Electric arc detection method and system based on lightweight neural network model

Abstract

The invention discloses an arc detection method and system based on a lightweight neural network model, and relates to the technical field of artificial intelligence. A to-be-detected image is input into an arc detection model; the arc detection model comprises a plurality of depth separable convolutions and is subjected to model training and model quantification in advance; and extracting multi-level features of the to-be-detected image according to each depth separable convolution through the arc detection model, and outputting a probability value that the to-be-detected image is an arc image according to the multi-level features. According to the invention, the arc detection model is established by using the depth separable convolution structure, the multilevel features of the input image can be extracted for arc detection, and the arc detection efficiency and precision are improved. According to the deep separable convolution structure, space convolution and channel convolution are separated in a convolution layer, network parameters and the calculation amount are remarkably reduced, and the system power consumption and the calculation efficiency are further optimized through model quantization. The problem that the arc signal detection method is difficult to give consideration to detection precision and system power consumption is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an arc detection method and system based on a lightweight neural network model. Background Art

[0002] With the rapid development of power systems, industrial equipment and intelligent monitoring technology, arc fault detection plays a vital role in electrical safety assurance. Arcs usually occur when current or voltage fluctuates greatly, especially in switchgear, power transformers, distribution systems and industrial machinery and equipment. Arcing is very common. Faults caused by arcs not only cause serious damage to equipment, but may also cause safety accidents such as fires. Therefore, real-time and accurate arc detection is of great significance for disaster prevention and loss reduction. However, traditional arc detection methods, such as ultrasonic methods and pulse current monitoring methods, have problems such as insufficient processing capabilities for complex arc signals, low detection accuracy and excessive system power consumption.

[0003] Therefore, the existing technology still needs to be improved and developed. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an arc detection method and system based on a lightweight neural network model in response to the above-mentioned defects of the prior art, aiming to solve the problem that the existing arc signal detection method is difficult to balance detection accuracy and system power consumption.

[0005] The technical solution adopted by the present invention to solve the problem is as follows: In a first aspect, an embodiment of the present invention provides an arc detection method based on a lightweight neural network model, the method comprising: Inputting the image to be detected into an arc detection model; the arc detection model includes a plurality of depth-separable convolutions and is pre-trained and quantized; The arc detection model extracts multi-level features of the image to be detected according to each of the depth-wise separable convolutions, and outputs a probability value that the image to be detected is an arc image according to the multi-level features.

[0006] In one embodiment, the model training method includes: Pre-build the initial arc detection model; Acquire a training data set; wherein the training data set includes a plurality of training samples and annotated data of each training sample, and each training sample is a preprocessed lightning image; The initial arc detection model is iteratively trained according to the training data set to obtain the trained arc detection model.

[0007] In one implementation, obtaining a training data set includes: Acquire a plurality of original lightning images, and for each of the original lightning images, preprocess the original lightning image by a preset image processing method to obtain a lightning image: Annotating each of the lightning images to obtain annotation data corresponding to each of the lightning images; A training data set is constructed according to each of the lightning images and the annotated data.

[0008] In one embodiment, preprocessing the original lightning image by a preset image processing method to obtain the lightning image includes: Binarizing the original lightning image to obtain a binary image; The arc region in the binary image is identified, and the binary image is segmented according to the arc region to obtain a lightning image.

[0009] In one embodiment, each depth-separable convolution includes a first convolution layer and a second convolution layer; the first convolution layer is used to perform spatial convolution to achieve separate convolution for each input channel; the second convolution layer is used to perform channel convolution to achieve point-by-point convolution on each output channel.

[0010] In one implementation, the first convolutional layer and the second convolutional layer are both fusion layers, and the fusion layer is obtained by merging a convolutional layer, a batch normalization layer, and an activation function.

[0011] In one embodiment, the model quantification method includes: Setting a quantization configuration, wherein the quantization configuration includes setting an activation value to use asymmetric quantization, setting a weight to use symmetric quantization, and setting a specified quantization format; adjusting the activation range of the arc detection model through a calibration operation; The adjusted arc detection model is converted into the quantized format through a large matrix multiplication engine.

[0012] In a second aspect, an embodiment of the present invention further provides an arc detection system based on a lightweight neural network model, the system comprising: An image input module, used for inputting the image to be detected into an arc detection model; the arc detection model includes a plurality of depth-separable convolutions and is pre-trained and quantized; The arc detection module is used to extract the multi-level features of the image to be detected according to each of the depth-separable convolutions through the arc detection model, and output the probability value of the image to be detected being an arc image according to the multi-level features.

[0013] In a third aspect, an embodiment of the present invention further provides a terminal, comprising a memory and one or more processors; the memory stores one or more programs; the program comprises instructions for executing an arc detection method based on a lightweight neural network model as described above; and the processor is used to execute the program.

[0014] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium on which a plurality of instructions are stored, wherein the instructions are suitable for being loaded and executed by a processor to implement the steps of any of the above-described arc detection methods based on a lightweight neural network model.

[0015] Beneficial effects of the present invention: In the embodiment of the present invention, the image to be detected is input into an arc detection model; the arc detection model includes a plurality of depth-separable convolutions, and is pre-trained and quantized; the arc detection model extracts multi-level features of the image to be detected according to each of the depth-separable convolutions, and outputs the probability value that the image to be detected is an arc image according to the multi-level features. The present invention adopts a depth-separable convolution structure to establish an arc detection model. The depth-separable convolution structure can extract multi-level features of the input image for arc detection, thereby improving the computational efficiency and accuracy of arc detection. Secondly, the depth-separable convolution structure separates spatial convolution and channel convolution in the convolution layer, significantly reducing network parameters and computational complexity, and model quantization further optimizes system power consumption and computational efficiency. This solves the problem that the existing arc signal detection method is difficult to balance detection accuracy and system power consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a flow chart of an arc detection method based on a lightweight neural network model provided in an embodiment of the present invention.

[0018] Figure 2 It is a schematic diagram of constructing a training data set provided by an embodiment of the present invention.

[0019] Figure 3 Schematic diagram of positive and negative training samples provided by an embodiment of the present invention.

[0020] Figure 4 It is a schematic diagram of the composition of the arc detection model provided by an embodiment of the present invention.

[0021] Figure 5 It is a structural schematic diagram of an arc detection model provided by an embodiment of the present invention.

[0022] Figure 6 It is a diagram showing the accuracy detection effect of the arc detection model provided by an embodiment of the present invention.

[0023] Figure 7 It is a parameter comparison diagram of the arc detection model provided by the embodiment of the present invention.

[0024] Figure 8 It is a schematic diagram of a quantization process provided by an embodiment of the present invention.

[0025] Fig. 9 It is a module schematic diagram of an arc detection system based on a lightweight neural network model provided in an embodiment of the present invention.

[0026] Fig.10 It is a principle block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The present invention discloses an arc detection method and system based on a lightweight neural network model. In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0028] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0029] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0030] In view of the above-mentioned defects of the prior art, the present invention provides an arc detection method based on a lightweight neural network model, wherein the method inputs the image to be detected into the arc detection model; the arc detection model includes a plurality of depth-separable convolutions, and is pre-trained and quantized; the arc detection model extracts the multi-level features of the image to be detected according to each of the depth-separable convolutions, and outputs the probability value that the image to be detected is an arc image according to the multi-level features. The present invention adopts a depth-separable convolution structure to establish an arc detection model, and the depth-separable convolution structure can extract the multi-level features of the input image for arc detection, thereby improving the computational efficiency and accuracy of arc detection. Secondly, the depth-separable convolution structure separates the spatial convolution and channel convolution in the convolution layer, significantly reducing the network parameters and the amount of calculation, and the model quantization further optimizes the system power consumption and computational efficiency. Thereby solving the problem that the existing arc signal detection method is difficult to balance the detection accuracy and system power consumption.

[0031] like Figure 1 As shown, the method includes: Step S100, inputting the image to be detected into an arc detection model; the arc detection model includes a plurality of depthwise separable convolutions and has been pre-trained and quantized; Step S200: extracting multi-level features of the image to be detected through the arc detection model according to each of the depthwise separable convolutions, and outputting a probability value that the image to be detected is an arc image according to the multi-level features.

[0032] Specifically, this embodiment establishes an arc detection model by adopting a lightweight neural network architecture to more efficiently analyze the characteristics of arc signals and achieve more accurate and rapid arc fault detection. The input data of the arc detection model is image data, which is defined as the image to be detected; the output data is a numerical value between 0 and 1, which is used to reflect the probability value of whether the image to be detected is an arc image. The arc detection model has been pre-trained and has learned the complex mapping relationship between input and output, so it can accurately identify whether it is an arc image based on the input image data. In order to improve the computational efficiency and accuracy of arc detection, this embodiment adopts a deep separable convolution structure to establish an arc detection model. Compared with the traditional convolutional neural network, the deep separable convolution structure separates the spatial convolution and channel convolution in the convolution layer, which significantly reduces the network parameters and the amount of calculation. The deep separable convolution learns the multi-level features of the arc signal during the training process. Therefore, during model reasoning, by extracting the multi-level features of the input image to be detected, it can be accurately identified whether it contains an arc. In addition, this embodiment further optimizes the power consumption and computational efficiency of the arc detection model by quantizing the arc detection model.

[0033] In one implementation, the model training method includes: Pre-build the initial arc detection model; Acquire a training data set; wherein the training data set includes a plurality of training samples and annotated data of each training sample, and each training sample is a preprocessed lightning image; The initial arc detection model is iteratively trained according to the training data set to obtain the trained arc detection model.

[0034] Specifically, in this embodiment, the model that has not been trained is defined as the initial arc detection model. The training data set consists of multiple training samples, each training sample corresponds to a preprocessed lightning image, and the lightning image needs to be labeled to indicate whether there is an arc and / or the arc state and features. The lightning images contained in the training data set have arcs of different states and features, for example, some lightning images have arcs with bifurcations, and some lightning images contain two arcs. A training sample in the training data set is input into the initial arc detection model, and after obtaining the probability value of the training sample predicted by the model as an arc image, the predicted data of the model is compared with the labeled data of the training sample and the model loss is calculated. It is determined whether the calculated model loss meets the training requirements. If not, it means that the current detection accuracy of the model does not meet the standard, and the model parameters are updated based on the model loss to improve the model performance. Repeat the above steps until the model loss meets the training requirements, and a trained arc detection model is obtained. In this embodiment, the arc detection model is iteratively trained through the training data set, so that the arc detection model can learn arc features of different forms and accurately classify arc images and non-arc images.

[0035] In one implementation, obtaining a training data set includes: Acquire a plurality of original lightning images, and for each of the original lightning images, preprocess the original lightning image by a preset image processing method to obtain a lightning image: Annotating each of the lightning images to obtain annotation data corresponding to each of the lightning images; A training data set is constructed according to each of the lightning images and the annotated data.

[0036] Specifically, Figure 2 As shown, in order to achieve the purpose of arc detection using neural networks or deep learning, this embodiment needs to collect a large amount of lightning image data to provide rich training samples for arc detection tasks. Lightning image data, namely the original lightning image, usually contains various arc shapes and background noise. Therefore, it is necessary to pre-process the original lightning image through a preset image processing method, and define the pre-processed image as a lightning image. Each lightning image is labeled to obtain a training sample, and the labeling method can be manual labeling. Each sample is labeled whether the image contains an arc and / or different states and characteristics of the arc, providing high-quality input data for subsequent deep learning models. As shown Figure 3 As shown, the left side is the positive sample containing the arc, and the right side is the negative sample not containing the arc.

[0037] In one implementation, preprocessing the original lightning image by a preset image processing method to obtain the lightning image includes: Binarizing the original lightning image to obtain a binary image; The arc region in the binary image is identified, and the binary image is segmented according to the arc region to obtain a lightning image.

[0038] Specifically, in order to highlight the characteristics of the arc and effectively improve the recognition accuracy of the subsequent neural network, this embodiment will perform a binarization process on the original lightning image. The binarization process can extract the arc edge information in the image and convert the complex lightning image into a clear binary image. At the same time, this embodiment will also perform image segmentation on the original lightning image to remove redundant background. In view of the more obvious characteristics and uncomplicated graphics of the arc data, an image size of 100x100 pixels can be selected to retain enough information while meeting the requirements of a shallow network depth. Figure 3 As shown in the left part of the figure, the lightning image after image segmentation eliminates redundant background information and retains the main features of the arc. The final classification constitutes the training data set of the arc detection model. The training data set contains positive and negative training samples. The positive training samples contain arcs, and the negative training samples do not contain arcs.

[0039] In one implementation, each depth-separable convolution includes a first convolution layer and a second convolution layer; the first convolution layer is used to perform spatial convolution to achieve separate convolution for each input channel; the second convolution layer is used to perform channel convolution to achieve point-by-point convolution on each output channel.

[0040] In summary, the depthwise separable convolution in this embodiment is composed of two convolutional layers, and the number of convolution kernels and channels are specially designed. Specifically, taking a depthwise separable convolution as an example, the depthwise separable convolution first performs spatial convolution, that is, each input channel is convolved separately, and then performs channel convolution, that is, point-by-point convolution is performed on each output channel. Or channel convolution is performed first and then spatial convolution. In other words, each depthwise separable convolution layer includes a spatial convolution + batch normalization layer + activation function, and a channel convolution + batch normalization layer + activation function. This embodiment can effectively reduce the number of convolution kernels and reduce computational complexity through depthwise separable convolution.

[0041] For example, the arc detection model includes 5 layers of depthwise separable convolutions. Figure 4 Convolutional layers 2 to 6 in are depth-separable convolutions, corresponding to Figure 5 DS convolution 1 to DS convolution 5 in are depthwise separable convolutions. Spatial convolution (depthwise) and channel convolution (pointwise) are the two specially designed convolution layers mentioned above. Figure 4 Convolutional layer 2 or Figure 5Taking DS convolution 1 as an example, the structure of each depth-separable convolution is (convolution layer + batch normalization layer + activation function) x2. In order to adapt to the image size of the training data set and extract the arc features, this embodiment designs a suitable intermediate feature layer size and convolution depth. Through multiple depth-separable convolutions, multi-level features of arc signals can be extracted. The multi-level features include low-level edge features to high-level arc shapes, such as Figure 6 As shown in Figure 2, the accuracy of arc detection based on multi-level features reaches 94.47%. The deep separable convolution structure provided in this embodiment greatly reduces the number of model parameters and calculations while maintaining high accuracy. Figure 7 As shown in the figure, the number of parameters is reduced by 20 times compared to the convolutional neural network with the same effect, and the amount of computation for a single forward propagation is reduced by 8.5 times, which enables the neural network to run efficiently in resource-limited environments such as embedded devices.

[0042] In one implementation, the first convolution layer and the second convolution layer are both fusion layers, and the fusion layer is obtained by merging a convolution layer, a batch normalization layer, and an activation function.

[0043] Specifically, in order to further optimize the power consumption and computing efficiency of the neural network model, this embodiment also performs quantization optimization on the arc detection model. The quantization optimization of the model requires the use of model layer fusion technology: In common convolutional neural networks, the convolution layer is usually followed by batch normalization and activation functions. Before quantization, the convolution layer, batch normalization layer, and activation function are fused to merge these continuous operations into one operation, thereby reducing the calculation steps and optimizing the quantization effect.

[0044] In one implementation, Figure 5 As shown, the specific structure of the arc detection model includes: a convolution layer, a first depth-separable convolution, a maximum pooling layer, a second depth-separable convolution, a third depth-separable convolution, a maximum pooling layer, a fourth depth-separable convolution, a fifth depth-separable convolution, and a global average pooling layer cascaded in sequence, wherein the convolution layer receives an input image to be detected, and the global average pooling layer outputs a probability value of whether the image to be detected is an arc image.

[0045] In short, the neural network architecture corresponding to the arc detection model in this embodiment includes multiple modules, including common structures such as convolution, batch normalization, activation function, pooling layer, and fully connected layer. The first layer of the arc detection model is a common convolution layer, that is, Figure 4 The convolutional layer 1 and Figure 5 The convolutional layer in refers to the ordinary convolutional layer, and the remaining 5 convolutional layers are depth-wise separable convolutional layers.

[0046] In one implementation, the model quantization method includes: Setting a quantization configuration, wherein the quantization configuration includes setting an activation value to use asymmetric quantization, setting a weight to use symmetric quantization, and setting a specified quantization format; adjusting the activation range of the arc detection model through a calibration operation; The adjusted arc detection model is converted into the quantized format through a large matrix multiplication engine.

[0047] Specifically, this embodiment uses FBGEMM (Facebook Large Matrix Multiplication Engine) as the quantization backend during the quantization process. Figure 8 As shown, model fusion is performed before quantization to optimize the quantization effect. Then the quantization configuration is set, including setting the activation value to use asymmetric quantization and the weight to use symmetric quantization. Assuming that the quantization goal is to convert the weights and activation values ​​of the model from 32-bit floating point numbers (FP32) to 8-bit integers (INT8), the quantization configuration also includes setting the data type or quantization format to INT8. After that, the activation range of the model is adjusted through calibration to ensure the accuracy after quantization: the input data flows through each layer of the quantized model (i.e., the arc detection model), and the observer captures the range of activation values ​​of each layer; according to the range collected by the calibration, appropriate quantization parameters are generated; and the activation range of each layer is adjusted by dynamic statistics of the calibration data. In short, the main purpose of calibration is to ensure that the quantization process can reasonably scale these values ​​by counting the dynamic range (i.e., numerical distribution) of the activation values ​​and / or weights of each layer in the arc detection model. After calibration, the arc detection model is converted to a quantized format. As shown Figure 7 As shown, this embodiment can reduce the computing memory consumption by 23.2% and the model file size by 70% by model quantization while reducing the model accuracy by only 1%, which greatly reduces the computing burden and is particularly suitable for hardware platforms with limited computing resources.

[0048] The advantages of the present invention are: 1. By adopting a lightweight neural network architecture, the characteristics of arc signals can be analyzed more efficiently, and more accurate and rapid arc fault detection can be achieved.

[0049] 2. By combining deep separable convolution and model quantization technology (i.e. combining neural network and hardware acceleration technology), the computational complexity and storage requirements of the model are significantly reduced while ensuring high detection accuracy. It has friendly hardware deployment characteristics, so that the optimized neural network can run efficiently on a low-power hardware platform, which is particularly suitable for real-time arc detection applications and other application scenarios with low power consumption and high real-time response requirements.

[0050] Based on the above embodiments, the present invention also provides an arc detection system based on a lightweight neural network model, such as Fig. 9 As shown, the system comprises: An image input module 01 is used to input the image to be detected into an arc detection model; the arc detection model includes a plurality of depth-separable convolutions and has been pre-trained and quantized; The arc detection module 02 is used to extract the multi-level features of the image to be detected according to each of the depth-separable convolutions through the arc detection model, and output the probability value of the image to be detected being an arc image according to the multi-level features.

[0051] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be shown as follows: Fig.10 As shown. The terminal includes a processor, a memory, a network interface, and a display screen connected via a system bus. Among them, the processor of the terminal is used to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, an arc detection method based on a lightweight neural network model is implemented. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.

[0052] Those skilled in the art will understand that Fig.10 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0053] In one implementation, the terminal has one or more programs stored in its memory, and is configured to be executed by one or more processors, wherein the one or more programs include instructions for performing an arc detection method based on a lightweight neural network model.

[0054] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0055] In summary, the present invention discloses an arc detection method and system based on a lightweight neural network model, wherein the method inputs an image to be detected into the arc detection model; the arc detection model includes a plurality of depth-separable convolutions, and is pre-trained and quantized; the arc detection model extracts multi-level features of the image to be detected according to each of the depth-separable convolutions, and outputs a probability value that the image to be detected is an arc image according to the multi-level features. The present invention adopts a depth-separable convolution structure to establish an arc detection model, and the depth-separable convolution structure can extract multi-level features of the input image for arc detection, thereby improving the computational efficiency and accuracy of arc detection. Secondly, the depth-separable convolution structure separates spatial convolution and channel convolution in the convolution layer, significantly reducing network parameters and computational complexity, and model quantization further optimizes system power consumption and computational efficiency. This solves the problem that the existing arc signal detection method is difficult to balance detection accuracy and system power consumption.

[0056] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. An arc detection method based on a lightweight neural network model, characterized in that: The method comprises: Inputting the image to be detected into an arc detection model; the arc detection model includes a plurality of depth-separable convolutions and is pre-trained and quantized; The arc detection model extracts multi-level features of the image to be detected according to each of the depth-wise separable convolutions, and outputs a probability value that the image to be detected is an arc image according to the multi-level features.

2. The arc detection method based on a lightweight neural network model according to claim 1, characterized in that: The model training method includes: Pre-build the initial arc detection model; Acquire a training data set; wherein the training data set includes a plurality of training samples and annotated data of each training sample, and each training sample is a preprocessed lightning image; The initial arc detection model is iteratively trained according to the training data set to obtain the trained arc detection model.

3. The arc detection method based on a lightweight neural network model according to claim 2 is characterized in that: Get the training dataset, including: Acquire a plurality of original lightning images, and for each of the original lightning images, preprocess the original lightning image by a preset image processing method to obtain a lightning image: Annotating each of the lightning images to obtain annotation data corresponding to each of the lightning images; A training data set is constructed according to each of the lightning images and the annotated data.

4. The arc detection method based on a lightweight neural network model according to claim 3 is characterized in that: Preprocessing the original lightning image by a preset image processing method to obtain a lightning image includes: Binarizing the original lightning image to obtain a binary image; The arc region in the binary image is identified, and the binary image is segmented according to the arc region to obtain a lightning image.

5. The arc detection method based on a lightweight neural network model according to claim 1, characterized in that: Each of the depth-separable convolutions includes a first convolution layer and a second convolution layer; the first convolution layer is used to perform spatial convolution to achieve separate convolution on each input channel; the second convolution layer is used to perform channel convolution to achieve point-by-point convolution on each output channel.

6. The arc detection method based on a lightweight neural network model according to claim 5, characterized in that: The first convolution layer and the second convolution layer are both fusion layers, and the fusion layer is obtained by merging a convolution layer, a batch normalization layer, and an activation function.

7. The arc detection method based on a lightweight neural network model according to claim 6, characterized in that: The method for quantifying the model includes: Setting a quantization configuration, wherein the quantization configuration includes setting an activation value to use asymmetric quantization, setting a weight to use symmetric quantization, and setting a specified quantization format; adjusting the activation range of the arc detection model through a calibration operation; The adjusted arc detection model is converted into the quantized format through a large matrix multiplication engine.

8. An arc detection system based on a lightweight neural network model, characterized in that: The system comprises: An image input module, used for inputting the image to be detected into an arc detection model; the arc detection model includes a plurality of depth-separable convolutions and is pre-trained and quantized; The arc detection module is used to extract the multi-level features of the image to be detected according to each of the depth-separable convolutions through the arc detection model, and output the probability value of the image to be detected being an arc image according to the multi-level features.

9. A terminal, characterized in that: The terminal includes a memory and one or more processors; the memory stores one or more programs; the program contains instructions for executing the arc detection method based on a lightweight neural network model as described in any one of claims 1-7; and the processor is used to execute the program.

10. A computer-readable storage medium having a plurality of instructions stored thereon, characterized in that: The instructions are suitable for being loaded and executed by a processor to implement the steps of the arc detection method based on a lightweight neural network model as described in any one of claims 1-7.

Images