embodiments and states. .

The above objects, features and advantages will become more apparent through the following detailed description in relation to the accompanying drawings, and accordingly, those of ordinary skill in the art to which the invention pertains will be able to easily practice the technical idea of the invention. .

In addition, in the description of the invention, if it is determined that a detailed description of a known technology related to the invention may unnecessarily obscure the gist of the invention, the detailed description thereof will be omitted. Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram illustrating a system configuration in which a block chain-based fire monitoring model 25 according to an embodiment of the present invention operates.

Referring to FIG. 1 , the fire monitoring method according to the present embodiment may be implemented on a block chain system 1000 .

The fire monitoring method may be implemented with each node 100 constituting the block chain system 1000 and the fire monitoring modules 200 associated with each node 100 .

The block chain system 1000 may manage a program used in the fire monitoring module 200 through each node 100 , and distribute the program through the node 100 corresponding to the monitoring module 200 .

Each node 100 may directly correspond to each fire monitoring module 200 or may correspond to a plurality of monitoring modules 200 in units of monitoring areas.

At this time, the fire monitoring module 200 may be configured to include, for example, a plurality of CCTV (Closed Circuit Television) 11, 12, 13, such as an image photographing device, the image obtained from the CCTV installed for the monitoring area. , and fire monitoring information can be provided to the user 5 .

The program used by the fire monitoring module 200 may be implemented based on an artificial neural network. The fire monitoring module 200 may self-analyze the captured image using the learned neural network model, determine whether a fire-related flame or smoke exists, and inform the user 5 of the occurrence of a fire.

In this embodiment, the neural network model used for fire monitoring is managed in units of the node 100 that generally performs the valid proof of the transaction in the block chain network, and the validity of the update information generated when the neural network model is trained can be verified. there is.

In addition, the version of a program or application including the neural network model can be managed for each node 100, and the prediction performance can be continuously upgraded by configuring the master node separately and managing the master neural network model through this.

Hereinafter, an operating method of the fire monitoring module 200 according to an embodiment of the present invention will be described with reference to FIG. 2 .

In the present embodiment, the fire monitoring module 200 may directly analyze the image captured from each photographing device 11 using the neural network model-based fire monitoring model 25 and inform the user 5 of the analysis result.

The fire monitoring model 25 is an algorithm or program that performs and learns fire monitoring through an input image, and is also called a deep neural network. The fire monitoring model 25 generates a captured image as an input value, and whether a fire has occurred and a location within the image, which are the recognition results, as output values. In addition, the neural network in the pre-designed model can be further trained using the output value.

Accordingly, the user 5 can check the analysis result of the fire monitoring module 200 and generate a warning according to the occurrence of a fire, and it is also possible to generate and input a response from the user 5 to verify the analysis result.

For example, the fire monitoring module 200 may use the neural network-based fire monitoring model 25 to output the presence or absence of a flame in the image and the location of the flame when it exists as coordinate values. It can be checked and the authenticity of the result (TRUE or FALSE) can be generated as a response.

If the output result of the fire monitoring model 25 is an incorrect answer, the user 5 provides a response to the user 5 to the monitoring module 200 and the monitoring module 200 updates information for learning the neural network model. (20) can be created.

That is, in the present embodiment, the learning of the fire monitoring model 25 may be performed through a user response to an image that is actually captured, and such updated information 20 is transmitted to the corresponding node 100 through the monitoring module 200 . It can be transmitted and used for update after validation of validity.

In addition, the update information 20 may be composed of blocks managed in the blockchain network, and the node 100 verifies the validity of the update information 20, creates a new block if it is valid, and updates the entire neural network model. can be used

That is, in this embodiment, the fire monitoring model 25 is managed based on the block chain network, and the update history can be managed with the block chain, making it impossible to arbitrarily change the fire monitoring model 25 .

Also, in the present embodiment, the node 100 may verify the validity of the received plurality of update information 20 .

Referring to FIG. 3 , in this embodiment, the node 100 of the block chain system 1000 may correspond to a plurality of monitoring modules 200 , and thus a plurality of update information 20 are received simultaneously or within a certain period of time. can do.

Upon receiving this, the node 100 may determine whether the updated information is actually generated during the fire monitoring process by collecting the updated information and verifying the validity. Alternatively, it is also possible to verify the validity by considering the version information of the fire monitoring model 25 that has generated the updated information together.

Furthermore, when a plurality of pieces of update information are received, the node 100 in the present embodiment can be created as a subject that manages a unit area corresponding to a specific monitoring area, so using the consistency of update information based on spatial characteristics. It is also possible to verify validity.

For example, since a specific node 100 and a node adjacent to each other may correspond to the monitoring modules 200 that monitor regions at similar locations, it is possible to check whether there is consistency or similarity in the characteristics of the update information and verify the validity. . Consistency or similarity can be judged based on the physical and temporal relevance of the fire occurrence.

Alternatively, it is also possible to use the distributed consensus algorithm of the blockchain in general for validation of validity.

According to this embodiment, since update information is generated for each node 100 through the block chain system 1000 and the validity is verified, there are several advantages compared to the existing centralized neural network model management method.

When a specific central server directly receives all input data of the neural network model and performs learning through it, data purification work to individually determine the learning potential of the input data may be required, and thus resources can be saved.

In addition, since the fire monitoring model according to the present embodiment performs learning using an actual photographed image, a central specific server may receive various information other than information related to fire in the image.

For example, design information of a specific facility in the image, personal information of individual workers, etc. may be included in the image, and if it is centralized and managed, confidential information may be leaked, so the management of personal information may incur costs.

However, according to the present embodiment, the update information of each fire monitoring model 25 is generated at the local stage based on the node 100 of the block chain network 1000, and the generated update information is of the fire monitoring model 25. As information about the network layers, it is data different from the captured image, so a risk factor due to information leakage can be removed.

Hereinafter, a learning method of the block chain-based fire monitoring model according to the present embodiment will be described with reference to FIG. 4 .

Distribute the fire monitoring model 25 for training (42). In this embodiment, the fire monitoring model 25 may be distributed to the fire monitoring module 200 corresponding to each node 100 using the block chain system 1000 .

Each node 100 may receive updated information according to the fire monitoring result of the fire monitoring model 25 (44).

Referring to FIG. 5 in relation to the update information, the update information may be generated using a monitoring result of the fire monitoring model 25 and a user response. In addition, the update information may be specifically configured as an update value of a layer constituting the neural network of the fire monitoring model 25 .

Referring to FIG. 5 , the fire monitoring model 25 may be designed based on a Convolution Neural Network (CNN) algorithm. The fire monitoring model 25 may be composed of at least one layer that performs convolution on the input image 50, and each layer applies filters of various sizes to the input image to determine pixel values. A convolution operation can be performed on .

The performed result can finally output the designed result such as the existence of an object, the location of the object, the meaning of the object, etc. through a fully-connected layer (FC), and in this embodiment, the fire monitoring model ( 25) can generate the presence and location of flame or smoke in the input image as a result.

A result output through the display device of the monitoring module 200 may be displayed to the user, and the authenticity of the monitoring result may be checked and inputted. Alternatively, it is also possible to directly designate an incorrectly recognized area to the user or provide an interface so that the user can directly correct it to change it.

When the user's response to the monitoring result is received through the above user interface, the user's response is reflected in the monitoring result of the fire monitoring model 25, and the monitoring result reflecting the user's response is an error as an input of the fire monitoring model 25 The correction value of the network layer that generated the value can be calculated.

Specifically, in order to update the fire monitoring model 25, learning can be performed in such a way that the error of the layer is calculated using a back-propagation process and the weight of the filter applied to the layer that generated the error is updated. there is.

Alternatively, it is also possible to generate error values as update information through a feed forward process of the update result reflecting the user's response to the final correct answer of the fire monitoring model 25 . At this time, the monitoring module 200 transmits the generated error values to the node 100, and the master node (not shown) performs reverse propagation to the network of the previous version of the fire monitoring model 25 using the received error values. And it is also possible to perform the update.

At this time, in this embodiment, since the update is performed based on the node 100 of the block chain system 1000, the validity of the received update information can be verified using a distributed consensus method before the update (46).

When the validity of the updated information is verified, the fire monitoring model 25 may be updated (48). The updated fire monitoring model 25 can be version managed as a block in the block chain system 1000, and thus it becomes impossible to arbitrarily change the fire monitoring model 25.

Therefore, it is possible to prevent a situation in which the fire monitoring model 25 is changed and malfunctions from external hacking or malicious code, and the fire monitoring model ( 25), it is possible to stably improve the performance of the entire fire monitoring system.

Hereinafter, the configuration of the node 100 of the block chain system 1000 according to an embodiment of the present invention will be described with reference to FIG. 6 .

Referring to FIG. 6 , the node 100 according to the present embodiment includes a fire monitoring model providing unit 110 , an updated information receiving unit 120 , an updated information verifying unit 130 , and a fire monitoring model updating unit 140 . can do.

The fire monitoring model providing unit 110 may provide a program for determining whether a fire situation exists in the image captured according to the present embodiment to the monitoring module 200 for performing actual monitoring.

The monitoring module 200 may correspond to the specific node 100 by downloading the fire monitoring model 25 directly from the web or by authenticating it through the web after installation, and the computer processor in which the monitoring module 200 is installed is directly installed. It is also possible to configure the node 100 of the blockchain system 1000 .

The update information receiving unit 120 receives the update information generated through the monitoring result performed by the corresponding monitoring module 200 .

In this case, the update information may be information generated by learning the monitoring result and the user's response to the monitoring result by the fire monitoring model 25, and specifically may include changed values of a specific network layer for updating the monitoring model. .

The update information includes correction information of at least one or more layers constituting the neural network of the fire monitoring model 25, and may be composed of values for changing the weight of a filter used for each layer.

The update information verification unit 130 may verify the validity of the received update information using the block chain system 1000 .

The validity can be verified by a distributed consensus method using the characteristics of the block chain system 1000, and it is also possible to verify the validity in consideration of the geographical characteristics of the node 100 and the timing characteristics of generation of update information.

The fire monitoring model update unit 140 updates the fire monitoring model 25 when the updated information is verified. Specifically, it is also possible to directly update the fire monitoring model 25 distributed from the node 100 to the monitoring module 200, or to request the update request to the master node that manages the entire master fire monitoring model. It is possible.

Since the update information generated through the above process is used for update after validation using the block chain system 1000, it is impossible to arbitrarily change the neural network model, thus improving the stability of neural network models applied to dangerous situations such as fire monitoring. can do it

In addition, the update information obtained in various environments can be stably used for the update, and the locally performed and learned result is reflected, and thus has strong adaptability to the user's actual use environment.

Hereinafter, a method for generating fire monitoring and update information by the fire monitoring module 200 according to the present embodiment will be described with reference to FIG. 7 .

First, the fire monitoring module 200 may receive a monitoring image from a photographing device (72). Specifically, it is possible to receive an image captured from a CCTV or a video captured from a device capable of collecting images in various environments, such as a drone or an IR (Infrared) camera, as a surveillance image.

When a monitoring image is input, a fire monitoring result may be output (74).

It is also possible for the user 5 to confirm this through the output result and sound the fire alarm, or to automatically operate the fire alarm according to the probability value of the fire monitoring result. Alternatively, if the user 5 judges that it is not a fire situation, it is also possible to confirm this as an erroneous judgment.

That is, in the present embodiment, it is also possible to perform the learning process of the fire monitoring model 25 using the response of the user 5 in addition to the quick response to the fire situation through the fire monitoring model 25 (76). When the result of the fire monitoring model 25 is incorrect, learning of the fire monitoring model 25 may be performed using the monitoring result output with respect to the captured image and the response of the user 5 .

Learning is performed and the fire monitoring model 25 may generate updated information (78).

Specifically, it is possible to generate update information of the network layer of the fire monitoring model 25 that needs to be changed in the learning process. A method such as a back propagation algorithm or a feed forward may be used.

The monitoring module 200 may transmit the update information generated through the above process to the node 100 of the block chain system 1000 ( 80 ). The next node 100 may verify the received update information using the block chain system 1000 as described above and allow it to be used for actual update.

According to the present invention, the actually captured image can be used for learning as learning data, and thus the performance of the fire monitoring model 25 can be improved.

In addition, according to the present invention, it is possible to update the fire monitoring model 25 by efficiently verifying the update information generated and performed in each monitoring area based on the block chain system 1000 .

In addition, according to the present invention, it is possible to prevent the leakage of unnecessary information according to the update of the fire monitoring model 25 by using the block chain system 1000 .

Above, various embodiments described herein may be implemented in a computer-readable recording medium using, for example, software, hardware, or a combination thereof.

According to the hardware implementation, the embodiments described herein include ASICs (application specific integrated circuits), DSPs (digital signal processors), DSPDs (digital signal processing devices), PLDs (programmable logic devices), FPGAs (field programmable gate arrays, It may be implemented using at least one of processors, controllers, micro-controllers, microprocessors, and other electrical units for performing functions. The described embodiments may be implemented in the control module itself.

According to the software implementation, embodiments such as the procedures and functions described in this specification may be implemented as separate software modules. Each of the software modules may perform one or more functions and operations described herein. The software code may be implemented as a software application written in a suitable programming language. The software code may be stored in the memory module and executed by the control module.

The above description is merely illustrative of the technical idea of the present invention, and those of ordinary skill in the art to which the present invention pertains may make various modifications, changes and substitutions within the scope without departing from the essential characteristics of the present invention. will be.

Accordingly, the embodiments disclosed in the present invention and the accompanying drawings are for explaining, not limiting, the technical spirit of the present invention, and the scope of the technical spirit of the present invention is not limited by these embodiments and the accompanying drawings . The protection scope of the present invention should be construed by the following claims, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of the present invention.

Claims (12)

distributing a fire monitoring model that performs at least one convolution operation on an image captured from a camera so that at least one fire monitoring module corresponding to each node of the block chain can be used;
receiving, by each node, update information of the fire monitoring model generated according to a fire monitoring result of the distributed fire monitoring model; and
The block chain-based learning method of a fire monitoring model, comprising the step of verifying the update information by each node and updating the fire monitoring model using the verified update information. The method of claim 1,
The update information is a block chain-based learning method of a fire monitoring model, characterized in that it is generated using a user response to a fire monitoring result provided through the fire monitoring module. 3. The method of claim 2,
The receiving is a block chain-based learning method of a fire monitoring model, characterized in that by comparing the output value of the fire monitoring model with the user response, weight change information of the layer performing the composite product operation is received as update information. The method of claim 1,
The updating step is
A block chain-based learning method of a fire monitoring model, characterized in that the received update information from each node is verified based on a block chain system, and the fire monitoring model distributed to each node is updated using the verified update information. 5. The method of claim 4,
The block chain system is a block chain-based learning method of a fire monitoring model, characterized in that it further comprises the step of managing the updated version of the fire monitoring model and redistributing the updated fire monitoring model to each fire monitoring module. The method of claim 1,
The fire monitoring model is a block chain-based learning method of a fire monitoring model, characterized in that it generates one-time update information by inputting a captured image. A fire monitoring model that releases a fire monitoring model that performs at least one convolution operation on an image taken from a camera so that at least one fire monitoring module corresponding to each node of the block chain can be used study;
an update information receiving unit for each node to receive update information of the fire monitoring model generated according to a fire monitoring result of the distributed fire monitoring model; and
and a fire monitoring model update unit that verifies the update information by each node and updates the fire monitoring model using the verified update information. 8. The method of claim 7,
The update information is a block chain-based learning system of a fire monitoring model, characterized in that it is generated using a user response to the identification result provided through the user terminal. 8. The method of claim 7,
The update unit,
A block chain-based learning system for a fire monitoring model, characterized in that the received update information from each node is verified based on a block chain system, and the fire monitoring model distributed to each node is updated using the verified update information. receiving a surveillance video;
outputting a fire monitoring result using a fire monitoring model that performs at least one composite product operation on the input monitoring image;
learning the fire monitoring model using the fire monitoring result and generating update information for the composite product operation; and,
and transmitting the updated information to a node in a block chain system that manages the fire monitoring model. 11. The method of claim 10,
The node receives the transmitted update information, verifies it based on the block chain system, and uses the verified update information to update the fire monitoring model distributed to each node. Way. 12. one or more computers and, when executed by the one or more computers, causes the one or more computers to perform the operations of each method of any one of claims 1 to 6 and 10 to 11, causing the computer to A program stored on a readable recording medium.

Images