Lightweight flame detection method and system based on YOLOv8

Abstract

The invention belongs to the technical field of fire detection, and particularly relates to a YOLOv8-based lightweight flame detection method and system, and the method comprises the steps: obtaining a to-be-detected flame image; constructing a YOLOv8 backbone network on the basis of the alternately arranged convolutional layers and a Ghost Bottleneck Hybrid module with different synchronization lengths, and extracting multi-scale features of the obtained flame image on the basis of the constructed backbone network; a YOLOv8 neck network of a lightweight Slim-Neck architecture is adopted, and multi-scale features of the obtained flame images of different levels are fused; and according to the obtained fused flame image multi-scale features and the YOLOv8 head network, carrying out classified detection on flames in the flame image, and completing YOLOv8-based lightweight flame detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of fire detection, and particularly relates to a lightweight flame detection method and system based on YOLOv8. Background Art

[0002] The statements in this part merely provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] In recent years, object detection algorithms based on deep learning have made remarkable progress in the field of flame detection. Among them, the YOLO series of algorithms have become a research hotspot in this field due to their advantages such as fast detection speed and high accuracy. However, mainstream models such as YOLOv8 usually have a large number of parameters and high computational complexity, and rely on high-performance GPUs for inference, making it difficult to be directly deployed on embedded devices with limited computing resources, which severely restricts their practical applications in edge computing scenarios such as industrial monitoring and forest fire prevention. Although existing research has tried to perform lightweight improvements on the YOLO model through means such as model pruning and knowledge distillation, it is often difficult to achieve a good balance between the model compression rate and detection accuracy, and cannot meet the dual requirements of real-time performance and accuracy for flame detection tasks.

[0004] During the flame detection process, flame targets exhibit high diversity in terms of morphology, scale, and color features. Their shapes are irregular and dynamically changing, the scale spans significantly from nearby open flames to distant smoke, and the color features are easily disturbed by factors such as environmental illumination and smoke concentration. In addition, flame targets are often confused with similar color regions (such as sunset glow and lights) in the background, further increasing the detection difficulty. Existing YOLOv models have limitations in feature extraction capabilities and multi-scale feature fusion mechanisms, making it difficult to effectively cope with the complex characteristics of flame targets, resulting in insufficient detection accuracy and robustness. Summary of the Invention

[0005] To solve the above problems, the present invention proposes a lightweight flame detection method and system based on YOLOv8. Taking YOLOv8 as the benchmark model, the backbone part is constructed based on the lightweight feature extraction module GBH, the SimAM parameter-free attention mechanism is introduced, and the Slim-neck is improved through GSConv to complete the construction of the lightweight flame detection model. Based on the constructed lightweight flame detection model, accurate recognition of flame targets with different distances and different morphologies in complex scenes is carried out.

[0006] According to some embodiments, the first aspect of the present invention provides a lightweight flame detection method based on YOLOv8, adopting the following technical solutions:

[0007] A lightweight flame detection method based on YOLOv8 includes:

[0008] Obtain the flame image to be detected;

[0009] Construct the YOLOv8 backbone network based on alternately arranged convolutional layers and Ghost Bottleneck Hybrid modules with different strides, and extract multi-scale features of the obtained flame image based on the constructed backbone network;

[0010] Adopt the YOLOv8 neck network with a lightweight Slim-Neck architecture to fuse the multi-scale features of the flame images at different levels obtained;

[0011] According to the fused multi-scale features of the flame image and the YOLOv8 head network obtained, perform classification detection of the flame in the flame image to complete lightweight flame detection based on YOLOv8.

[0012] As a further technical limitation, in the YOLOv8 neck network with a lightweight Slim-Neck architecture, introduce a Slim-Neck structure based on GSConv, use GSConv to replace the standard dense convolution, use VoVGSCSP to replace the C3 module, and perform fusion and enhancement of multi-scale features through lightweight design.

[0013] As a further technical limitation, in the process of extracting multi-scale features of the obtained flame image, extract flame image features based on depth convolution, expand the number of channels through a pointwise convolutional layer, and capture diverse features of the flame image; adopt the SimAM parameter-free attention mechanism for adaptive weighting of the extracted flame image features to enhance the robustness of the backbone network.

[0014] As a further technical limitation, perform multi-scale convolution on the fused multi-scale features of the flame image, combine depthwise separable convolution kernels and point convolution to extract semantic information of the deep network, combine with the detection head of YOLOv8 to generate the target bounding box and class probability of the flame image, identify the position and class of the flame, and complete lightweight flame detection based on YOLOv8.

[0015] As a further technical limitation, the YOLOv8 neck network with a lightweight Slim-Neck architecture is a Slim-neck architecture improved based on GSConv, that is, GSConv combines grouped convolution and depthwise separable convolution in a cascaded manner to reduce the computational complexity while maintaining the channel interaction ability.

[0016] As a further technical limitation, the Ghost Bottleneck Hybrid module with different strides extracts local details and global context information of the image through multi-scale convolution, and gradually reduces the resolution by combining downsampling with different strides to obtain feature maps of different scales of the flame image.

[0017] According to some embodiments, the second solution of the present invention provides a lightweight flame detection system based on YOLOv8, adopting the following technical solution:

[0018] A lightweight flame detection system based on YOLOv8, comprising:

[0019] An acquisition module configured to acquire a flame image to be detected;

[0020] An extraction module configured to construct a YOLOv8 backbone network based on alternately arranged convolutional layers and Ghost Bottleneck Hybrid modules with different strides, and extract multi-scale features of the acquired flame image based on the constructed backbone network;

[0021] A fusion module configured to adopt a YOLOv8 neck network with a lightweight Slim-Neck architecture to fuse the multi-scale features of the acquired flame images at different levels;

[0022] A detection module configured to perform classification detection of flames in the flame image according to the obtained fused multi-scale features of the flame image and the YOLOv8 head network, and complete lightweight flame detection based on YOLOv8.

[0023] According to some embodiments, the third solution of the present invention provides a computer-readable storage medium, adopting the following technical solution:

[0024] A computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, it implements the steps in a lightweight flame detection method based on YOLOv8 as described in the first solution of the present invention.

[0025] According to some embodiments, the fourth solution of the present invention provides an electronic device, adopting the following technical solution:

[0026] An electronic device, comprising a memory, a processor, and a program stored on the memory and running on the processor, and when the processor executes the program, it implements the steps in a lightweight flame detection method based on YOLOv8 as described in the first solution of the present invention.

[0027] According to some embodiments, the fifth solution of the present invention provides a computer program product, adopting the following technical solution:

[0028] A computer program product, comprising software code, and the program in the software code executes the steps in a lightweight flame detection method based on YOLOv8 as described in the first solution of the present invention.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0030] Based on YOLOv8 as the benchmark model, the present invention constructs the backbone part based on the lightweight feature extraction module GBH, introduces the parameter-free attention mechanism SimAM, improves Slim-neck through GSConv, completes the construction of the lightweight flame detection model, and accurately identifies flame targets at different distances and in different forms in complex scenarios based on the constructed lightweight flame detection model, realizing high-precision and high-efficiency flame detection and promoting the wide application of flame detection technology in actual scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings forming a part of this embodiment are used to provide a further understanding of this embodiment. The schematic embodiments and descriptions thereof of this embodiment are used to explain this embodiment and do not constitute an improper limitation to this embodiment.

[0032] Figure 1 It is a flowchart of a lightweight flame detection method based on YOLOv8 in Embodiment 1 of the present invention;

[0033] Figure 2 It is a network architecture diagram of YOLOv8-GBH in Embodiment 1 of the present invention;

[0034] Figure 3 It is a structural diagram of the feature extraction module GhostBottleneckHybrid with a stride of 1 in Embodiment 1 of the present invention;

[0035] Figure 4 It is a structural diagram of the feature extraction module Ghost Bottleneck Hybrid with a stride of 2 in Embodiment 1 of the present invention;

[0036] Figure 5 It is a structural diagram of the GhostHybridConv module in Embodiment 1 of the present invention;

[0037] Figure 6 It is a schematic diagram of the relationship between precision and confidence during the flame detection process in Embodiment 1 of the present invention;

[0038] Figure 7 It is a schematic diagram of the relationship between accuracy and recall during the flame detection process in Embodiment 1 of the present invention;

[0039] Figure 8 It is a schematic diagram of the relationship between recall and confidence during the flame detection process in Embodiment 1 of the present invention;

[0040] Figure 9 It is an effect diagram of the flame detection in Embodiment 1 of the present invention;

[0041] Figure 10 This is the structural block diagram of a lightweight flame detection system based on YOLOv8 in the second embodiment of the present invention. Specific implementation mode

[0042] The present invention will be further described below in conjunction with the drawings and embodiments.

[0043] It should be noted that the following detailed descriptions are all exemplary and are intended to provide further descriptions of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present application belongs.

[0044] It should be noted that the terms used herein are only for describing specific implementation modes and are not intended to limit the exemplary implementation modes according to the present invention. As used herein, unless otherwise clearly specified in the context, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0045] In the present invention, terms such as "upper", "lower", "left", "right", "front", "rear", "vertical", "horizontal", "side", "bottom", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings. They are only relationship terms determined for the convenience of describing the structural relationship of each component or element of the present invention and do not specifically refer to any component or element of the present invention. It should not be construed as a limitation of the present invention.

[0046] In the present invention, terms such as "fixed connection", "connected", "connected" should be understood in a broad sense, indicating that it can be a fixed connection, an integral connection or a detachable connection; it can be directly connected or indirectly connected through an intermediate medium. For those skilled in the relevant scientific research or technology in this field, the specific meaning of the above terms in the present invention can be determined according to specific circumstances and should not be construed as a limitation of the present invention.

[0047] Without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0048] Embodiment 1

[0049] Embodiment 1 of the present invention introduces a lightweight flame detection method based on YOLOv8.

[0050] As Figure 1 shown, a lightweight flame detection method based on YOLOv8 includes:

[0051] Obtain the flame image to be detected;

[0052] Construct the YOLOv8 backbone network based on alternately arranged convolutional layers and Ghost Bottleneck Hybrid modules with different strides, and extract multi-scale features of the acquired flame images based on the constructed backbone network;

[0053] Adopt the YOLOv8 neck network with a lightweight Slim-Neck architecture to fuse the multi-scale features of the acquired flame images at different levels;

[0054] Based on the obtained fused multi-scale features of the flame images and the YOLOv8 head network, perform classification detection of the flame in the flame images to complete lightweight flame detection based on YOLOv8.

[0055] In this embodiment, when YOLOv8-GBH processes the input image, the image resolution is required to be 640x640 and in the RGB color space, that is, each pixel consists of three channels: red (R), green (G), and blue (B). To maintain the stability of model processing, the aspect ratio of the input image should be as close to 1:1 as possible; if the aspect ratio is inconsistent, YOLOv8-GBH will automatically perform padding to ensure that the original ratio of the image is not damaged and at the same time avoid target deformation or information loss. The YOLOv8-GBH flame detection algorithm supports various common RGB image formats, such as JPEG, PNG, BMP, etc., which can ensure the integrity and compatibility of image data and are suitable for different application scenarios.

[0056] Aiming at the problems of the YOLOv8 in flame detection applications, such as a relatively deep network architecture, a large number of parameters, and being unfavorable for deployment on edge devices, this embodiment adopts the Figure 2 YOLOv8-GBH network structure shown, including a backbone network (Backbone), a neck network (Neck), and a head network (Head); among them, the backbone network is responsible for extracting rich feature representations from the input image and generating high-level feature maps through multiple convolutional operations; the neck network further fuses features at different levels on this basis, uses the Path Aggregation Network (PAN) structure to integrate multi-scale information, and enhances the model's detection ability for targets of different sizes; the head network then performs final target classification and bounding box regression based on these fused feature maps and directly outputs the detection results; enabling YOLOv8 to efficiently and accurately complete the target detection task.

[0057] In the backbone network, this embodiment introduces the GhostBottleneckHybrid (GBH) module. The network structure alternately arranges standard convolutional layers (Conv) and GBH modules with two different strides, achieving effective extraction of multi-scale flame features; it not only maintains high-efficient feature extraction ability but also significantly reduces computational complexity and resource consumption.

[0058] In the YOLOv8 neck network, this embodiment introduces the Slim-Neck structure based on GSConv, achieving efficient multi-scale feature fusion and enhancement through lightweight design. In Slim-Neck, GSConv is used to replace the standard dense convolution, and VoVGSCSP replaces the C3 module; the GSConv module is used for downsampling operations to ensure effective transfer of feature maps between different scales, while maintaining high-efficient feature extraction and significantly reducing the number of parameters and computational costs; the VoVGSCSP module further processes the fused feature maps by combining Ghost convolution and CSPNet to enrich the feature representation.

[0059] As Figure 3 shown, for the feature extraction module Ghost Bottleneck Hybrid with a stride of 1, the left branch is the feature extraction part, and the right branch designs an identity mapping, enhancing the stability of network training and retaining the information of the original input through residual connection; it effectively alleviates the vanishing gradient problem in deep networks, promotes the direct flow of information, and enables the model to more stably learn fine-grained features.

[0060] When performing feature extraction, in this embodiment, deep convolution is used to extract features, and a pointwise convolution layer is used to expand the number of channels, making the feature representation more diverse and rich, which helps to capture more complex feature patterns and enhances the learning ability and representation performance of the model. Combined with the SimAM attention mechanism, the feature map is adaptively weighted. SimAM enhances the importance of important features in a simple and efficient way, improving the representation ability of the model. This combination not only enriches the feature expression but also ensures that the model can more accurately focus on key information, thereby improving the overall detection and classification performance. The cross-stage local operation is introduced. After the Channel Split operation, an equal division operation is performed on the feature map in the channel dimension. One part is directly passed to the subsequent layer, and the other part uses deep convolution to further refine the feature expression, aiming to improve the diversity of the feature expression. At the same time, the computational burden is reduced through lightweight operations. After splicing and merging, a 1x1 efficient convolution layer is used to restore the number of channels, ensuring the compactness and efficiency of the output feature map and avoiding unnecessary redundant information. The design of the Inverted Residual structure promotes more non-linear transformations and enhances the model's learning ability for features of different scales. It enables the network to achieve more complex feature mappings at a lower computational cost, improving the flexibility and expressiveness of the model. The result after feature fusion is added to the output on the shortcut path to obtain the final output feature map of the module. This structure not only promotes the implementation of non-linear transformations but also enhances the richness of the feature expression, improving the representation ability and generalization performance of the model.

[0061] As Figure 4 shown, the feature extraction module with a stride of 2 is shown in the figure. The input tensor passes through the GhostHybridConv multi-scale convolution layer as Figure 5 shown, generating a more diverse feature map. The GhostHybridConv module significantly improves the efficiency and diversity of feature extraction by combining multi-scale convolution and lightweight Ghost convolution techniques. It can achieve rich feature representations with lower computational complexity and resource consumption. By introducing two different sizes of deep convolution kernels, 3x3 and 5x5, local and global spatial information is captured in a single layer, enhancing the ability to detect flame targets of different scales.

[0062] The feature map after multi-scale feature extraction is then downsampled with a stride of 2 through depthwise separable convolution. This process reduces the spatial dimension while retaining rich channel information. Pointwise convolution is used to fuse the features after depth convolution, increasing the information exchange between different channels. This feature map is added to the identity mapping to obtain the final output feature map.

[0063] It should be noted that the input image in this embodiment first enters the backbone network. In the original YOLOv8, the image undergoes feature extraction and downsampling through a series of standard convolutional layers (Conv) and CSP (Cross Stage Partial) modules. In the improved algorithm, the core module of the backbone network is replaced by the GhostBottleneckHybrid module, which significantly optimizes the computational efficiency and feature extraction ability through designs such as depthwise separable convolution, point convolution, and multi-scale convolution. Specifically, the input image first undergoes initial feature extraction through a standard convolutional layer (Conv) to generate an initial feature map, and then is alternately processed by GhostBottleneckHybrid modules with a stride of 1 and a stride of 2. Among them, the module with a stride of 1 extracts spatial features through depthwise separable convolution, adjusts the number of channels through point convolution, combines the Split Path structure to extract multi-scale features, and simultaneously introduces the SimAM attention mechanism to enhance important features, which is particularly suitable for capturing the detailed information of flame targets in the shallow network. The GhostBottleneckHybrid module with a stride of 2 is used for downsampling and deep feature extraction. By introducing multi-scale convolutions (such as 3x3 and 5x5 convolutions) through the GhostHybridConv module, combined with depthwise separable convolution and point convolution, it significantly reduces the computational amount while improving the adaptability to flame targets and can better extract semantic information in the deep network. Compared with the CSP module of the original YOLOv8, the improved backbone network structure significantly reduces the number of parameters and computational complexity while maintaining a high feature extraction ability, especially suitable for deployment on resource-constrained devices.

[0064] Then, it enters the neck part. The improved model has been optimized specifically compared to YOLOv8. GSConv is used to replace the standard dense convolution, and VoVGSCSP replaces the C3 module. GSConv reduces the computational amount and the number of parameters through the combination of grouped convolution and depthwise separable convolution, while VoVGSCSP optimizes feature fusion through a one-time aggregation method, maintaining the accuracy while reducing the computational complexity.

[0065] Finally, it is input into the detection head (Detect) to generate the bounding boxes and class probabilities of flame targets. The detection head outputs the position and class information of the flame through the anchor box mechanism and the classifier.

[0066] Case analysis

[0067] Based on the flame detection method proposed in this embodiment, a certain flame image to be detected is obtained. After obtaining the image, preprocessing and data augmentation of the flame image are carried out. Subsequently, after manual annotation, the obtained flame images are divided into a training set, a validation set, and a test set; a lightweight flame detection model is constructed based on the training set, and the constructed lightweight flame detection model is tested and verified using the test set and the validation set respectively, and finally flame detection is completed.

[0068] In this embodiment, YOLOv8 is used as the benchmark model, and a lightweight improvement scheme is proposed for the flame detection task. The original YOLOv8 backbone network adopts a feature extraction architecture composed of C2f modules and standard convolutions; to optimize the model efficiency, a new GhostBottleneckHybrid module is used to replace the original feature extraction module, that is, depthwise separable convolutions and pointwise convolutions are used, which will significantly reduce the computational load.

[0069] For a standard convolution, assuming the convolution kernel size is D k ×D k , the number of input channels is M, the number of output channels is N, and the size of the output feature map is D F ×D F , then after passing through the standard convolution, the number of parameters can be calculated as D k ×D k ×M×N, and the amount of computation is D k ×D k ×M×N×D F ×D F . The depthwise separable convolution includes a depthwise convolution and a pointwise convolution. The convolution kernel size of the depthwise convolution is D k ×D k ×1, the number of convolution kernels is M, and each one has to do D F ×D F times of multiplication and addition operations, so the amount of computation is D k ×D k ×M×D F ×D F , the number of parameters is D k ×D k ×M, the convolution kernel size of the pointwise convolution is 1×1×M, the number of convolution kernels is N, and each one has to do D F ×D F times of multiplication and addition operations, so the amount of computation is M×M×D F ×D F , the number of parameters is M×N. Therefore, the amount of computation of the depthwise separable convolution is D k ×D k ×M×D F ×D F +M×M×D F ×D F, with the number of parameters being D k ×D k ×M + M×N. Parameter ratio: Computation ratio: It can be seen from this that when the convolution kernel is the commonly used 3*3 convolution, the number of parameters and the amount of computation using depthwise separable convolution are reduced to about one-ninth of the original.

[0070] In the Neck part, a Slim-neck architecture improved based on GSConv is adopted; GSConv reduces the computational complexity while maintaining the channel interaction ability through the cascaded combination of grouped convolution and depthwise separable convolution. Specifically, the CSP module in the original Neck is replaced with the VoVGSCSP module, which optimizes the information fusion path through cross-stage feature aggregation. Experiments show that this improvement reduces the floating-point operation amount (FLOPs) of the Neck part by an average of 15.72%.

[0071] Combined with the collaborative optimization of Backbone and Neck, the final model has achieved significant improvements in various performance indicators, and the detection results are shown in Table 1.

[0072] Table 1 Detection Results

[0073]

[0074] As Figure 6 、 Figure 7 and Figure 8 shown, the improved model in this embodiment has achieved significant improvements in various performance indicators: the accuracy (Precision) reaches 93.4%, the recall rate (Recall) is 92.7%, and mAP@0.5 is increased to 93.8%. In terms of model lightweighting, compared with the baseline model YOLOv8n, the number of parameters is reduced by about 68.4%, and the floating-point operation amount is reduced by 50%, greatly improving the deployment feasibility of the model on edge devices.

[0075] It should be noted that despite significant lightweighting processing, the detection accuracy of the model still remains at a high level. The average accuracy of the improved lightweight model is only 0.8% lower than that of the original YOLOv8n model, and the detection performance is basically the same as that of the original model. This achievement fully demonstrates the effectiveness of the proposed architecture in achieving model lightweighting while maintaining detection performance.

[0076] In this embodiment, YOLOv8 is used as the benchmark model. The backbone part is constructed based on the lightweight feature extraction module GBH. The SimAM parameter-free attention mechanism is introduced, and the Slim-neck is improved through GSConv to complete the construction of the lightweight flame detection model. Based on the constructed lightweight flame detection model, accurate recognition of flame targets with different distances and forms in complex scenarios is carried out, realizing high-precision and high-efficiency flame detection, and promoting the wide application of flame detection technology in actual scenarios.

[0077] Benefiting from a smaller model size and lower computing requirements, the application scenarios of this lightweight flame detection model have been significantly expanded. It can be easily embedded in mobile devices or embedded systems and is suitable for a wider range of environmental and task requirements. Whether it is a portable device with limited resources or an industrial monitoring scenario with strict efficiency requirements, this model can meet different resource constraints and performance requirements, demonstrating extremely high flexibility and adaptability. This wide applicability not only broadens the application boundaries of flame detection technology but also provides new solutions for the fields of edge computing and the Internet of Things.

[0078] Embodiment 2

[0079] Embodiment 2 of the present invention introduces a lightweight flame detection system based on YOLOv8.

[0080] As Figure 10 shown, a lightweight flame detection system based on YOLOv8 includes:

[0081] An acquisition module configured to acquire a flame image to be detected;

[0082] An extraction module configured to construct a YOLOv8 backbone network based on alternately arranged convolutional layers and Ghost Bottleneck Hybrid modules with different strides, and extract multi-scale features of the acquired flame image based on the constructed backbone network;

[0083] A fusion module configured to adopt a YOLOv8 neck network with a lightweight Slim-Neck architecture to fuse the multi-scale features of the acquired flame images at different levels;

[0084] A detection module configured to perform classification detection of flames in the flame image according to the obtained multi-scale features of the fused flame image and the YOLOv8 head network, and complete lightweight flame detection based on YOLOv8.

[0085] The detailed steps are the same as those of a lightweight flame detection method based on YOLOv8 provided in Embodiment 1 and will not be elaborated here.

[0086] Embodiment 3

[0087] Embodiment 3 of the present invention provides a computer-readable storage medium.

[0088] A computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, it implements the steps in a lightweight flame detection method based on YOLOv8 as described in Embodiment 1 of the present invention.

[0089] The detailed steps are the same as those of the lightweight flame detection method based on YOLOv8 provided in Embodiment 1, and will not be elaborated here.

[0090] Embodiment 4

[0091] Embodiment 4 of the present invention provides an electronic device.

[0092] An electronic device includes a memory, a processor, and a program stored on the memory and running on the processor. When the processor executes the program, it implements the steps in a lightweight flame detection method based on YOLOv8 as described in Embodiment 1 of the present invention.

[0093] The detailed steps are the same as those of the lightweight flame detection method based on YOLOv8 provided in Embodiment 1, and will not be elaborated here.

[0094] Embodiment 5

[0095] Embodiment 5 of the present invention provides a computer program product.

[0096] A computer program product includes software code, and the program in the software code executes the steps in a lightweight flame detection method based on YOLOv8 as described in Embodiment 1 of the present invention.

[0097] The detailed steps are the same as those of the lightweight flame detection method based on YOLOv8 provided in Embodiment 1, and will not be elaborated here.

[0098] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can be in the form of a computer program product implemented 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. The solutions in the embodiments of the present invention can be implemented in various computer languages, for example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript.

[0099] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0100] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0101] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, such that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0102] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0103] Obviously, those skilled in the art can make various changes and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

[0104] The above are only the preferred embodiments of the present embodiment and are not used to limit the present embodiment. For those skilled in the art, the present embodiment can have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present embodiment shall be included within the protection scope of the present embodiment.

Claims

1. A lightweight flame detection method based on YOLOv8, characterized in that, Including: Obtain the flame image to be detected; Construct a YOLOv8 backbone network based on alternately arranged convolutional layers and Ghost Bottleneck Hybrid modules with different strides, and extract multi-scale features of the obtained flame image based on the constructed backbone network; Adopt a YOLOv8 neck network with a lightweight Slim-Neck architecture to fuse the multi-scale features of the obtained flame images at different levels; According to the fused multi-scale features of the flame image and the YOLOv8 head network obtained, perform classification detection of the flame in the flame image, and complete lightweight flame detection based on YOLOv8.

2. The lightweight flame detection method based on YOLOv8 as described in claim 1, wherein, In the YOLOv8 neck network with the lightweight Slim-Neck architecture, introduce a Slim-Neck structure based on GSConv, use GSConv to replace the standard dense convolution, use VoVGSCSP to replace the C3 module, and perform fusion and enhancement of multi-scale features through lightweight design.

3. The lightweight flame detection method based on YOLOv8 according to claim 1, characterized in that, During the process of extracting multi-scale features of the obtained flame image, extract flame image features based on depth convolution, expand the number of channels through pointwise convolutional layers, and capture diverse features of the flame image; adopt the SimAM parameter-free attention mechanism for adaptive weighting of the extracted flame image features to enhance the robustness of the backbone network.

4. A lightweight flame detection method based on YOLOv8 as described in claim 1, characterized in that, Perform multi-scale convolution on the fused multi-scale features of the flame image, combine depthwise separable convolution kernels and point convolution to extract semantic information of the deep network, combine with the detection head of YOLOv8 to generate the target bounding box and class probability of the flame image, identify the position and class of the flame, and complete lightweight flame detection based on YOLOv8.

5. A lightweight flame detection method based on YOLOv8 as described in claim 1, characterized in that, The YOLOv8 neck network with the lightweight Slim-Neck architecture adopts a Slim-neck architecture improved based on GSConv, that is, GSConv combines grouped convolution and depthwise separable convolution in series to reduce the computational complexity while maintaining the channel interaction ability.

6. A lightweight flame detection method based on YOLOv8 as described in claim 1, characterized in that, The Ghost Bottleneck Hybrid module with different strides extracts local details and global context information of the image through multi-scale convolution, and gradually reduces the resolution by combining downsampling with different strides to obtain feature maps of different scales of the flame image.

7. A lightweight flame detection system based on YOLOv8, characterized in that, Including: An acquisition module configured to obtain the flame image to be detected; An extraction module configured to construct a YOLOv8 backbone network based on alternately arranged convolutional layers and Ghost Bottleneck Hybrid modules with different strides, and extract multi-scale features of the obtained flame image based on the constructed backbone network; A fusion module configured to adopt a YOLOv8 neck network with a lightweight Slim-Neck architecture to fuse the multi-scale features of the obtained flame images at different levels; A detection module configured to perform classification detection of the flame in the flame image according to the fused multi-scale features of the flame image and the YOLOv8 head network obtained, and complete lightweight flame detection based on YOLOv8.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the steps of a lightweight flame detection method based on YOLOv8 as described in any one of claims 1-6.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the program, it implements the steps of a lightweight flame detection method based on YOLOv8 as described in any one of claims 1-6.

10. A computer program product comprising software code, characterized in that, The program in the software code executes the steps of a lightweight flame detection method based on YOLOv8 as described in any one of claims 1-6.

Images