A lightweight method for instance segmentation of lightning images

Abstract

The present invention belongs to the technical field of high-voltage lightning strike fault prevention, and more specifically, relates to a lightweight lightning image instance segmentation method. The method includes: First, obtain lightning images existing in the target area during thunderstorms and clean them, removing images that do not meet the conditions, and the remaining images should have moderate exposure; Then, input the cleaned lightning image data into the YOLO-lightning network model to perform instance segmentation on the lightning image data; Finally, output the contour information of the lightning to be recognized in the image. The present invention solves the problems of insufficient accuracy and limited feature extraction ability in the instance segmentation of complex-shaped targets in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of high-voltage lightning strike fault prevention, and more specifically, relates to a lightweight lightning image instance segmentation method. Background Art

[0002] High-voltage transmission lines are the core infrastructure of the power system, and ensuring their safe and stable operation is crucial for the reliability of the power grid. However, since high-voltage transmission lines are often located in extensive and complex natural environments, they are extremely vulnerable to the threat of lightning disasters. Lightning is recognized as one of the main causes of transmission line faults and tripping. In particular, lightning with complex shapes (such as forked lightning) may cause multi-point grounding, which may in turn lead to simultaneous tripping of multiple circuits. Such faults not only seriously threaten the safety of transmission lines but may also trigger large-scale power outages, affecting the continuous operation of the power grid.

[0003] As transmission lines are gradually extended, especially in complex terrains such as mountains and plateaus, the risk of lightning-induced faults is increasing. This risk not only stems from complex terrains but is also closely related to the lightning protection design of transmission lines. Due to the undulation of the terrain and the change of the lightning protection angle, lightning is prone to form a branched structure, increasing the difficulty and challenge of fault troubleshooting.

[0004] Existing lightning strike fault detection means mainly rely on traditional methods such as video monitoring, lightning location systems, and ground inspections. However, these methods often can only provide macroscopic information about the occurrence of lightning strikes and cannot capture the fine features of lightning, especially the specific morphology of the lightning path and its branches. At the same time, due to the high randomness and complexity of lightning, existing technologies have significant deficiencies in identifying the development path, bifurcation, and deformation of lightning. This makes the work of preventing and coping with lightning disasters more complex and unable to respond promptly and accurately to potential threats.

[0005] The development of deep learning and computer vision technologies has brought new opportunities for the accurate identification of lightning paths. As an advanced computer vision method, instance segmentation technology can perform pixel-level segmentation on each lightning channel in an image, capturing the complex morphology and branched structure of lightning. However, existing instance segmentation methods still face challenges such as insufficient accuracy and weak feature extraction ability when dealing with targets with complex shapes and fine edges like lightning, especially when dealing with the details and complex edges of lightning, the effect is not ideal.

[0006] Therefore, in response to the lightning strike prevention requirements of high-voltage transmission lines, designing an efficient and lightweight lightning instance segmentation model to improve the ability to identify complex lightning paths and branches has become a key technical means to prevent lightning hazards. Such a model can not only improve the accuracy of lightning warning but also provide more targeted technical support for the lightning protection design and operation and maintenance of transmission lines, thus greatly reducing the threat of lightning strikes to the operation of the power grid.

[0007] Chinese patent document CN118096654A discloses a method for identifying road surface diseases based on deep learning. The method includes: obtaining a road surface image to be identified captured for a target road section; inputting the road surface image to be identified into a pre-trained road surface disease instance segmentation model, so that the road surface disease instance segmentation model determines the category information and contour information of the road surface diseases included in the road surface image to be identified. Among them, the road surface disease instance segmentation model is based on the Yolov8-seg model, introduces a convolutional attention mechanism into the c2f module, replaces the SPPF module in the original model with the SPPFCSPC module, and adds an Inner Iou loss function on the basis of the original model; according to the category information and contour information of the road surface diseases, determine the length, width or disease area of the road surface diseases. However, this instance segmentation method still faces problems such as insufficient accuracy and weak feature extraction ability when dealing with targets with complex shapes and fine edges such as lightning.

[0008] In view of this, the present invention designs a lightweight lightning image instance segmentation method to solve the problems existing in the prior art. Summary of the Invention

[0009] The present invention aims to overcome at least one defect of the above-mentioned prior art, and provides a lightweight lightning image instance segmentation method to solve the problems of insufficient accuracy and limited feature extraction ability in the instance segmentation of targets with complex shapes in the prior art.

[0010] The detailed technical solution of the present invention is as follows:

[0011] A lightweight lightning image instance segmentation method, the method includes:

[0012] S1. First, obtain lightning images existing in the target area during thunderstorms, and perform cleaning to remove images that do not meet the conditions, and retain images with moderate exposure;

[0013] S2. Then, input the cleaned lightning image data into the YOLO-lightning network model to perform instance segmentation on the lightning image data;

[0014] The YOLO-lightning network model is obtained by lightweight improvement of the YOLOv8-seg network model: use the C2f_PBConv module to replace the original C2f module, and introduce a lightweight Triple Fusion module in the Neck part for multi-scale feature fusion; the YOLO-lightning network model specifically includes a Backbone part, a Neck part and a head part;

[0015] Among them, the Backbone part includes Conv, C2f_PBConv, and SPPF modules; the Conv module includes a Conv2d two-dimensional convolutional layer, a BN layer, and a SiLU activation function; the SPPF module includes a Conv module and a MaxPool2d layer;

[0016] The Neck part includes C2f_PBConv modules, Triple-Fusion modules, Concatenate, and Conv modules;

[0017] The head part includes a detection head and a segmentation head;

[0018] S3. Finally, output the contour information of the lightning to be recognized in the image.

[0019] Preferably according to the present invention, the instance segmentation of the lightning image means that the lightning image to be recognized is first input into the Backbone part to obtain a feature map, which is mainly used to extract the feature information of the image data; the Neck part fuses the feature maps extracted by the Backbone to enhance the representation ability and semantic information of the feature maps; the head part takes the output feature map of the Neck part as input to obtain the contour information of the lightning to be recognized.

[0020] Preferably according to the present invention, the C2f_PBConv module includes: a feature input Input, a convolutional layer Conv, a Split layer, a PBConv_Neck module, a feature concatenation layer Concatenate, and an output convolutional layer.

[0021] Preferably according to the present invention, the specific processing flow of the C2f_PBConv module is as follows:

[0022] S11. The features of the input image are first processed by the convolutional layer Conv to perform preliminary feature extraction on the input image features;

[0023] S12. Then it enters the Split layer for processing. The processed features are divided into two parts. One part of the features is directly passed through the skip connection Shortcut to the feature concatenation layer Concatenate, and the other part enters n repeated PBConv_Neck modules for in-depth processing;

[0024] S13. The features processed by n PBConv_Neck modules and the features that reach the feature concatenation layer Concatenate through the skip connection are concatenated in Concatenate to retain multi-scale and multi-channel feature information;

[0025] S14. Finally, the fused features are compressed and output through the last convolutional layer Conv to generate the segmentation result.

[0026] Preferably according to the present invention, the PBConv_Neck module replaces the traditional bottleneck layer Bottleneck and includes partial batch normalization convolution PBConv, SiLU activation layer, depth convolution Dw_Conv, pointwise convolution Pw_Conv, Dropout random inactivation layer and residual connection; the input features entering the PBConv_Neck through the Split layer sequentially pass through PBConv, SiLU activation layer, Dw_Conv, Pw_Conv, Dropout and residual connection to extract richer local and global features.

[0027] Preferably according to the present invention, the specific processing flow of the PBConv_Neck module is as follows:

[0028] S21. The input image features pass through the partial batch normalization convolution PBConv to perform convolution operations on some channels, and the convolved features pass through batch normalization BatchNormalization to make the feature distribution more stable;

[0029] S22. After the output of PBConv, the features enter the SiLU activation layer and are processed by the SiLU activation function;

[0030] S23. After being processed by the SiLU activation function, the features enter the depth convolution Dw_Conv, and each channel independently performs convolution operations to capture spatial features;

[0031] S24. Then it enters the pointwise convolution Pw_Conv, and the features of different channels are combined through 1x1 convolution, greatly reducing the number of parameters and the amount of calculation, and further realizing lightweight design;

[0032] S25. To avoid overfitting of the model, it enters the Dropout layer to randomly discard some neurons;

[0033] S26. Finally, the Add layer is used to fuse the features processed by the Dropout layer with the original input features of the PBConv_Neck module.

[0034] Preferably according to the present invention, the TripleFusion module combines deformable convolution DeformableConvolution and CBAM, and improves the flexibility and accuracy of feature extraction by dynamically adjusting the convolution kernel and the attention mechanism.

[0035] Preferably according to the present invention, the specific processing process of the TripleFusion module is as follows:

[0036] S31. Input the low - level features Lowerstagelayerfeature, corresponding - layer features Correspondingstagelayerfeature, and high - level features Higherstagelayerfeature from the Backbone part of the YOLO - lightning network into the TripleFusion module;

[0037] S32. The DeformableConvolution performs deformable convolution processing on the low - level features Lowerstagelayerfeature: The DeformableConvolution enhances the ability to capture irregular shapes by dynamically adjusting the positions of the convolution kernels;

[0038] S33. The low - level features Lowerstagelayerfeature processed by the DeformableConvolution, together with the corresponding - layer features and high - level features, enter the Concatenate layer for feature concatenation to obtain the fused features;

[0039] S34. The fused features are processed through the CBAM attention mechanism module;

[0040] The CBAM attention mechanism module weights according to the different feature importances in the channel and spatial dimensions, strengthening the attention to key features;

[0041] S35. The fused features processed by CBAM are output as the fused feature Fusionfeature, and the fused feature Fusionfeature contains feature information of multiple scales and multiple levels.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] (1) The present invention constructs the C2f_PBConv module by using the PBConv_Neck with partial batch - normalization convolution PBConv as the core, and makes a lightweight improvement to the original C2f module. C2f_PBConv is optimized on the basis of C2f. By only performing convolution processing on some channels and combining batch normalization technology, the computational amount is significantly reduced. While reducing the computational complexity of the model, the PBConv module still maintains a high feature extraction accuracy, and is particularly suitable for instance segmentation tasks of complex and variable - shaped targets such as lightning, enabling the model to operate efficiently while reducing the number of parameters.

[0044] (2) The TripleFusion module of the present invention combines the Deformable Convolution and the Convolutional Block Attention Module (CBAM). Through the combination of these two lightweight technologies, the improvement of accuracy and efficiency is achieved. The Deformable Convolution dynamically adapts to the changes of complex-shaped targets such as lightning, effectively processes multi-scale low-level features, and enhances the capture of details. The CBAM module weights important regions through channel and spatial attention mechanisms, further improving the focusing ability on key features. Without significantly increasing parameters, the TripleFusion module expands the receptive field through multi-scale feature fusion, improves the feature extraction ability of the model, and ensures the efficient operation of the model. Brief Description of the Drawings

[0045] Figure 1 is the structural diagram of the C2f_PBConv module of the present invention.

[0046] Figure 2 is the structural diagram of the PBConv module of the present invention.

[0047] Figure 3 is the structural diagram of the TripleFusion module of the present invention.

[0048] Figure 4 is the structural diagram of the YOLO-lightning network model of the present invention.

[0049] Figure 5 is the lightning image obtained after instance segmentation using the existing instance segmentation method.

[0050] Figure 6 is the lightning image obtained after instance segmentation using the instance segmentation method of the present invention. Detailed Embodiment

[0051] The following further describes the present disclosure in conjunction with the drawings and embodiments.

[0052] It should be noted that the following detailed description is exemplary and is intended to provide further description of the present disclosure. 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 disclosure belongs.

[0053] Embodiment 1

[0054] This embodiment provides a lightweight lightning image instance segmentation method, and the method includes:

[0055] First, collect the image data of lightning and preprocess the data;

[0056] Then, the preprocessed lightning image data is input into the YOLO-lightning network for instance segmentation of the lightning image data;

[0057] As Figure 4 shown, the YOLO-lightning network is obtained by lightweight improvement of the YOLOv8-seg network: the original C2f module is replaced with the C2f_PBConv module, and a lightweight Triple Fusion module is introduced in the Neck part for multi-scale feature fusion. The YOLO-lightning network model specifically includes a Backbone part, a Neck part, and a head part;

[0058] Among them, the Backbone part includes Conv, C2f_PBConv, and SPPF modules; the Conv module includes a Conv2d two-dimensional convolutional layer, a BN layer, and a SiLU activation function; the SPPF module includes a Conv module and a MaxPool2d layer;

[0059] The Neck part includes C2f_PBConv modules, Triple-Fusion modules, Concatenate, and Conv modules;

[0060] The head part includes a detection head and a segmentation head;

[0061] The lightning image to be recognized is first passed into the Backbone backbone part to extract the feature information of the image data to obtain a feature map; then the Neck neck part fuses the feature maps extracted by the Backbone to enhance the representation ability and semantic information of the feature maps; finally, the head head part takes the output feature map of the Neck part as input to obtain the contour information of the lightning to be recognized.

[0062] Figure 1 shows the overall architecture of the C2f_PBConv module, which is optimized on the basis of the C2f module in YOLOv8-seg. The C2f_PBConv module includes a feature input Input, a convolutional layer Conv, a Split layer, a PBConv_Neck module, a feature concatenation layer Concatenate, and an output convolutional layer.

[0063] The specific processing flow of the C2f_PBConv module is as follows:

[0064] The features of the input image input are first processed by the convolutional layer Conv to perform preliminary feature extraction on the input image features; this convolutional operation is similar to the traditional C2f module and Bottleneck operations;

[0065] Next, it enters the Split layer for processing. The processed features are divided into multiple parts. One part of the features is directly passed to the Concatenate layer through the skip connection Shortcut, and the other part enters n repeated PBConv_Neck modules for in-depth processing;

[0066] After being processed by n PBConv_Neck modules, the directly passed skip connection and the features processed by the PBConv_Neck module are merged here through the feature concatenation layer Concatenate to ensure the retention of multi-scale and multi-channel feature information.

[0067] Finally, the fused features are compressed and output through the last convolutional layer Conv to generate the segmentation result.

[0068] The PBConv_Neck module is designed based on the partially batch-normalized convolution PBConv, aiming to improve the efficiency of feature extraction and reduce the computational cost through partial-channel convolution and batch normalization techniques. The core of its design is to achieve the lightweight of the model by effectively reducing unnecessary computations while maintaining a high feature extraction ability. The PBConv_Neck module replaces the traditional bottleneck layer Bottleneck, as Figure 1 shown in the right half, including PBConv, SiLU activation, Dw_Conv, Pw_Conv, Dropout, and residual connection; the input features pass through PBConv, SiLU activation, Dw_Conv, Pw_Conv, Dropout, and residual connection in sequence to extract richer local and global features.

[0069] The specific processing flow is as follows:

[0070] First, as Figure 2 shown, the input image features pass through the partially batch-normalized convolution PBConv to perform convolution operations on some channels, avoiding redundant processing of all channels. In this way, the computational cost is significantly reduced, and the overall computational efficiency is improved. The convolved features will go through batch normalization BatchNormalization, which can make the feature distribution more stable, effectively prevent gradient vanishing or explosion, and accelerate the training of the model. Figure 2 In, Identity represents the identity mapping, filters represents the filter, and convolution represents the convolution.

[0071] After the output of PBConv, the feature map is processed by the SiLU activation function, which is an improved lightweight activation function. It not only retains the non-linear expression ability but also reduces the loss of negative value information that may be caused by traditional activation functions (such as ReLU) by optimizing the gradient flow. SiLU helps to accelerate the training and inference processes of the network.

[0072] After being processed by the SiLU activation function, the feature map enters the depthwise convolution Dw_Conv. Each channel independently performs the convolution operation to capture spatial features. The introduction of depthwise convolution significantly reduces the computational complexity in the convolution operation.

[0073] Then it enters the pointwise convolution Pw_Conv: The features of different channels are combined through 1x1 convolution, greatly reducing the number of parameters and computational complexity, and further realizing the lightweight design.

[0074] To avoid overfitting of the model, it enters the Dropout layer to randomly discard some neurons. This mechanism enhances the generalization ability of the model and makes the model more robust during training and inference. The introduction of Dropout can ensure the robustness of the model while maintaining lightweight.

[0075] Finally, the Add layer is used to fuse the processed features with the original input. This design ensures the continuity of information transmission through residual connections, effectively alleviating the problem of gradient disappearance. At the same time, the introduction of the Add layer does not significantly increase the computational complexity but can maintain the integrity and stability of the features during the forward propagation process.

[0076] In the comparison between the C2f_PBConv module and the traditional C2f module, by optimizing the convolution operation, feature extraction method, and adding a regularization mechanism, the C2f_PBConv module has significant advantages in lightweight:

[0077] Higher computational efficiency: The traditional C2f module uses conventional convolution to process all channels, resulting in a large amount of computational complexity, especially when dealing with large-scale inputs, the computational efficiency is low. While in the C2f_PBConv module, multiple PBConv_Neck structures are adopted, and the PBConv in it only performs convolution processing on some channels, avoiding redundant calculations and significantly reducing the computational complexity. The introduction of batch normalization BatchNormalization further improves the computational efficiency and training stability, speeds up the inference speed FPS, and better meets the requirements of lightweight design.

[0078] More refined feature extraction: The C2f_PBConv module effectively separates spatial information and channel information by introducing depthwise separable convolutions Dw_Conv and Pw_Conv. While reducing the number of parameters, it improves the accuracy of feature extraction. In contrast, the traditional C2f module uses regular convolutions and cannot effectively separate space and channels, resulting in insufficient feature extraction accuracy. Especially when dealing with complex, multi-branched, or multi-scale targets, the performance of the traditional module is poor. The lightweight PBConv_Neck design enables C2f_PBConv to maintain high computational performance while improving accuracy.

[0079] Stronger generalization ability: A Dropout layer is introduced in PBConv_Neck, which further enhances the generalization ability of the C2f_PBConv module. Dropout effectively reduces the model's dependence on specific data by randomly discarding some neurons, avoiding overfitting. This makes the C2f_PBConv module more robust when dealing with complex and morphologically variable targets. In contrast, the traditional C2f module is prone to overfitting when dealing with diverse datasets due to the lack of a clear regularization mechanism.

[0080] The core design idea of the TripleFusion module is to enhance the detection and segmentation capabilities for complex-shaped targets (such as lightning) through the fusion of multi-scale features. As Figure 3 shown, the TripleFusion module combines Deformable Convolution and CBAM, improving the flexibility and accuracy of feature extraction by dynamically adjusting the convolution kernel and attention mechanism.

[0081] The specific processing process of the TripleFusion module is as follows:

[0082] The lower-stage layer features Lowerstagelayerfeature, corresponding-stage layer features Correspondingstagelayerfeature, and higher-stage layer features Higherstagelayerfeature from the Backbone part of the YOLO-lightning network are input into the TripleFusion module; these features represent different levels of spatial resolution and semantic information, and fusing these features can capture both local and global information of the target.

[0083] The Deformable Convolution of the deformable convolution performs deformable convolution processing on the lower-stage layer features. The deformable convolution enhances the model's ability to capture irregular shapes by dynamically adjusting the positions of the convolution kernels, which is particularly suitable for processing targets with complex shapes such as lightning. This operation can effectively extract local features in lightning images and capture the shape changes of objects.

[0084] The lower-stage layer features processed by the deformable convolution, together with the corresponding layer features and high-level features, enter the Concatenate layer for feature concatenation, fusing feature information at different scales together to ensure that the network can consider multi-scale local and global features simultaneously during the processing.

[0085] The fused features are processed through the CBAM attention mechanism module;

[0086] The CBAM attention mechanism module is a module that combines channel attention and spatial attention. It can weight according to different feature importances in the channel and spatial dimensions, thereby strengthening the attention to key features. Especially for targets with complex backgrounds and many details such as lightning, CBAM can effectively improve the network's attention to key regions, contributing to more accurate instance segmentation.

[0087] Finally, the fused features processed by CBAM are output as the Fusion feature. This feature will contain feature information at multiple scales and multiple levels. Strengthened by the attention mechanism, it can more accurately capture the details and shape changes of the target and be used for subsequent segmentation operations.

[0088] Compared with traditional multi-scale feature fusion methods, the Triple Fusion module has achieved significant advantages through lightweight design in the following aspects:

[0089] Dynamic feature capture: Through the Deformable Convolution, the network can dynamically adjust the positions of the convolution kernels according to the shape changes of the target, thereby flexibly capturing the target features with irregular shapes. This lightweight mechanism reduces the calculation of redundant features and is particularly suitable for processing targets with complex shapes and many branch changes, such as lightning. Compared with traditional convolution operations, the deformable convolution can improve the adaptability to complex targets without significantly increasing the computational amount.

[0090] Multi-scale Feature Fusion: The TripleFusion module combines low-level, middle-level, and high-level feature information through the Concatenate operation, achieving lightweight multi-scale feature fusion. Compared with traditional multi-scale fusion methods, the TripleFusion module avoids redundant calculations, ensures the capture of local and global information of the target while reducing redundant operations. This lightweight feature fusion scheme is particularly suitable for processing targets with large ranges and obvious morphological changes, and can improve segmentation accuracy while maintaining the efficiency of the model.

[0091] Attention Mechanism Enhancement: The introduction of the Convolutional Block Attention Module (CBAM) weights the features in two dimensions, channel attention and spatial attention, effectively enhancing the attention to key regions. The weighting operation of CBAM helps the network focus on key features, reduces the processing of unimportant background information, and thus realizes the reasonable allocation of computing resources. This design improves the feature extraction accuracy and the overall computing efficiency of the model in complex scenarios without increasing the computational complexity.

[0092] Computational Efficiency Optimization: Although the TripleFusion module combines multiple feature processing mechanisms, by introducing lightweight technologies such as deformable convolution and CBAM, its feature extraction and fusion process is more efficient and flexible. Deformable convolution reduces invalid calculations, ensuring that only necessary feature regions are processed, while CBAM optimizes the allocation of computing resources through the attention mechanism, focusing on important features. This lightweight design significantly improves the computational efficiency, ensuring that the model can still perform rapid inference when processing complex targets.

[0093] Finally, the lightning image obtained by instance segmentation is as Figure 6 shown. It can be seen that the lightning image obtained by instance segmentation using the improved method of the present invention, compared with Figure 5 the lightning image obtained by unimproved instance segmentation shown, enhances the ability to capture the detailed features of lightning.

[0094] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solutions of the present invention, rather than limitations on the specific implementation manners of the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the claims of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. A lightweight lightning image instance segmentation method, characterized in that: The method comprises: S1. First, obtain lightning images in the target area during thunderstorms, clean them, remove images that do not meet the requirements, and retain images with moderate exposure; S2. Then, the cleaned lightning image data is input into the YOLO-lightning network model to perform instance segmentation on the lightning image. The YOLO-lightning network model is obtained by lightweight improvement of the YOLOv8-seg network model: the C2f_PBConv module is used to replace the original C2f module, and the lightweight TripleFusion module is introduced in the Neck part for multi-scale feature fusion; the YOLO-lightning network model specifically includes the Backbone part, the Neck part and the head part; The Backbone part includes Conv, C2f_PBConv and SPPF modules; the Conv module includes a Conv2d two-dimensional convolution layer, a BN layer and a SiLU activation function; the SPPF module includes a Conv module and a MaxPool2d layer; The Neck part includes a C2f_PBConv module, a TripleFusion module, a Concatenate and a Conv module; The head part includes a detection head and a segmentation head; S3. Finally, the contour information of the lightning to be identified in the image is output.

2. The lightweight lightning image instance segmentation method according to claim 1, characterized in that: The instance segmentation of lightning images means that the lightning image to be identified is firstly transmitted to the Backbone part to extract the feature information of the image data to obtain a feature map; then the Neck part fuses the feature map extracted by the Backbone to enhance the representation ability and semantic information of the feature map; finally, the head part uses the output feature map of the Neck part as input to obtain the contour information of the lightning to be identified.

3. The lightweight lightning image instance segmentation method according to claim 1, characterized in that: The specific processing flow of the C2f_PBConv module is as follows: S11, the features of the input image are first processed by the convolution layer Conv to perform preliminary feature extraction on the features of the input image; S12, then enter the Split layer for processing, the processed features are divided into two parts, one part of the features is directly passed to the feature concatenation layer Concatenate through the jump connection Shortcut, and the other part enters n repeated PBConv_Neck modules for deep processing; S13, the features processed by n PBConv_Neck modules and the features directly reaching the feature concatenation layer Concatenate through the jump connection are concatenated in Concatenate to retain the multi-scale and multi-channel feature information; S14. Finally, the fused features are compressed and output through the last convolutional layer Conv.

4. The lightweight lightning image instance segmentation method according to claim 3, characterized in that: The PBConv_Neck module includes a partial batch normalization convolution PBConv, a SiLU activation layer, a depth convolution Dw_Conv, a point-by-point convolution Pw_Conv, a Dropout random inactivation layer and a residual connection; the input features entering the PBConv_Neck through the Split layer pass through PBConv, a SiLU activation layer, Dw_Conv, Pw_Conv, Dropout and a residual connection in sequence.

5. The lightweight lightning image instance segmentation method according to claim 4, characterized in that: The specific processing flow of the PBConv_Neck module is as follows: S21, the input image features pass through partial batch normalization convolution PBConv, perform convolution operation on some channels, and the convolved features undergo batch normalization BatchNormalization; S22, after PBConv output, the feature enters the SiLU activation layer and is processed by the SiLU activation function; S23, the features are processed by SiLU activation function and then enter the deep convolution Dw_Conv. Each channel performs convolution operation independently to capture spatial features. S24, then enter the point-by-point convolution Pw_Conv, and combine the features of different channels through 1x1 convolution; S25, enter the Dropout layer to randomly discard some neurons; S26. Finally, the Add layer is used to merge the features processed by the Dropout layer with the features of the original input of the PBConv_Neck module.

6. The lightweight lightning image instance segmentation method according to claim 1, characterized in that: The TripleFusion module combines deformable convolution and CBAM.

7. The lightweight lightning image instance segmentation method according to claim 6, characterized in that: The specific processing process of the TripleFusion module is as follows: S31, input the lower-level feature Lowerstagelayerfeature, the corresponding layer feature Correspondingstagelayerfeature and the higher-level feature Higherstagelayerfeature from the Backbone part of the YOLO-lightning network into the TripleFusion module; S32. Deformable Convolution performs deformable convolution processing on the lower-stagelayerfeature: Deformable Convolution enhances the ability to capture irregular shapes by dynamically adjusting the position of the convolution kernel; S33, the lower-level features Lowerstagelayerfeature processed by deformable convolution and the corresponding layer features and high-level features enter the feature concatenation layer for feature concatenation to obtain the fused features; S34, the fused features are processed by the CBAM attention mechanism module; The CBAM attention mechanism module weights the importance of different features in channel and spatial dimensions to strengthen the focus on key features; S35. The fused features processed by CBAM are output as fused features Fusionfeature, and the fused features Fusionfeature contain feature information of multiple scales and multiple levels.

Images