Improved fire detection algorithm based on YOLOv8 in combination with SCConv and C2f

Abstract

The invention discloses an improved fire detection algorithm (yov8-C2f-SCConv) based on YOLOv8 in combination with SCConv and C2f, the algorithm uses space and channel redundancy between features to carry out CNN (Convolutional Neural Networks) compression, and an efficient convolution module named SCConv (space and channel reconstruction convolution) is provided to reduce redundant calculation and promote representative feature learning. The SCConv provided by the invention comprises two units, namely a space reconstruction unit (SRU) and a channel reconstruction unit (CRU). The SRU uses a separation-reconstruction method to suppress space redundancy, and the CRU uses a split-conversion-fusion strategy to reduce channel redundancy. Experimental results show that the model embedded with SCConv can significantly reduce complexity and calculation cost by reducing redundant features, the fire detection precision is improved by 12% compared with that of a common YOLOv8 network, and a new technical solution is provided for fire monitoring and early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing and machine learning, and in particular to an improved fire detection algorithm based on YOLOv8 combined with spatial and channel reconstruction convolution (SCConv) and C2f strategy. Background Art

[0002] This algorithm is based on YOLOv8 and combines SCConv and C2f to improve the accuracy of forest fire detection. This algorithm is specifically designed to address the challenge of detecting small targets in fires.

[0003] Patent document CN202211419841.7 discloses a method for detecting infrared small targets based on an improved Yolov5 network. This method uses the neck structure (Neck) in the Yolov5 network for feature fusion to obtain a predicted feature map, which is then predicted by the Yolov5 network's Prediction component. This method addresses the class imbalance between small targets and backgrounds, thereby simplifying the task of detecting infrared small targets. This invention improves YOLOv8 by combining SCConv and the C2f strategy with YOLOv8. By integrating the efficient convolution module of spatial and channel reconstruction convolution (SCConv) and the C2f strategy, the invention aims to enhance the global and local analysis capabilities of the feature extractor, improve the recognition performance of small-scale targets, and enhance the performance of drone monitoring systems.

[0004] While existing deep learning-based methods have made progress in global feature extraction, they are still limited by the local nature of convolutional networks and struggle to capture broad dependencies, which can lead to false alarms. However, this algorithm, which combines SCConv and C2f, improves the accuracy of the model and can effectively avoid false alarms. Summary of the Invention

[0005] The purpose of this invention is to provide an improved YOLOv8 algorithm by integrating SCConv and C2f strategies to improve the accuracy and efficiency of fire detection.

[0006] In order to achieve the above object, the present invention provides the following technical solutions: An improved fire detection algorithm based on YOLOv8 combined with SCConv and C2f is proposed. It consists of two units, namely the spatial reconstruction unit (SRU) and the channel reconstruction unit (CRU). By arranging the SRU and CRU in sequence, SCConv not only reduces the model parameters and floating-point operations (FLOPS), but also enhances the feature expression capability by reducing the spatial and channel redundancy in the feature map.

[0007] Preferably, the SCConv module includes a spatial reconstruction unit (SRU) and a channel reconstruction unit (CRU) to reduce spatial and channel redundancy in the feature map respectively.

[0008] Preferably, the C2f strategy introduces an attention mechanism into the Feature Pyramid Network (FPN) to enhance the feature expression capability of complex fire scenes.

[0009] Preferably, the SRU unit suppresses spatial redundancy by performing separation and reconstruction operations. The SRU first separates the input feature map into information-rich features and information-less features, and then combines the information-rich features with the information-less features through a reconstruction operation to generate a feature map with richer information.

[0010] Preferably, the CRU unit adopts a split-transform-fusion strategy to reduce redundancy in the channel dimension. The CRU first divides the spatially refined feature map into two parts, applies a different transformation strategy to each part, and finally merges these feature maps through a fusion strategy to generate the final channel-refined feature.

[0011] In the above technical solution, the improved YOLOv8 algorithm provided by the present invention has the following beneficial effects: 1. This invention, by introducing the spatial reconstruction unit (SRU) and channel reconstruction unit (CRU), the SCConv module significantly reduces model parameters and floating-point operations (FLOPs), making the model more lightweight and efficient.

[0012] 2. This invention enhances feature expression by reducing spatial and channel redundancy in feature maps, thereby improving the accuracy and robustness of fire detection. This means the algorithm has greater adaptability and generalization capabilities in complex fire scenarios, enabling more accurate identification and location of fire targets, helping to improve disaster response and rescue efficiency.

[0013] 3. This invention utilizes the C2f strategy to improve feature fusion efficiency, enabling the model to better utilize feature information at different scales, enhancing its perception and understanding of complex fire scenarios. This feature fusion mechanism helps reduce information loss and fuzziness, improving the accuracy and stability of fire detection.

[0014] 4. This invention uses the RIOU loss function to optimize bounding box prediction, which is particularly effective in detecting small objects in high-resolution images. This optimization effectively addresses the challenge of detecting small objects in fire scenes, improving the algorithm's ability to detect fire objects of various sizes and shapes, and further improving the reliability and comprehensiveness of fire monitoring systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a fire detection algorithm based on YOLOv8 combined with SCConv and C2f in one embodiment of the present application; Figure 2 This is a diagram of the SCConv architecture that combines a spatial reconstruction unit (SRU) and a channel reconstruction unit (CRU) in another embodiment of the present application; Figure 3 This is an architectural diagram of a spatial reconstruction unit (SRU) used in another embodiment of the present application; Figure 4 This is an architectural diagram of a channel reconstruction unit (CRU) used in another embodiment of the present application; Reference numerals: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] Figure 1 This is a flow chart of a fire detection algorithm based on YOLOv8 combined with SCConv and C2f in one embodiment of the present application. Figure 1 As shown, the fire detection method based on YOLOv8 combined with SCConv and C2f strategy in this embodiment includes the following steps: S1. Use SRU and CRU to reduce the spatial and channel redundancy in the feature map to obtain a fused feature map; S2. Perform fire detection based on the fused feature map to obtain a fire detection result.

[0018] The fire detection method of this embodiment can be flexibly applied to various computer devices, including laptops, desktop computers, and servers. These devices can execute the method by installing corresponding software systems. Furthermore, the specific type of computer device is not particularly limited, thus providing a wide range of applicability.

[0019] The fire detection method of this embodiment can be executed separately on the client device or the server device, or through the cooperation of the two. The server can be a separate server or a cloud system built by multiple servers. For example, if the fire detection method is integrated into the client, when the client device receives a small target detection request, it can use its own hardware environment to perform the detection operation. This method is suitable for scenarios where target detection needs to be performed locally, such as in an edge computing environment or for applications with high requirements for privacy protection. In addition, if the small target detection method is integrated into the server device, when the server device receives a detection request, it can use its own hardware environment to execute the detection method. This method is suitable for scenarios where a large number of detection requests need to be processed centrally or where large-scale data processing needs to be performed in the cloud.

[0020] Figure 2 In another embodiment of the present application, the SCConv architecture diagram using a combination of a spatial reconstruction unit (SRU) and a channel reconstruction unit (CRU) is constructed as follows: The spatial reconstruction unit (SRU) and channel reconstruction unit (CRU) are placed in sequence. Specifically, for the intermediate input feature X in the bottleneck residual block, the spatial refined feature X is first obtained by the SRU operation. w , and then the CRU operation is used to obtain the channel-refined feature Y. The spatial and channel redundancy between features is exploited in the SCConv module, which can be seamlessly integrated into any CNN architecture to reduce the redundancy between intermediate feature maps and improve the feature expression of CNN.

[0021] Figure 3 This is an architecture diagram of the spatial reconstruction unit (SRU) used in another embodiment of the present application, which utilizes separation and reconstruction operations. The separation operation aims to separate feature maps that are rich in information related to spatial content from feature maps with less information. The information richness of different feature maps is evaluated by utilizing the scaling factor in the group normalization (GN) layer. Specifically, given an intermediate feature map , where N is the batch axis, C is the channel axis, H and W are the spatial height and width axes respectively. First, the input feature X is normalized by subtracting the mean µ and dividing by the standard deviation σ, as follows: Where µ and σ are the mean and standard deviation in X, ε is a decimal added for the stability of division, and γ and β are trainable affine transformation parameters. γ ∈R C To measure the spatial pixel variance of each batch and channel. The normalized correlation weight W is rich in spatial information. γ ∈R Cis obtained by formula (2), which indicates the importance of different feature maps.

[0022]

[0023] The feature map reweighted by the weight γ is mapped to the interval (0, 1) by the Sigmoid function and gated by a threshold. The weights above the threshold are set to 1, resulting in a weight with rich information content W1, while the weights below the threshold are set to 0, resulting in a weight with less information content W2. The entire process of obtaining W is given by formula (3).

[0024]

[0025] To reduce spatial redundancy, this paper proposes a reconstruction operation that adds information-rich features to less informative features to generate more informative and space-efficient features. Rather than simply adding these two components, this approach uses a cross-reconstruction operation to fully combine the two weighted features with different information content and enhance the information flow between them. Finally, the cross-reconstructed features are concatenated to produce a spatially refined feature map.

[0026] Figure 4 This is an architecture diagram of a channel reconstruction unit (CRU) used in another embodiment of the present application. The present invention replaces the standard convolution with the CRU. The method includes the following steps: S4, Split, in the processing space refinement feature X w When , its channels are first split into two parts according to the ratio α, with αC and (1−α)C channels respectively. This ratio α is between 0 and 1. Subsequently, the number of channels of the feature map is further compressed by 1x1 convolution to enhance computational efficiency. The present invention adopts a compression ratio r to adjust the number of feature channels to balance the computational cost of the channel reconstruction unit (CRU). After the segmentation and compression operations, the spatially refined feature X w is subdivided into the upper part X up and the lower part X low .

[0027] S5. Transformation: In the channel reconstruction unit (CRU), the present invention transforms the input feature X up First, the upper layer is transformed. Efficient convolution operations - group convolution (GWC) and point-by-point convolution (PWC) are used to replace the standard k×k convolution to extract high-level representative information and reduce computational costs. The upper layer transform extracts rich features Y1 by combining GWC and PWC while reducing computational costs. In the lower layer transform stage, X lowThe shallow detail feature map is generated by 1×1 PWC transformation to enrich the feature extraction. In addition, the previous feature X is reused. low To obtain more feature maps. Finally, the generated and reused features are concatenated to form the output Y2 of the next stage.

[0028] S6, Fusion (Fuse), the present invention adopts a simplified SKNet method to adaptively fuse the output features Y1 and Y2 of the upper and lower transformation stages. First, a global average pooling operation is applied to collect global spatial information and calculate the channel-level statistics S m Next, the global channel descriptors S1 and S2 of the upper and lower layers are stacked, and a channel-level soft attention operation is used to generate the feature importance vector , Finally, under the guidance of the feature importance vector, the upper-layer feature Y1 and the lower-layer feature Y2 are fused in a channel-level manner to obtain the channel-refined feature Y.

[0029] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments after learning the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0030] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be encompassed therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. An improved fire detection algorithm based on YOLOv8 combined with SCConv and C2f, characterized in that: The method includes: Utilize the spatial and channel redundancy between features for CNN compression, adopt the separation-reconstruction method to suppress spatial redundancy, and use the split-transform-fusion strategy to reduce channel redundancy and promote representative feature learning; Fire detection is performed based on the fused feature map to obtain a fire detection result.

2. The improved fire detection algorithm based on YOLOv8 combined with SCConv and C2f according to claim 1, characterized in that: The method comprises the following steps: S1. Acquire an image to be detected, and preprocess the image to be detected to obtain a preprocessed image to be detected; S2. Input the image to be detected into a pre-trained fire detection model to obtain the corresponding fire detection result; wherein, the fire detection model includes a backbone network module for compressing and refining features, a feature fusion module for fusing features at different levels, a bottleneck residual block for reducing the number of parameters and the amount of calculation, a loss function for optimizing bounding box prediction, an SPPF module for pooling processing, and a detection head module for performing fire detection on the fused feature map.

3. The improved fire detection algorithm based on YOLOv8 combined with SCConv and C2f according to claim 2, characterized in that: The steps of constructing the backbone network module include: The image to be detected is first passed through a standard convolutional layer for preliminary feature extraction. Batch normalization and the SiLU activation function are then used to enhance the performance of the feature extraction layer. Next, a bottleneck layer is introduced to reduce dimensionality and computational complexity while preserving key features. SCConv layers are then constructed, utilizing spatial and channel-wise reconstructive convolutions to reduce redundancy. Finally, multiple C2f SCConv layers are deployed as a composite convolution operation to further compress and refine features.

4. The improved fire detection algorithm based on YOLOv8 combined with SCConv and C2f according to claim 2, characterized in that: The feature fusion methods at different levels include: Features at different levels are fused through upsampling and cascade operations to improve the network's adaptability to targets of different scales, and multiple C2f SCConv layers are used to further optimize the fused features.

5. The improved fire detection algorithm based on YOLOv8 combined with SCConv and C2f according to claim 2, characterized in that: The loss function for optimizing bounding box prediction includes: RIOU loss function.

6. The improved fire detection algorithm based on YOLOv8 combined with SCConv and C2f according to claim 2, characterized in that: The steps of constructing the bottleneck residual block for reducing the number of parameters and the amount of computation include: The image to be detected is subjected to a 1x1 convolution for dimensionality reduction, and then a 3x3 convolution is performed to extract more complex spatial features. Finally, a 1x1 convolution is performed again to increase the dimensionality and restore the original depth of the feature map. At the same time, the input feature map is added to the final output through a direct residual connection to promote the direct flow of gradients and obtain a multi-size feature map.

7. The improved fire detection algorithm based on YOLOv8 combined with SCConv and C2f according to claim 2, characterized in that: Pooling processing method of the SPPF module used for pooling processing: The image to be detected is pooled using a MaxPool operation.

8. The improved fire detection algorithm based on YOLOv8 combined with SCConv and C2f according to claim 2, characterized in that: The method for the detection head module to detect the fused feature map includes: The Feature Pyramid Network (FPN) is used to construct a feature pyramid from low to high levels to obtain high-level features with strong semantic information and retain detailed information.

9. The improved fire detection algorithm based on YOLOv8 combined with SCConv and C2f according to claim 2, characterized in that: After S2, it also includes: S3. Delete the overlapping detection frames to obtain the final fire detection result.

Images