Method, device and system for identifying invading foreign matter on power transmission line
Through the improved SimplifyNet model, the computational complexity problem of invasive foreign object recognition on power transmission lines in resource-constrained environments is solved, and efficient inference performance and concise model structure are achieved.
Patent Information
- Application Number
- CN202510140214.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-08
- Publication Date
- 2025-05-16
AI Technical Summary
The prior art has complex deployment and difficult to achieve efficient computing and reasoning in resource-constrained environments in terms of intrusive foreign matter recognition on transmission lines.
Using the improved SimplifyNet model, which includes initial convolutional layer, multiple convolution pooling groups and fully connected layers, simplifies network structure and reduces computational complexity through time-varying activated deep training strategies and stacked activation functions.
Efficient inference performance is achieved in resource-constrained environments, suitable for real-time foreign object recognition tasks, and the model structure is simple, the number of parameters is small, and the convergence speed is fast.
Smart Images
Figure CN120012859A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of target recognition, and in particular relates to an edge recognition method, device and system for intruding foreign objects on a transmission line. Background Art
[0002] As a key link in the power transmission system, the stability of transmission lines is very important. In actual operation, transmission lines may encounter interference from various foreign objects, such as plastic films, plant debris, bird nests, etc. These foreign objects will not only cause physical damage to the transmission lines, but also trigger electrical problems such as circuit short circuits and arc discharges. In extreme cases, they may also induce large-scale power outages, which will bring serious economic and social consequences.
[0003] In order to promptly remove foreign objects from transmission lines, the existing technology uses manual inspections or automated detection systems for detection. Manual inspections are not only time-consuming and labor-intensive, but also have limitations in detection efficiency, making it difficult to immediately discover and handle foreign objects. Technological advances have led to the gradual introduction of automated detection systems into the monitoring system of transmission lines. Automated detection systems are more convenient than manual inspections, and the results are obtained more quickly and accurately, but these methods are too complex and relatively difficult to deploy. In terms of foreign object identification on transmission lines, it is worth further improving how to maintain efficient computing and reasoning capabilities in an environment where edge client resources are limited. Summary of the invention
[0004] To address the deficiencies in the prior art, the present invention provides a method, device and system for identifying intruding foreign objects on transmission lines, which simplifies the model structure, reduces the computational complexity, avoids the use of deep and complex operations, and achieves efficient reasoning performance in a resource-constrained environment.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is: In a first aspect, a method for identifying intruding foreign objects on a transmission line is provided. The method is executed by an edge client deployed at a set position of the transmission line, and includes: collecting images of the transmission lines in a set area; inputting the collected images of the transmission lines into a trained improved SimplifyNet model, and outputting the identification results of the intruding foreign objects on the transmission line.
[0006] In combination with the first aspect, the improved SimplifyNet model includes: an initial convolution layer, multiple convolution pooling groups and a fully connected layer; the initial convolution layer is used to extract the structural feature map of the transmission line image; the convolution pooling group is used to extract the detail features of the transmission line image from the structural feature map of the transmission line image, and the multiple convolution pooling groups are serially connected and have the same structure; the fully connected layer is used to establish a mapping from the detail features extracted by the convolution pooling group to the classification results, and output the classification results.
[0007] Combined with the first aspect, each convolution and pooling group includes a convolution layer, a pooling layer and an activation layer, wherein the convolution layer is used to extract detail features from the structural feature map, and the convolution layer adopts a 1×1 convolution kernel to reduce the amount of calculation while doubling the number of channels; the pooling layer is used to reduce the dimension of the extracted detail features so that the size of the extracted detail features is halved; the activation layer uses a deep training strategy of time-varying activation and a stacked activation function to convolve the nonlinear ability of the pooling group.
[0008] Combined with the first aspect, a deep training strategy of time-varying activation is used during the training process to enhance the nonlinear ability of the convolutional pooling group. The deep training strategy of time-varying activation is: as the training cycle increases, the activation function is gradually simplified to an identity mapping. Specifically, during training, in each iteration cycle of training, the activation function is Combined with the identity mapping, we get the time-varying activation function , for: , in, is a hyperparameter used to balance the nonlinearity of the activation function; , e is the current training cycle, E is the total training cycle; at the beginning of training, e =0, , indicating that the improved SimplifyNet model has stronger nonlinearity; when the training is finished, e = E , , indicating that no activation is required between convolutional layers.
[0009] Combined with the first aspect, for the input feature x∈R H×W×C , where H, W and C are the width, height and channel values of the input feature respectively, R is the feature domain, and the stacking activation function is: , where h∈{1, 2, ..., H}, w∈{1, 2, ..., W}, c∈{1, 2, ..., C}; is the output activation function, is the basis function, is the weight term for the stacking position (i, j) in the cth channel, is the input feature of the c-th channel position (i+h, j+w), is the bias term in the cth channel, The stacking range.
[0010] Combined with the first aspect, the basis function of the stacked activation function Use the parameterized rectified linear function PReLU.
[0011] In combination with the first aspect, during training, the improved SimplifyNet model is trained using a training data set, and the method for obtaining the training data set includes: Obtain historical transmission line intrusion foreign object images and perform preprocessing to obtain the original data set; The original data set is annotated, and the images in the original data set are processed by image data enhancement technology to expand the data of the original data set to obtain a data set that meets the set requirements, and the data set that meets the set requirements is used as a training data set; Among them, image data enhancement techniques include image rotation, mirror flipping, brightness adjustment, Gaussian noise addition, image scaling and cropping.
[0012] In a second aspect, a device for identifying an intruding foreign object on a transmission line is provided, comprising: an edge client deployed at a set position of the transmission line, the edge client comprising: A data acquisition module, used to collect images of power transmission lines in a set area; The foreign object recognition module is used to input the collected transmission line images into the trained improved SimplifyNet model and output the recognition results of the intruding foreign objects on the transmission line.
[0013] Combined with the second aspect, the improved SimplifyNet model includes: an initial convolutional layer, multiple convolutional pooling groups, and a fully connected layer; The initial convolutional layer is used to extract the structural feature map of the transmission line image; A convolution pooling group is used to extract detail features of the power transmission line image from the structural feature map of the power transmission line image. Multiple convolution pooling groups are serially connected and have the same structure. The fully connected layer is used to establish a mapping from the detailed features of the transmission line image extracted by the convolution pooling group to the classification results, and output the classification results.
[0014] Combined with the second aspect, the deep training strategy of time-varying activation is: as the training cycle increases, the activation function is gradually simplified to an identity mapping, specifically: during training, in each iteration cycle of training, the activation function is Combined with the identity mapping, we get the time-varying activation function , for: , in, is a hyperparameter used to balance the nonlinearity of the activation function; , e is the current training cycle, E is the total training cycle; at the beginning of training, e =0, , indicating that the improved SimplifyNet model has stronger nonlinearity; when the training is finished, e = E , , indicating that no activation is required between convolutional layers; For input feature x∈R H×W×C , where H, W and C are the width, height and channel values of the input feature respectively, R is the feature domain, and the stacking activation function is: , where h∈{1, 2, ..., H}, w∈{1, 2, ..., W}, c∈{1, 2, ..., C}; is the output activation function, is the basis function, is the weight term for the stacking position (i, j) in the cth channel, is the input feature of the c-th channel position (i+h, j+w), is the bias term in the cth channel, is the stacking range; Basis functions for stacked activation functions Use the parameterized rectified linear function PreLU.
[0015] In combination with the second aspect, during training, the improved SimplifyNet model is trained using a training data set, and the method for obtaining the training data set includes: Obtain historical transmission line intrusion foreign object images and perform preprocessing to obtain the original data set; The original data set is annotated, and the images in the original data set are processed by image data enhancement technology to expand the data of the original data set to obtain a data set that meets the set requirements, and the data set that meets the set requirements is used as a training data set; Among them, image data enhancement techniques include image rotation, mirror flipping, brightness adjustment, Gaussian noise addition, image scaling and cropping.
[0016] In a third aspect, a system for identifying intruding foreign objects on a transmission line is provided, comprising: a memory for storing instructions; a processor for executing the instructions so that the device performs operations to implement the method for identifying intruding foreign objects on a transmission line as described in the first aspect.
[0017] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed, the method for identifying intruding foreign objects on the transmission line as described in the first aspect is implemented.
[0018] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention adopts a deep training strategy. In the early stage of network training, the learning ability of the network is enhanced by using a deeper convolutional structure and a nonlinear activation function. As the training progresses, the activation function gradually weakens, and finally the complexity of the network is reduced by merging convolutional layers; (2) The present invention adopts a stacking activation function. In order to enhance the nonlinear expression ability of SimplifyNet, the present invention proposes a stacking-based activation function, which can stack multiple nonlinear transformations in parallel, thereby improving the nonlinear ability of the simple network; (3) The present invention has efficient reasoning performance. By reducing complex operations and hierarchical structures, SimplifyNet has a faster execution speed in the reasoning stage and can be applied to tasks that require real-time processing, such as foreign object recognition. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 It is a schematic diagram of the main process of a method for identifying the edge of an intruding foreign object on a power transmission line provided by an embodiment of the present invention; Figure 2 Schematic diagram of the creation and training process of the improved SimplifyNet model in an embodiment of the present invention; Figure 3 Schematic diagram of the structure of the improved SimplifyNet model in the embodiment of the present invention. DETAILED DESCRIPTION
[0020] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.
[0021] Embodiment 1 like Figure 1As shown, a method for identifying intruding foreign objects on a transmission line is executed by an edge client deployed at a set location of the transmission line, including: collecting transmission line images in a set area; inputting the collected transmission line images into a trained improved SimplifyNet model, and outputting the identification results of intruding foreign objects on the transmission line.
[0022] The creation and training process of the improved SimplifyNet model in the present invention is as follows: Figure 2 shown.
[0023] Step 1: Collect the image data of foreign objects on the power transmission line. The edge client extracts the intrusion foreign object images from the monitoring video, and obtains a total of 5,000 images. The intrusion foreign object images are processed through data enhancement technologies such as image rotation, mirror flipping, brightness adjustment, Gaussian noise addition, image scaling and cropping to obtain the processed intrusion foreign object images, that is, the processed foreign object dataset. The processed foreign object dataset contains five types of image data: bird's nest, balloon, kite, garbage, and no foreign object.
[0024] Step 2: randomly divide the processed data set into a training set, a validation set, and a test set in a ratio of 6:2:2. The training set contains 3,000 images, the validation set contains 1,000 images, and the test set contains 1,000 images.
[0025] Step 3: Build an improved SimplifyNet model, such as Figure 3 As shown in the figure, the improved SimplifyNet model includes: an initial convolution layer, multiple convolution pooling groups and a fully connected layer; the initial convolution layer uses a large-size convolution kernel to extract the structural feature map of the transmission line image; the convolution pooling group uses a small-size convolution kernel to extract the detail features of the transmission line image from the structural feature map of the transmission line image, and the multiple convolution pooling groups are serially connected and have the same structure; in each convolution pooling group, features are extracted by the convolution layer to double the number of convolution channels, and then the dimension is reduced by the pooling layer to halve the size of the feature map; the fully connected layer establishes a mapping from the detail features extracted by the convolution pooling group to the classification results, and outputs the classification results. The improved SimplifyNet model avoids the use of deep and complex hierarchical structures and does not use self-attention mechanisms or shortcut connections; the convolutional pooling group contains a convolutional layer, a pooling layer and an activation layer, where the convolutional layer uses a 1×1 convolution kernel to extract detail features and reduce the amount of computation, while doubling the number of channels; the pooling layer is used to reduce the dimensionality of the extracted detail features, so that the size of the extracted detail features is halved; the activation layer enhances the nonlinear ability of the convolutional pooling group through a deep training strategy of time-varying activation and a stacked activation function.
[0026] The improved SimplifyNet model is trained using a deep training strategy with time-varying activations. The main idea of the deep training strategy with time-varying activations is that as the training cycle increases, the activation function is gradually simplified to an identity mapping. Specifically, during training, in each iteration of the training, for the activation function , the present invention combines it with the identity mapping to obtain the time-varying activation function , It is expressed as: , in, is a hyperparameter used to balance the nonlinearity of the activation function; , e is the current training cycle, E is the total training cycle; at the beginning of training, e =0, , indicating that the improved SimplifyNet model has stronger nonlinearity; when the training is finished, e = E , , indicating that no activation is required between convolutional layers.
[0027] During training, the activation functions are stacked to improve the nonlinearity of the activation layer. Specifically, given an input feature x∈R H×W×C , where H, W and C are the width, height and channel value of the input feature respectively, R is the feature domain, and the stacked activation function formula is: , where h∈{1, 2, ..., H}, w∈{1, 2, ..., W}, c∈{1, 2, ..., C}; is the output activation function, is the basis function, is the weight term for the stacking position (i, j) in the cth channel, is the input feature of the c-th channel position (i+h, j+w), is the bias term in the cth channel, The stacking range.
[0028] The basis function of the stacked activation function in the present invention Use the parametric rectified linear unit (PReLU).
[0029] Step 4, by setting the training hyperparameters, the four edge client devices are equipped with Intel Core i5-7200CPU@2.5 GHz, 8 GB memory, running Windows 10 operating system, and PyTorch 1.10.2 version is installed.
[0030] To quantify the performance of the model, the present invention uses cross entropy error as the loss function. For parameter optimization, all parameters are optimized using the Adam optimizer. The learning rate is initialized to 0.001. The batch size is 32 and the training cycle is 100.
[0031] Step 5: On the edge client, use the training dataset to train the model.
[0032] Step 6: Model verification: Use the test data set to verify the model trained in step 5.
[0033] The following is a comparative experiment between the improved SimplifyNet model in the present invention and other models in the prior art: The present invention compares the improved SimplifyNet model with other baseline models including VGG, ResNet, MobileNet, YOLO, etc. The comparison indicators are precision, recall rate and F1, and the results are shown in Table 1.
[0034] Table 1 Performance comparison between the improved SimplifyNet model and the baseline model
[0035] The verification results of the improved SimplifyNet model in the present invention show that the average detection accuracy of the model can reach 87.2%. In terms of accuracy, it is 1.1% higher than that of YOLOv8, and its performance is comparable to that of the YOLOv8 model. However, the complexity of the model is much simpler than that of the YOLO series, and is suitable for edge deployment.
[0036] The present invention designs a new neural network architecture SimplifyNet, whose design concept is based on simplifying the structure and computational complexity of the network, avoiding the use of deep and complex operations, and thus achieving efficient reasoning performance in a resource-constrained environment. Through the improved SimplifyNet technology, it is intended to improve the computational efficiency in a resource-constrained environment by reducing the network depth and avoiding the use of complex operation modules (such as self-attention mechanisms). SimplifyNet has a very simple design concept, using only one convolutional layer at each stage, and gradually reducing nonlinear activation functions, making the network more efficient during reasoning. In addition, through the combination of a deep training strategy with time-varying activation and a stacked activation function, SimplifyNet significantly improves the nonlinear representation ability of the network while maintaining simplicity. Compared with the prior art, the improved SimplifyNet model has achieved performance comparable to that of a complex neural network in the task of identifying foreign objects intruding into a power transmission line, but has the advantages of a simple structure, fewer parameters, and faster convergence speed.
[0037] Embodiment 2 Based on the method for identifying an intruding foreign object on a transmission line described in Example 1, this embodiment provides an intruding foreign object identification device on a transmission line, including an edge client deployed at a set position of the transmission line, and the edge client includes: A data acquisition module, used to collect images of power transmission lines in a set area; The foreign object recognition module is used to input the collected transmission line images into the trained improved SimplifyNet model and output the recognition results of the intruding foreign objects on the transmission line.
[0038] The improved SimplifyNet model includes: an initial convolution layer, multiple convolution pooling groups and a fully connected layer; the initial convolution layer is used to extract the structural feature map of the transmission line image; the convolution pooling group is used to extract the detail features of the transmission line image from the structural feature map of the transmission line image, and the multiple convolution pooling groups are serially connected and have the same structure; the fully connected layer is used to establish a mapping from the detail features extracted by the convolution pooling group to the classification results, and output the classification results.
[0039] The deep training strategy of time-varying activation is: as the training cycle increases, the activation function is gradually simplified to an identity mapping, specifically: during training, in each iteration of training, the activation function is Combined with the identity mapping, we get the time-varying activation function , for: , in, is a hyperparameter used to balance the nonlinearity of the activation function; , e is the current training cycle, E is the total training cycle; at the beginning of training, e =0, , indicating that the improved SimplifyNet model has stronger nonlinearity; when the training is finished, e = E , , indicating that no activation is required between convolutional layers; For input feature x∈R H×W×C , where H, W and C are the width, height and channel values of the input feature respectively, R is the feature domain, and the stacking activation function is: , where h∈{1, 2, ..., H}, w∈{1, 2, ..., W}, c∈{1, 2, ..., C}; is the output activation function, is the basis function, is the weight term for the stacking position (i, j) in the cth channel, is the input feature of the c-th channel position (i+h, j+w), is the bias term in the cth channel, is the stacking range; Basis functions for stacked activation functions To use the parameterized rectified linear function PreLU.
[0040] During training, the improved SimplifyNet model is trained using a training data set, and the method for obtaining the training data set includes: obtaining historical images of foreign objects intruding into power transmission lines and preprocessing them to obtain the original data set; annotating the original data set, and processing the images in the original data set through image data enhancement technology to expand the data of the original data set to obtain a data set that meets the set requirements, and using the data set that meets the set requirements as the training data set; wherein the image data enhancement technology includes image rotation, mirror flipping, brightness adjustment, Gaussian noise addition, image scaling and cropping.
[0041] Embodiment 3 Based on the method for identifying an intruding foreign object on a transmission line described in the first embodiment, this embodiment provides an intruding foreign object identification system on a transmission line, including: A memory for storing instructions; The processor is used to execute the instructions so that the device performs operations to implement the method for identifying foreign objects intruding into the transmission line as described in the first embodiment.
[0042] Embodiment 4 Based on the method for identifying intruding foreign objects on a transmission line described in Example 1, this embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, the method for identifying intruding foreign objects on a transmission line described in Example 1 is implemented.
[0043] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take 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.
[0044] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes 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 a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0045] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0046] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0047] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.
Claims
1. A method for identifying foreign objects intruding on a transmission line, characterized in that: The method is performed by an edge client deployed at a set location of a power transmission line, and includes: Collect transmission line images in a set area; The collected transmission line images are input into the trained improved SimplifyNet model to output the recognition results of intruding foreign objects on the transmission line.
2. The method for identifying foreign objects intruding into a transmission line according to claim 1, characterized in that: The improved SimplifyNet model includes: an initial convolutional layer, multiple convolutional pooling groups, and a fully connected layer; The initial convolutional layer is used to extract the structural feature map of the transmission line image; A convolution pooling group is used to extract detail features of the power transmission line image from the structural feature map of the power transmission line image. Multiple convolution pooling groups are serially connected and have the same structure. The fully connected layer is used to establish a mapping from the detailed features of the transmission line image extracted by the convolution pooling group to the classification results, and output the classification results.
3. The method for identifying foreign objects intruding into a transmission line according to claim 2, characterized in that: Each convolution and pooling group contains a convolution layer, a pooling layer and an activation layer. The convolution layer is used to extract detail features from the structural feature map. The convolution layer uses a 1×1 convolution kernel to reduce the amount of calculation and double the number of channels. The pooling layer is used to reduce the dimension of the extracted detail features, so that the size of the extracted detail features is halved. The activation layer enhances the nonlinear ability of the convolution and pooling group through a deep training strategy of time-varying activation and a stacked activation function.
4. The method for identifying foreign objects intruding into a transmission line according to claim 3, characterized in that: The deep training strategy of time-varying activation is: as the training cycle increases, the activation function is gradually simplified to an identity mapping, specifically: during training, in each iteration of training, the activation function is Combined with the identity mapping, we get the time-varying activation function , for: ， in, is a hyperparameter used to balance the nonlinearity of the activation function; , e is the current training cycle, E is the total training cycle; at the beginning of training, e =0, , indicating that the improved SimplifyNet model has stronger nonlinearity; when the training is finished, e = E , , indicating that no activation is required between convolutional layers.
5. The method for identifying foreign objects intruding into a transmission line according to claim 3, characterized in that: For input feature x∈R H×W×C , where H, W and C are the width, height and channel values of the input feature respectively, R is the feature domain, and the stacking activation function is: ， where h∈{1, 2, ..., H}, w∈{1, 2, ..., W}, c∈{1, 2, ..., C}; is the output activation function, is the basis function, is the weight term for the stacking position (i, j) in the cth channel, is the input feature of the c-th channel position (i+h, j+w), is the bias term in the cth channel, The stacking range.
6. The method for identifying foreign objects intruding into a transmission line according to claim 3, characterized in that: Basis functions for stacked activation functions Use the parameterized rectified linear function PReLU.
7. The method for identifying foreign objects intruding into a transmission line according to claim 3, characterized in that: During training, the improved SimplifyNet model is trained using a training data set, and the method for obtaining the training data set includes: Obtain historical transmission line intrusion foreign object images and perform preprocessing to obtain the original data set; The original data set is annotated, and the images in the original data set are processed by image data enhancement technology to expand the data of the original data set to obtain a data set that meets the set requirements, and the data set that meets the set requirements is used as a training data set; Among them, image data enhancement techniques include image rotation, mirror flipping, brightness adjustment, Gaussian noise addition, image scaling and cropping.
8. A device for identifying foreign objects intruding on a transmission line, characterized in that: include: An edge client deployed at a set location of a transmission line, the edge client comprising: A data acquisition module, used to collect images of power transmission lines in a set area; The foreign object recognition module is used to input the collected transmission line images into the trained improved SimplifyNet model and output the recognition results of the intruding foreign objects on the transmission line.
9. The device for identifying foreign objects intruding into a transmission line according to claim 8, characterized in that: The improved SimplifyNet model includes: an initial convolutional layer, multiple convolutional pooling groups, and a fully connected layer; The initial convolutional layer is used to extract the structural feature map of the transmission line image; A convolution pooling group is used to extract detail features of the power transmission line image from the structural feature map of the power transmission line image. Multiple convolution pooling groups are serially connected and have the same structure. The fully connected layer is used to establish a mapping from the detailed features of the transmission line image extracted by the convolution pooling group to the classification results, and output the classification results.
10. The device for identifying foreign objects intruding into a transmission line according to claim 9, characterized in that: The deep training strategy of time-varying activation is: as the training cycle increases, the activation function is gradually simplified to an identity mapping, specifically: during training, in each iteration of training, the activation function is Combined with the identity mapping, we get the time-varying activation function , for: ， in, is a hyperparameter used to balance the nonlinearity of the activation function; , e is the current training cycle, E is the total training cycle; at the beginning of training, e =0, , indicating that the improved SimplifyNet model has stronger nonlinearity; when the training is finished, e = E , , indicating that no activation is required between convolutional layers; For input feature x∈R H×W×C , where H, W and C are the width, height and channel values of the input feature respectively, R is the feature domain, and the stacking activation function is: ， where h∈{1, 2, ..., H}, w∈{1, 2, ..., W}, c∈{1, 2, ..., C}; is the output activation function, is the basis function, is the weight term for the stacking position (i, j) in the cth channel, is the input feature of the c-th channel position (i+h, j+w), is the bias term in the cth channel, is the stacking range; Basis functions for stacked activation functions Use the parameterized rectified linear function PreLU.
11. The device for identifying foreign objects intruding into a transmission line according to claim 9, characterized in that: During training, the improved SimplifyNet model is trained using a training data set, and the method for obtaining the training data set includes: Obtain historical transmission line intrusion foreign object images and perform preprocessing to obtain the original data set; The original data set is annotated, and the images in the original data set are processed by image data enhancement technology to expand the data of the original data set to obtain a data set that meets the set requirements, and the data set that meets the set requirements is used as a training data set; Among them, image data enhancement techniques include image rotation, mirror flipping, brightness adjustment, Gaussian noise addition, image scaling and cropping.
12. A system for identifying foreign objects intruding on a transmission line, characterized in that: include: A memory for storing instructions; The processor is used to execute the instruction so that the device performs the operation of implementing the method for identifying foreign objects intruding on the transmission line as described in any one of claims 1 to 7.
13. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, the method for identifying foreign objects intruding into a transmission line as described in any one of claims 1 to 7 is implemented.