A real-time fire monitoring image enhancement system

Abstract

The present application relates to a kind of real-time fire monitoring image enhancement systems, including noise removal sub-model and resolution recovery sub-model;The noise removal sub-model removes the noise of original fire monitoring image, and the image after processing is input resolution recovery sub-model, and resolution is promoted and image is reconstructed.The present application improves the image definition and resolution, and guarantees monitoring quality.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and fire monitoring technology, specifically to a real-time fire monitoring image enhancement system. Background Technology

[0002] Fire accidents have always been a potential hazard accompanying human life and production, and their randomness is one of the key factors contributing to the frequent occurrence of accidents in reality. While human control is insufficient to completely prevent fire accidents, preventative measures are essential. The emergence of fire monitoring has filled a gap in fire prevention within the firefighting field, becoming an important means of disaster early warning in modern society. Among these, image monitoring provides more dynamic information, serving as a crucial basis for fire rescue tactics and reducing fire losses. However, it is often susceptible to interference from smoke and ambient light at the fire scene, leading to missing image information. Furthermore, camera equipment relies on electronic components for photoelectric signal conversion, and its performance and stability are affected by the temperature at the fire scene; overheating often results in a large amount of thermal noise in the image. Therefore, enhancing fire monitoring images is crucial for ensuring sufficient information acquisition and maintaining the quality and quantity of monitoring work. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a real-time fire monitoring image enhancement system that removes noise and smoke from fire monitoring images, improves image clarity and resolution, and ensures monitoring quality.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A real-time fire monitoring image enhancement system includes a noise removal sub-model and a resolution restoration sub-model. The noise removal sub-model removes noise from the original fire monitoring image and inputs the processed image into the resolution restoration sub-model for resolution enhancement and image reconstruction.

[0006] Furthermore, the noise removal sub-model includes an encoder-decoder; the encoder-decoder adopts a symmetric convolutional neural network, and the network is configured with 16 convolutional layers with different convolutional kernels, 16 batch normalization layers, and 16 activation layers.

[0007] Furthermore, the first 8 convolutional layers of the codec each use skip connections under the spatial attention module, while the remaining convolutional layers use direct skip connections.

[0008] Furthermore, the spatial attention module employs a convolutional neural network with the sigmoid function as the activation function, and the output is the weights for enhancing or suppressing the information flow.

[0009] Furthermore, the resolution restoration sub-model includes a multi-segment resolution enhancement module and an information flow stack module integrating a small neural network.

[0010] Furthermore, the multi-segment resolution enhancement module employs five convolutional neural networks connected by a symmetrical encoding and decoding structure. Each convolutional neural network includes 14 convolutional layers with different convolutional kernels, 14 batch normalization layers, and 14 activation layers.

[0011] Furthermore, the first three convolutional layers of the multi-segment resolution enhancement module each use skip connections under spatial attention, while the remaining convolutional layers use direct skip connections.

[0012] Furthermore, the integrated small neural network information flow stack module includes: nested network loss functions and repeatedly computed information flow structures.

[0013] Compared with the prior art, the present invention has the following advantages:

[0014] This invention achieves noise and smoke removal in fire monitoring images by establishing noise removal and resolution restoration sub-models. The spatial attention module technology used in the sub-models effectively filters noisy data, while skip connection technology compensates for the weakening of information intensity in the model. Multi-segment resolution enhancement technology and information flow stacking technology refine the resolution reconstruction process. These technologies and model structures ultimately improve the model's performance and ensure the quality of fire monitoring image enhancement. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the model architecture of the present invention;

[0016] Figure 2 This is a schematic diagram of the noise removal sub-model of the present invention;

[0017] Figure 3 This is a schematic diagram of the spatial attention module of the present invention;

[0018] Figure 4 This is a schematic diagram of the resolution restoration sub-model of the present invention;

[0019] Figure 5 This is a diagram illustrating the noise reduction effect;

[0020] Figure 6 This is a schematic diagram of the multi-segment resolution enhancement effect. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] Please refer to Figure 1The present invention provides a real-time fire monitoring image enhancement system, including a noise removal sub-model and a resolution restoration sub-model; the noise removal sub-model removes noise from the original fire monitoring image and inputs the processed image into the resolution restoration sub-model for resolution enhancement and image reconstruction.

[0023] In this embodiment, reference Figure 2 The noise removal sub-model takes a noisy image as input and passes it through a 16-layer encoder for noise data identification and filtering. Each layer of the encoder has a similar structure: convolutional layer, batch normalization layer, and activation layer. At the end of the encoder's processing of the image information stream, the information stream becomes a latent variable, awaiting the decoder's data spatial mapping, i.e., the noise reduction process. The decoder consists of 16 upsampling layers corresponding to the encoder's structure, with each layer showing: deconvolutional layer, batch normalization layer, and activation layer. Information streams are shorted between encoder-decoder layers, and spatial attention modules are used only in the shorting of the first 8 layers to enhance and filter the information stream.

[0024] In this embodiment, the spatial attention module is a small, lightweight convolutional neural network. By convolving the input information stream, weights corresponding to each data point can be obtained, i.e., a weight map whose size matches the information stream. After activation by a special activation function, sigmoid, the weight map is limited to the range of 0-1. The dot product weighting of the weight map and the information stream can effectively enhance or suppress each data point in the information stream.

[0025] In this embodiment, the multiple convolutional neural networks integrated in the resolution restoration sub-model are similar in structure to the noise removal sub-model, also featuring an encoding / decoding structure and skip connections under spatial attention. Information flow between the multiple networks is interconnected, and each network is responsible for only a small amount of resolution reconstruction. Therefore, in the last network, the restoration efficiency of the information flow is additive.

[0026] Figure 5 The diagram illustrates the noise removal effect. By simulating noise, a Gaussian distribution is used to randomly add noise to the fire image, with noise types including white, black, and mixed-color noise. The noise samples are shown in the leftmost column of images, the middle column is the image before noise addition, and the right column is the denoised image output by the noise removal sub-model. Visually, the powerful noise removal capability of this sub-model is evident.

[0027] Figure 6The diagram illustrates the multi-segment resolution enhancement effect. By simulating smoke blurring flames, Gaussian blurring is used to impair the resolution of a fire image. The leftmost image shows the input to this sub-model, while the middle images represent intermediate values ​​processed by the model, demonstrating the gradual reconstruction of resolution. The rightmost images show the model's final output and the image before blurring; the two images are visually very similar, indicating the effectiveness and stability of the sub-model in reconstructing image resolution.

[0028] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0029] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0030] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0031] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0032] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A real-time fire monitoring image enhancement system, characterized in that, It includes a noise removal sub-model and a resolution restoration sub-model; the noise removal sub-model removes noise from the original fire monitoring image, and inputs the processed image into the resolution restoration sub-model for resolution enhancement and image reconstruction; The noise removal sub-model includes an encoder-decoder; the encoder-decoder adopts a symmetric convolutional neural network, which has 16 convolutional layers with different kernels, 16 batch normalization layers and 16 activation layers. The first 8 convolutional layers of the codec each use skip connections under the spatial attention module, while the remaining convolutional layers use direct skip connections. The spatial attention module uses a convolutional neural network with the sigmoid activation function, and the output is the weights for enhancing or suppressing the information flow. The resolution restoration sub-model includes a multi-segment resolution enhancement module and an information flow stack module integrating a small neural network; The multi-segment resolution enhancement module uses five convolutional neural networks connected by a symmetrical encoding and decoding structure. Each convolutional neural network includes 14 convolutional layers with different convolutional kernels, 14 batch normalization layers, and 14 activation layers. The first three convolutional layers of the multi-segment resolution enhancement module each use skip connections under spatial attention, while the remaining convolutional layers use direct skip connections. The integrated small neural network information flow stack module includes: nested network loss functions and repeatedly computed information flow structures.

Images