A method and device for locating status indicator lights of complex electrical instrument equipment

Abstract

The present invention discloses a method and device for locating status indicators of complex electrical instrument equipment, relating to the technical field of status indicator light positioning. The method comprises the following steps: acquiring data, including real-time collected image data of status indicators of electrical instrument equipment; preprocessing the data; inputting the preprocessed data into a status indicator light detection model based on a deep neural network, and outputting positioning information of each indicator light. The status indicator light detection model based on the deep neural network is used to extract features from the preprocessed data to obtain image features at different scales; performing feature fusion on the image features at different scales to obtain a feature map; processing the feature map based on the feature map using a region proposal network and a binary classifier to obtain positioning information of the indicator lights; and outputting precise positioning information of each indicator light based on the positioning information of each indicator light based on spatial proximity analysis. The present invention enhances the detection capability of multi-scale indicator lights, effectively reducing false detections and missed detections.

Description

Technical Field

[0001] The present invention relates to the technical field of status indicator light positioning, and in particular to a status indicator light positioning method and device for complex electrical instrument equipment. Background Art

[0002] Traditional deep learning technology has been widely used in the field of target detection, but in complex industrial scenarios, there are still many bottlenecks in the detection and positioning of status indicator lights. First, the color and brightness characteristics of status indicator lights are easily significantly affected by lighting conditions. In strong light, low light, and even shadow environments, the detection performance of existing methods is generally poor. Secondly, the scene of multiple indicator lights in complex backgrounds further exacerbates the difficulty of detection. Due to the diverse equipment designs and significant background interference, false detections and missed detections are frequent in traditional methods. In addition, the detection of unlit indicator lights is even more difficult. Existing algorithms rely heavily on brightness features, and the recognition accuracy is greatly reduced when dealing with such situations. Finally, in the scene of multiple indicator lights, the spatial distribution information of each indicator light is not fully utilized in existing methods, which leads to insufficient reliability of positioning results and inaccurate positioning.

[0003] Therefore, the applicant has developed a status indicator light positioning method for complex electrical instrument equipment to solve the above problems. Summary of the Invention

[0004] The present invention proposes a status indicator light positioning method for complex electrical instrument equipment to solve the problem of inaccurate positioning of the existing status indicator lights.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] In one aspect, the present invention provides a method for locating status indicator lights of complex electrical instrument equipment, comprising:

[0007] Acquiring data, the data including real-time collected image data of status indicator lights of electrical instrument equipment;

[0008] Preprocessing the data to obtain preprocessed data;

[0009] Inputting the preprocessed data into a status indicator light detection model based on a deep neural network, outputting the positioning information of each indicator light, the status indicator light detection model based on the deep neural network is used to extract features from the preprocessed data to obtain image features of different scales, performing feature fusion on the image features of different scales to obtain a feature map, and processing the feature map based on the region proposal network and the binary classifier to obtain the positioning information of the indicator light;

[0010] Based on the spatial proximity analysis according to the positioning information of each indicator light, the precise positioning information of each indicator light is output.

[0011] Furthermore, data is obtained, including:

[0012] Real-time acquisition of image data of a first electrical instrument device status indicator light, wherein the image data of the first electrical instrument device status indicator light is a visible light image;

[0013] collecting image data of a second electrical instrument device status indicator light in real time, wherein the image data of the second electrical instrument device status indicator light is an infrared image;

[0014] The data is obtained by fusing the first electrical instrument equipment status indicator light image data and the second electrical instrument equipment status indicator light image data.

[0015] Furthermore, the data is preprocessed to obtain preprocessed data, including:

[0016] performing data cleaning on the data to obtain cleaned data;

[0017] Perform image enhancement processing on the cleaned data to obtain preprocessed data.

[0018] Furthermore, the pre-processed data is input into a status indicator light detection model based on a deep neural network to output the location information of each indicator light, including:

[0019] Performing feature extraction on the preprocessed data to obtain a multi-layer feature map;

[0020] Capture image features at different scales based on multi-layer feature maps;

[0021] The image features of different scales are fused based on a feature pyramid network to obtain a feature map;

[0022] Generate an anchor point on each of the feature maps based on a region proposal network through a preset sliding window;

[0023] Scoring each anchor point through a binary classifier to obtain anchor box parameters, and predicting the anchor box offset and scaling factor;

[0024] Generate a candidate region frame on the feature map according to the anchor frame parameters, the offset, and the scaling factor;

[0025] Based on the binary classifier, it is determined whether the candidate area frame contains an indicator light, and the category probability distribution of each candidate area frame is calculated. Based on the category probability distribution, the indicator light category corresponding to the category probability distribution with the maximum value in the category set is used as the candidate area frame category. Based on the regression head, the candidate area frame containing the indicator light is optimized to obtain the positioning information of each indicator light.

[0026] Furthermore, the regression formula for generating the candidate region frame on the feature map is as follows:

[0027]

[0028] Furthermore, the calculation formula for determining the category of the candidate region frame is as follows:

[0029] The anchor box parameters are , Respectively represent the horizontal coordinate, vertical coordinate, width and height of the center point of the anchor box.

[0030] The anchor box offset is , respectively represent the horizontal and vertical offsets of the center point of the candidate box relative to the center point of the anchor box, and the scaling factors of the width and height of the candidate box relative to the width and height of the anchor box.

[0031] The candidate region frame is represented as , Respectively represent the horizontal coordinate, vertical coordinate, width and height of the center point of the candidate box.

[0032] The category probability distribution is P(c), which represents the probability that the candidate region box belongs to category c.

[0033] The category set is C, and the category of the candidate region box is c*.

[0034] Furthermore, outputting precise positioning information of each indicator light based on spatial proximity analysis according to the positioning information of each indicator light includes:

[0035] Obtaining the actual two-dimensional coordinates of each indicator light according to the positioning information of each indicator light;

[0036] Obtaining the two-dimensional predicted coordinates of each indicator light according to the predicted information of each indicator light;

[0037] Constructing a spatial proximity matrix according to the actual coordinates of each of the indicator lights, and calculating an expansion vector of the spatial proximity matrix;

[0038] Constructing a prediction space proximity matrix according to the predicted coordinates of each of the indicator lights, and calculating an expansion vector of the prediction space proximity matrix;

[0039] Calculating cosine similarity based on the expansion vector of the spatial proximity matrix and the expansion vector of the predicted spatial proximity matrix;

[0040] According to the cosine similarity-based matrix dimensionality reduction and similarity optimization method, redundant or erroneous detection items in the spatial proximity matrix and the predicted spatial proximity matrix are eliminated to obtain accurate positioning results of each indicator light.

[0041] Furthermore, the calculation formula for calculating the cosine similarity based on the expansion vector of the spatial proximity matrix and the expansion vector of the predicted spatial proximity matrix is:

[0042]

[0043] Where A represents the expansion vector of the spatial proximity matrix, B represents the expansion vector of the predicted spatial proximity matrix, m is the total length of the vector, that is, the total number of indicator light pairs, k is the vector index, and A k Indicates the actual distance between the kth pair of indicator lights, B k represents the predicted distance between the kth pair of indicator lights, represents the cosine similarity.

[0044] Furthermore, removing redundant or erroneous detection items from the spatial proximity matrix and the predicted spatial proximity matrix includes:

[0045] First, gradually remove indicator light information from a matrix with a higher dimension between the spatial proximity matrix and the predicted spatial proximity matrix until the matrix dimensions of the predicted spatial proximity matrix match those of the spatial proximity matrix;

[0046] The number of indicator light information removed during the dimensionality reduction process is fixed, but different indicator light information can be selected for removal, and the cosine similarity is calculated for all possible dimensionality reduction combinations;

[0047] The group with the highest cosine similarity is selected for matching. At this time, each dimension of the predicted spatial proximity matrix and the spatial proximity matrix are matched, and a mapping between the predicted coordinates of the indicator light and the actual coordinates of the indicator light can be established to obtain the accurate positioning results of each indicator light.

[0048] Furthermore, it also includes:

[0049] A preset state mapping table includes an indicator light number, the two-dimensional actual coordinates of the indicator light, an indicator light state flag, and a device state description corresponding to the indicator light state. The indicator light number uniquely identifies each indicator light, and the indicator light state flag describes the indicator light state.

[0050] Obtaining status information and positioning information of all indicator lights, wherein the status information is image data of the status indicator lights of the electrical instrument equipment and the status flags, and the positioning information is the precise positioning result of each indicator light;

[0051] The state information and the positioning information are converted into a readable device state according to the preset state mapping table.

[0052] Another aspect of the present invention provides a status indicator light positioning device for complex electrical instrument equipment, comprising:

[0053] An acquisition module, the acquisition module is used to acquire data, the data including real-time collected image data of status indicator lights of electrical instrument equipment;

[0054] A preprocessing module, configured to preprocess the data to obtain preprocessed data;

[0055] a detection module, the detection module being configured to input the preprocessed data into a status indicator light detection model based on a deep neural network, and output location information of each indicator light; the status indicator light detection model based on the deep neural network being configured to perform feature extraction on the preprocessed data to obtain image features at different scales, perform feature fusion on the image features at different scales to obtain a feature map, and process the feature map based on the feature map using a region proposal network and a binary classifier;

[0056] An analysis module is used to output precise positioning information of each indicator light based on spatial proximity analysis according to the positioning information of each indicator light.

[0057] The beneficial effects of the present invention are:

[0058] This paper proposes a method and device for locating status indicators on complex electrical instrumentation equipment, based on a comprehensive algorithmic framework using deep learning and spatial proximity matrices. This framework employs a deep neural network-based status indicator detection model and incorporates a feature pyramid network to enhance the model's ability to detect indicators at multiple scales. To address the localization challenges in multi-indicator scenarios, the algorithm incorporates an optimization strategy based on a spatial proximity matrix. This strategy leverages the spatial distribution of indicators to achieve precise positioning through matrix matching, effectively reducing false detections and missed detections. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a framework flow chart of a method for locating status indicator lights of complex electrical instrument equipment in an embodiment of the present application;

[0060] Figure 2This is a data processing flow chart of a status indicator light detection model based on a deep neural network in an embodiment of the present application;

[0061] Figure 3 This is a flow chart of the method for spatial proximity analysis in an embodiment of the present application. DETAILED DESCRIPTION

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the embodiments of the present invention. It should be understood that the described embodiments are only a portion of the embodiments of the present invention, not all of them. Generally, the components of the embodiments of the present invention described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations.

[0063] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0064] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not require further definition or explanation in subsequent drawings.

[0065] Furthermore, the terms “first”, “second”, etc. are merely used for distinguishing descriptions and should not be understood as indicating or implying relative importance.

[0066] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0067] like Figure 1 As shown, this embodiment provides a method for locating status indicator lights of complex electrical instrument equipment, including:

[0068] Data collection and preprocessing steps: Get data The data includes real-time image data of status indicators of electrical instrument equipment, which is pre-processed to obtain pre-processed data. In this embodiment, the monitoring device is moved to a designated target location via a slide system to capture real-time images of the device's status. These captured images undergo rigorous pre-processing steps to eliminate noise and adjust image quality, thereby improving the recognition accuracy and robustness of subsequent models in complex environments.

[0069] The data acquisition includes:

[0070] Real-time acquisition of image data of a first electrical instrument device status indicator light, wherein the image data of the first electrical instrument device status indicator light is a visible light image;

[0071] collecting image data of a second electrical instrument device status indicator light in real time, wherein the image data of the second electrical instrument device status indicator light is an infrared image;

[0072] The data is obtained by fusing the first electrical instrument equipment status indicator light image data and the second electrical instrument equipment status indicator light image data. This invention utilizes multimodal data fusion technology, combining the complementarity of visible light and infrared images, to significantly improve the robustness of status indicator light detection, enabling the algorithm to maintain high detection accuracy under varying lighting conditions and complex backgrounds.

[0073] The data is preprocessed to obtain preprocessed data, including:

[0074] Performing data cleaning on the data to obtain cleaned data. The data cleaning stage mainly filters the image quality and removes abnormal data caused by occlusion, blur or excessive noise;

[0075] Image enhancement processing is performed on the cleaned data to obtain preprocessed data, and brightness adjustment, contrast enhancement, and data amplification are applied to optimize image quality.

[0076] After data cleaning and image enhancement, the processed data I pre .

[0077] Indicator light detection and positioning step: input the pre-processed data into a status indicator light detection model based on a deep neural network, and output the positioning information of each indicator light. The status indicator light detection model based on a deep neural network is used to extract features from the pre-processed data to obtain image features of different scales, and perform feature fusion on the image features of different scales to obtain a feature map. The feature map is processed based on a region proposal network and a binary classifier to obtain the positioning information of the indicator light. Based on the positioning information of each indicator light, based on spatial proximity analysis, the precise positioning information of each indicator light is output. The area of ​​the indicator light is automatically extracted from the device image using a detection algorithm based on deep learning. The detected indicator light is then accurately positioned using a positioning algorithm based on a spatial proximity matrix to ensure that the detected indicator light can be accurately mapped to a specific location on the device to avoid false detection and missed detection.

[0078] The pre-processed data is input into a status indicator light detection model based on a deep neural network, and the location information of each indicator light is output, including:

[0079] Performing feature extraction on the preprocessed data to obtain a multi-layer feature map;

[0080] Capture image features at different scales based on multi-layer feature maps;

[0081] The image features of different scales are fused based on a feature pyramid network to obtain a feature map;

[0082] Generate an anchor point on each of the feature maps based on a region proposal network through a preset sliding window;

[0083] Scoring each anchor point through a binary classifier to obtain anchor box parameters, and predicting the anchor box offset and scaling factor;

[0084] Generate a candidate region frame on the feature map according to the anchor frame parameters, the offset, and the scaling factor;

[0085] Determine whether the candidate area frame contains an indicator light according to the binary classifier, and calculate the category probability distribution of each candidate area frame;

[0086] According to the category probability distribution, the indicator light category corresponding to the category probability distribution with the maximum value in the category set is taken as the candidate area frame category, and the candidate area frame containing the indicator light is optimized based on the regression head to obtain the positioning information of each indicator light.

[0087] Specifically:

[0088] The input image is , H, W, C represent the height, width, and number of channels of the image respectively. After feature extraction, a multi-layer feature map is obtained. , the resolution of each layer of feature maps decreases layer by layer with the depth of the network, thereby capturing image features of different scales.

[0089] Through the top-down feature fusion strategy of the feature pyramid network, low-level detail information is introduced by upsampling on the high-level semantic features to obtain the fused feature map , the fusion formula is:

[0090]

[0091] The Region Proposal Network (RPN) generates anchors on each feature map through a sliding window and uses a binary classifier to score each anchor to obtain the anchor box parameters. , generate candidate region boxes on the feature map The regression formula for calculating the candidate box is:

[0092]

[0093] The binary classifier determines whether the candidate area contains the indicator light and calculates the category probability distribution of each area. The regression head further optimizes the area box to make it closer to the actual target and finally outputs a set of accurately located detection boxes B.

[0094] For the n indicator lights in the image, their two-dimensional coordinates are expressed as ,in The distance d between indicator lights i and j ij It can be expressed as:

[0095]

[0096] Then construct the spatial proximity matrix D:

[0097]

[0098] The same method is used to construct the predicted spatial proximity matrix based on the detected indicator light coordinates , calculate D and The expanded vectors A and B are used to calculate the cosine similarity ρ. The closer ρ is to 1, the more consistent the detected indicator light layout is with the known indicator lights on the device. The formula for calculating cosine similarity is:

[0099]

[0100] Using matrix dimensionality reduction and similarity optimization methods, redundant or incorrect detection items in the matrix are removed to make the two matrices have the same dimensions. First, some indicator light information is gradually removed from the matrix with higher dimensions until the prediction matrix The dimension of the matrix D is matched with the original matrix D, and the cosine similarity is calculated for all possible dimensionality reduction combinations, and the group with the highest similarity is selected for matching. The precise positioning result of the indicator light is obtained.

[0101] Status reporting: A preset status mapping table is used, including the indicator light number, the actual two-dimensional coordinates of the indicator light, the indicator light status flag, and the device status description corresponding to the indicator light state. The system obtains the status information and location information of all indicator lights. The status information is the device status image and the status flag, and the location information is the precise location of each indicator light. This status information and location information are converted into readable device status according to the preset status mapping table. The system obtains the status information of each indicator light through identification and location, and integrates this information with the overall device status to form a complete status report.

[0102] According to the preset state mapping table Integrate the status information S and positioning information B of all indicators and convert them into readable device status R. The status mapping formula is:

[0103]

[0104] like Figure 2 As shown in the figure, the data processing flow of the status indicator light detection model based on deep neural network includes:

[0105] The preprocessed data first enters the basic feature extraction module and then passes through four residual blocks in sequence. The image outputs from the four residual blocks have the following dimensions: height / 4, width / 4; height / 8, width / 8; height / 16, width / 16; and height / 32, width / 32. The output of each residual block is fed into the corresponding first convolutional layer, with a 1x1 kernel. The output of each first convolutional layer is then upsampled by a factor of 2 and fed into the corresponding second convolutional layer, with a 3x3 kernel. This generates a feature map containing image features at different scales output from each second convolutional layer. These features at different scales are then fused using a feature pyramid network to generate a feature map. A convolutional neural network is used for feature extraction, capturing high-level image features layer by layer through multiple residual blocks, ensuring that the distinctive features of the indicator light can be separated from the complex background. The feature pyramid network further enhances the detection capabilities of indicator lights at multiple scales, utilizing feature maps at different scales to ensure accurate detection at both high and low resolutions.

[0106] The feature map is input into the region proposal network, where a 3x3 convolution kernel is applied to each location in the feature map. This convolution kernel is further decomposed into two 1x1 convolution kernels to handle different tasks in parallel. The output of the first 1x1 convolution kernel undergoes tensor deformation, softmax activation, and tensor deformation again to complete the category prediction. The second 1x1 convolution kernel focuses on bounding box regression, calculating the bounding box coordinates of each candidate region to obtain the precise location of the candidate region. The outputs of these two convolution kernels are aggregated to obtain the final object region proposal.

[0107] The feature map and the target region proposal are jointly input into the region of interest pooling module, and then the output of the region of interest pooling module is input into the fully connected layer. After the fully connected layer, the binary classifier classifies each candidate box, and the regression head predicts the precise boundary of the candidate box to obtain the predicted category and predicted bounding box.

[0108] like Figure 3 As shown in Figure 2, the spatial proximity analysis process includes:

[0109] Dimension matching: Reduce the higher dimensions of the original matrix and the predicted matrix to match the dimension information;

[0110] Similarity optimization: Calculate the similarity of the matrix after dimension matching and continuously select the better results;

[0111] Best match: Get the result with the highest similarity from all the results, and then get the best match between the predicted matrix and the original matrix.

[0112] Another aspect of the present invention provides a status indicator light positioning device for complex electrical instrument equipment, comprising:

[0113] An acquisition module, the acquisition module is used to acquire data, the data including real-time collected image data of status indicator lights of electrical instrument equipment;

[0114] A preprocessing module, configured to preprocess the data to obtain preprocessed data;

[0115] a detection module, the detection module being configured to input the preprocessed data into a status indicator light detection model based on a deep neural network, and output location information of each indicator light; the status indicator light detection model based on the deep neural network being configured to perform feature extraction on the preprocessed data to obtain image features at different scales, perform feature fusion on the image features at different scales to obtain a feature map, and process the feature map based on the feature map using a region proposal network and a binary classifier;

[0116] An analysis module is used to output precise positioning information of each indicator light based on spatial proximity analysis according to the positioning information of each indicator light.

[0117] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for locating status indicator lights of complex electrical instrument equipment, characterized in that: include: Acquiring data, the data including real-time collected image data of status indicator lights of electrical instrument equipment; Preprocessing the data to obtain preprocessed data; Inputting the preprocessed data into a status indicator light detection model based on a deep neural network, outputting the positioning information of each indicator light, the status indicator light detection model based on the deep neural network is used to extract features from the preprocessed data to obtain image features of different scales, performing feature fusion on the image features of different scales to obtain a feature map, and processing the feature map based on the region proposal network and the binary classifier to obtain the positioning information of the indicator light; Outputting precise positioning information of each indicator light based on spatial proximity analysis according to the positioning information of each indicator light; The pre-processed data is input into a status indicator light detection model based on a deep neural network, and the location information of each indicator light is output, including: Performing feature extraction on the preprocessed data to obtain a multi-layer feature map; Capture image features at different scales based on multi-layer feature maps; The image features of different scales are fused based on a feature pyramid network to obtain a feature map; Generate an anchor point on each of the feature maps based on a region proposal network through a preset sliding window; Scoring each anchor point through a binary classifier to obtain anchor box parameters, and predicting the anchor box offset and scaling factor; Generating a candidate region frame on the feature map according to the anchor frame parameters, offset and scaling factor; Determine whether the candidate area frame contains an indicator light according to the binary classifier, and calculate the category probability distribution of each candidate area frame; According to the category probability distribution, the indicator light category corresponding to the maximum category probability distribution in the category set is selected as the candidate region frame category, and the candidate region frame containing the indicator light is optimized based on the regression head to obtain the positioning information of each indicator light; Outputting precise positioning information of each indicator light based on spatial proximity analysis according to the positioning information of each indicator light includes: Obtain the two-dimensional coordinates of each indicator light according to the positioning information of each indicator light; Constructing a spatial proximity matrix according to the two-dimensional coordinates of each of the indicator lights, and calculating an expansion vector of the spatial proximity matrix; Constructing a prediction space proximity matrix according to the two-dimensional coordinates of each of the indicator lights, and calculating an expansion vector of the prediction space proximity matrix; Calculating cosine similarity based on the expansion vector of the spatial proximity matrix and the expansion vector of the predicted spatial proximity matrix; Eliminating redundant or erroneous detection items in the spatial proximity matrix and the predicted spatial proximity matrix according to the cosine similarity-based matrix dimensionality reduction and similarity optimization method to obtain accurate positioning results of each indicator light; Eliminating redundant or erroneous detection items in the spatial proximity matrix and the predicted spatial proximity matrix, including: First, gradually remove indicator light information from a matrix with a higher dimension between the spatial proximity matrix and the predicted spatial proximity matrix until the matrix dimensions of the predicted spatial proximity matrix match those of the spatial proximity matrix; The number of indicator light information removed during the dimensionality reduction process is fixed, but different indicator light information can be selected for removal, and the cosine similarity is calculated for all possible dimensionality reduction combinations; Select the group with the highest cosine similarity for matching, construct a mapping between the predicted coordinates of the indicator light and the actual coordinates of the indicator light, and obtain the precise positioning result of each indicator light; For the n indicator lights in the image, their two-dimensional coordinates are expressed as , where i=1,2,...,n, the distance between indicator light i and indicator light j It can be expressed as: ， Then construct the spatial proximity matrix D: ， The same method is used to construct the predicted spatial proximity matrix based on the detected indicator light coordinates , calculate D and The expanded vectors A and B are used to calculate the cosine similarity ρ. The closer ρ is to 1, the more consistent the detected indicator light layout is with the known indicator lights on the device.

2. A method for locating status indicator lights of complex electrical instrument equipment according to claim 1, characterized in that: Get data, including: Real-time acquisition of image data of a first electrical instrument device status indicator light, wherein the image data of the first electrical instrument device status indicator light is a visible light image; collecting image data of a second electrical instrument device status indicator light in real time, wherein the image data of the second electrical instrument device status indicator light is an infrared image; The data is obtained by fusing the first electrical instrument equipment status indicator light image data and the second electrical instrument equipment status indicator light image data.

3. A method for locating status indicator lights of complex electrical instrument equipment according to claim 1 or 2, characterized in that: Preprocessing the data to obtain preprocessed data includes: performing data cleaning on the data to obtain cleaned data; Perform image enhancement processing on the cleaned data to obtain preprocessed data.

4. The method for locating status indicator lights of complex electrical instrument equipment according to claim 1, characterized in that: The calculation formula for generating the candidate region frame on the feature map is as follows: ； The anchor box parameters are , Respectively represent the horizontal coordinate, vertical coordinate, width and height of the center point of the anchor point frame. The offset of the anchor point frame is , They respectively represent the horizontal offset and vertical offset of the center point of the candidate box relative to the center point of the anchor box, and the scaling factors of the width and height of the candidate box relative to the width and height of the anchor box. The candidate area box is represented as (x, y, w, h), where x, y, w, and h represent the horizontal coordinate of the center point of the candidate box, the vertical coordinate of the center point, the width, and the height, respectively.

5. The method for locating status indicator lights of complex electrical instrument equipment according to claim 1, characterized in that: The calculation formula for calculating the cosine similarity based on the expansion vector of the spatial proximity matrix and the expansion vector of the predicted spatial proximity matrix is: ， Where A represents the expansion vector of the spatial proximity matrix, B represents the expansion vector of the predicted spatial proximity matrix, m is the total length of the vector, that is, the total number of indicator light pairs, k is the vector index, and A k Indicates the actual distance between the kth pair of indicator lights, B k represents the predicted distance between the kth pair of indicator lights, and ρ(A, B) represents the cosine similarity.

6. The method for locating status indicator lights of complex electrical instrument equipment according to claim 1, characterized in that: Also includes: A preset state mapping table, wherein the state mapping table includes an indicator light number, the actual two-dimensional coordinates of the indicator light, an indicator light state flag, and a device state description corresponding to the indicator light state; Obtaining status information and positioning information of all indicator lights, wherein the status information is image data of the status indicator lights of the electrical instrument equipment and the status flags, and the positioning information is the precise positioning result of each indicator light; The state information and the positioning information are converted into a readable device state according to the preset state mapping table.

7. A status indicator light positioning device for complex electrical instrument equipment, characterized in that: include: An acquisition module, the acquisition module is used to acquire data, the data including real-time collected image data of status indicator lights of electrical instrument equipment; A preprocessing module, configured to preprocess the data to obtain preprocessed data; a detection module, the detection module being configured to input the preprocessed data into a status indicator light detection model based on a deep neural network, and output location information of each indicator light; the status indicator light detection model based on the deep neural network being configured to perform feature extraction on the preprocessed data to obtain image features at different scales, perform feature fusion on the image features at different scales to obtain a feature map, and process the feature map based on the feature map using a region proposal network and a binary classifier; An analysis module configured to output precise positioning information of each indicator light based on spatial proximity analysis according to the positioning information of each indicator light; The pre-processed data is input into a status indicator light detection model based on a deep neural network, and the location information of each indicator light is output, including: Performing feature extraction on the preprocessed data to obtain a multi-layer feature map; Capture image features at different scales based on multi-layer feature maps; The image features of different scales are fused based on a feature pyramid network to obtain a feature map; Generate an anchor point on each of the feature maps based on a region proposal network through a preset sliding window; Scoring each anchor point through a binary classifier to obtain anchor box parameters, and predicting the anchor box offset and scaling factor; Generating a candidate region frame on the feature map according to the anchor frame parameters, offset and scaling factor; Determine whether the candidate area frame contains an indicator light according to the binary classifier, and calculate the category probability distribution of each candidate area frame; According to the category probability distribution, the indicator light category corresponding to the maximum category probability distribution in the category set is selected as the candidate region frame category, and the candidate region frame containing the indicator light is optimized based on the regression head to obtain the positioning information of each indicator light; Outputting precise positioning information of each indicator light based on spatial proximity analysis according to the positioning information of each indicator light includes: Obtain the two-dimensional coordinates of each indicator light according to the positioning information of each indicator light; Constructing a spatial proximity matrix according to the two-dimensional coordinates of each of the indicator lights, and calculating an expansion vector of the spatial proximity matrix; Constructing a prediction space proximity matrix according to the two-dimensional coordinates of each of the indicator lights, and calculating an expansion vector of the prediction space proximity matrix; Calculating cosine similarity based on the expansion vector of the spatial proximity matrix and the expansion vector of the predicted spatial proximity matrix; Eliminating redundant or erroneous detection items in the spatial proximity matrix and the predicted spatial proximity matrix according to the cosine similarity-based matrix dimensionality reduction and similarity optimization method to obtain accurate positioning results of each indicator light; Eliminating redundant or erroneous detection items in the spatial proximity matrix and the predicted spatial proximity matrix, including: First, gradually remove indicator light information from a matrix with a higher dimension between the spatial proximity matrix and the predicted spatial proximity matrix until the matrix dimensions of the predicted spatial proximity matrix match those of the spatial proximity matrix; The number of indicator light information removed during the dimensionality reduction process is fixed, but different indicator light information can be selected for removal, and the cosine similarity is calculated for all possible dimensionality reduction combinations; Select the group with the highest cosine similarity for matching, construct a mapping between the predicted coordinates of the indicator light and the actual coordinates of the indicator light, and obtain the precise positioning result of each indicator light; For the n indicator lights in the image, their two-dimensional coordinates are expressed as , where i=1,2,...,n, the distance between indicator light i and indicator light j It can be expressed as: ， Then construct the spatial proximity matrix D: ， The same method is used to construct the predicted spatial proximity matrix based on the detected indicator light coordinates , calculate D and The expanded vectors A and B are used to calculate the cosine similarity ρ. The closer ρ is to 1, the more consistent the detected indicator light layout is with the known indicator lights on the device.

Images