Movable monitoring equipment and method for monitoring power supply guaranteeing equipment

Abstract

The invention provides movable monitoring equipment and a power supply guaranteeing equipment monitoring method, the movable monitoring equipment comprises a movable device and a power supply guaranteeing equipment monitoring device, the monitoring device is composed of a central processing unit, an infrared image acquisition unit, an image processing unit and a temperature identification unit, and the temperature identification unit identifies the temperature of the power supply guaranteeing equipment through acquiring an infrared image. Judging whether the image is fuzzy or not by using a fuzzy detection model, if so, processing the image, and acquiring equipment temperature information through a temperature identification model; and if the image is not fuzzy, directly carrying out temperature identification. The method improves the monitoring efficiency and accuracy, and is suitable for monitoring the operation state of the power supply equipment in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment monitoring, and in particular to a movable monitoring device and a method for monitoring power supply equipment. Background Art

[0002] With the continuous increase in electricity demand, the stable operation of power supply equipment has become increasingly important. The normal and stable operation of power supply equipment affects the safe and stable operation of the power system. Temperature is an important factor affecting the normal and stable operation of power supply equipment, so it is necessary to monitor the operating temperature of the power supply equipment in real time. Existing monitoring equipment for power supply equipment mainly relies on sensors, that is, sensors are installed near or on the power supply equipment, and the temperature information of the power supply equipment is collected by the sensors, so as to monitor the power supply equipment through the collected temperature information. However, the sensor can only be applied to simple environmental conditions. For extreme environmental conditions, the working performance of the sensor is unstable, resulting in inaccurate temperature information collected, which leads to low monitoring accuracy of the power supply equipment. Summary of the invention

[0003] The object of the present invention is to provide a mobile monitoring device and a method for monitoring power supply equipment to solve the technical problem that the working performance of the sensor is unstable, resulting in inaccurate collected temperature information.

[0004] To achieve the purpose of the present invention, according to a first aspect of the present invention, there is provided a mobile monitoring device based on artificial intelligence, comprising a mobile device and a power supply equipment monitoring device, wherein the mobile device is connected to the power supply equipment monitoring device; the power supply equipment monitoring device comprises a central processing unit, an infrared image acquisition unit, an image blurring unit, an image processing unit and a temperature recognition unit;

[0005] The infrared image acquisition unit is used to acquire an infrared image of a target power supply equipment; the target power supply equipment is any power supply equipment in the power system;

[0006] The image processing unit is used to perform image blur detection on the infrared image based on a blur detection model, and output a blur detection result of the infrared image;

[0007] The central processing unit is used for:

[0008] If the blur detection result is that the infrared image is a blurry image, the image processing unit is controlled to perform image processing on the infrared image to obtain a target recognition image of the infrared image; the temperature recognition unit is controlled to recognize the target recognition image through a temperature recognition model to obtain temperature information of the target power supply equipment; the image processing process includes grayscale processing, resolution reconstruction processing and enhancement processing;

[0009] If the blur detection result is that the infrared image is a non-blurred image, the temperature recognition unit is controlled to recognize the infrared image through the temperature recognition model to obtain the temperature information of the target power supply equipment.

[0010] According to a second aspect of the present invention, a method for monitoring power supply equipment is provided, which is applied to the artificial intelligence-based mobile monitoring device described in the first aspect, and the method comprises:

[0011] Collecting an infrared image of the target power supply equipment; inputting the infrared image into a fuzzy detection model to perform image fuzzy detection, and obtaining a fuzzy detection result output by the fuzzy detection model;

[0012] If the blur detection result is that the infrared image is a blurry image, the infrared image is processed to obtain a target recognition image of the infrared image; the target recognition image is input into a temperature recognition model for recognition to obtain the temperature information of the target power supply equipment output by the temperature recognition model; the image processing process includes grayscale processing, resolution reconstruction processing and enhancement processing;

[0013] If the blur detection result is that the infrared image is a non-blurred image, the infrared image is input into the temperature recognition model for recognition, and the temperature information of the target power supply equipment output by the temperature recognition model is obtained.

[0014] The present invention provides a mobile monitoring device and a power supply equipment monitoring method, which acquires infrared images of target power supply equipment, identifies the infrared images, and obtains the temperature information of the target power supply equipment. First, since the environment has little influence on the image acquisition process, the detailed features of the acquired infrared images are completely preserved regardless of whether it is a simple environment or an extreme environment, so the temperature information of the target power supply equipment can be accurately identified. Even in extreme environments, the infrared images processed by image blur detection, grayscale processing, resolution reconstruction processing, and enhancement processing can restore the detailed features of the original infrared images, and can accurately identify the temperature information of the target power supply equipment. At the same time, in the whole process, the artificial intelligence models of the fuzzy detection model and the temperature recognition model are also used, thereby improving the monitoring accuracy of the power supply equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] 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 of the present invention. For ordinary technicians in this field, without paying creative labor, other drawings obtained based on these drawings still belong to the scope of the present invention.

[0016] Figure 1 is a structural diagram of a mobile monitoring device based on artificial intelligence provided by an embodiment of the present invention;

[0017] Figure 2 It is a flow chart of a method for monitoring power supply equipment provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0018] The detailed description of the drawings is intended as an illustration of the current embodiment of the present application, and is not intended to represent the only form in which the present application can be implemented. It should be understood that the same or equivalent functions can be accomplished by different embodiments intended to be included in the spirit and scope of the present application.

[0019] See also Figure 1 , Figure 1 A structural diagram of an artificial intelligence-based mobile monitoring device provided in an embodiment of the present invention, wherein the artificial intelligence-based mobile monitoring device includes a movable device and a power supply equipment monitoring device, wherein the movable device is connected to the power supply equipment monitoring device, wherein the movable device is such as wheels, universal wheels, servo motors, etc., therefore, the entire power supply equipment monitoring device can be controlled by the movable device to move on the power supply equipment.

[0020] Furthermore, the power supply equipment monitoring device in the embodiment of the present invention includes a central processing unit (such as a CPU), an infrared image acquisition unit (such as an infrared image acquisition camera), an image blurring unit, an image processing unit and a temperature recognition unit.

[0021] Furthermore, the infrared image acquisition unit acquires an infrared image of the target power supply protection equipment, wherein the target power supply protection equipment is any power supply protection equipment in the power system.

[0022] Furthermore, the image processing unit inputs the infrared image into the blur detection model, performs image blur detection on the infrared image through the blur detection model, and outputs the blur detection result of the infrared image, wherein the blur detection result may be that the infrared image is a blurred image or that the infrared image is a non-blurred image.

[0023] Furthermore, if the blur detection result is that the infrared image is a blurred image, the central processing unit controls the image processing unit to perform image processing on the infrared image to obtain a target recognition image of the infrared image.

[0024] Furthermore, the central processing unit controls the temperature recognition unit to input the target recognition image into the temperature recognition model, and the target recognition image is recognized by the temperature recognition model to obtain the temperature information of the target power supply equipment. Among them, the image processing process in the embodiment of the present invention includes grayscale processing, resolution reconstruction processing and enhancement processing.

[0025] Furthermore, if the blur detection result shows that the infrared image is a non-blurred image, the central processing unit controls the temperature recognition unit to input the infrared image into the temperature recognition model, recognizes the infrared image through the temperature recognition model, and obtains the temperature information of the target power supply equipment.

[0026] The artificial intelligence-based mobile monitoring device provided by the present invention collects infrared images of target power supply equipment, identifies the infrared images, and obtains the temperature information of the target power supply equipment. First, since the environment has little influence on the image acquisition process, the detailed features of the collected infrared images are retained intact regardless of whether it is a simple environment or an extreme environment, so the temperature information of the target power supply equipment can be accurately identified. Even in the face of extreme environments, the infrared image processed by image blur detection, grayscale processing, resolution reconstruction processing and enhancement processing can restore the detailed features of the original infrared image, and can accurately identify the temperature information of the target power supply equipment. At the same time, in the whole process, the artificial intelligence model of the fuzzy detection model and the temperature recognition model is also used, thereby improving the monitoring accuracy of the power supply equipment.

[0027] In one embodiment, the power supply equipment monitoring device further includes an environment monitoring unit, an image compression unit, and a data communication unit. Among them, the image compression unit compresses the infrared image to obtain compressed information. The environment monitoring unit collects environmental information of the environment in which the target power supply equipment is located. The central processing unit determines the data transmission method according to the environmental information, and transmits the compressed information and temperature information to the user terminal through the data communication unit in the form of data transmission. After receiving the compressed information and temperature information, the user terminal decompresses the compressed information to obtain an infrared image, and further performs data verification based on the infrared image and temperature information.

[0028] In the embodiment of the present invention, on the one hand, the compressed information occupies less communication resources and has a faster transmission speed. On the other hand, the compressed information and temperature information are transmitted to the user terminal in different data transmission modes according to different environmental information, thereby reducing the interference of environmental factors on data communication. Therefore, the user terminal can quickly and accurately receive the infrared image and temperature information sent by the mobile monitoring device, thereby improving the speed of data verification.

[0029] Reference Figure 2 , Figure 2It is a flow chart of the power supply equipment monitoring method provided by the present invention. The execution subject of the power supply equipment monitoring method provided by the present invention is a mobile monitoring device based on artificial intelligence, including a mobile device and a power supply equipment monitoring device. The power supply equipment monitoring device includes a central processing unit, an infrared image acquisition unit, an image blurring unit, an image processing unit and a temperature recognition unit. Therefore, the power supply equipment monitoring method includes:

[0030] Step 10: Collect infrared images of the target power supply equipment.

[0031] Optionally, the infrared image acquisition unit acquires an infrared image of any target power supply equipment in the power system, and the infrared image acquisition unit is such as an infrared camera.

[0032] Step 20, inputting the infrared image into a fuzzy detection model to perform image fuzzy detection, and obtaining a fuzzy detection result output by the fuzzy detection model.

[0033] Further, the infrared image is input into the fuzzy detection model, and the fuzzy detection model is used to perform image fuzziness detection on the infrared image to obtain the fuzzy detection result output by the fuzzy detection model, wherein the fuzzy detection result may be that the infrared image is a fuzzy image or that the infrared image is a non-fuzzy image. The fuzzy detection model is trained based on the sample image and the low-resolution image model corresponding to the sample image. The specific process is: the infrared image is input into the fuzzy detection model to obtain the fuzzy confidence of the infrared image output by the fuzzy detection model. Further, the fuzzy confidence of the infrared image is compared with the preset threshold, wherein the preset threshold is set according to actual conditions. If the comparison result is that the fuzzy confidence of the infrared image is greater than the preset threshold, the infrared image is determined to be a fuzzy image. If the comparison result is that the fuzzy confidence of the infrared image is less than or equal to the preset threshold, the infrared image is determined to be a non-fuzzy image.

[0034] Step 30, if the blur detection result is that the infrared image is a blurry image, image processing is performed on the infrared image to obtain a target recognition image of the infrared image; the target recognition image is input into the temperature recognition model for recognition, and the temperature information of the target power supply equipment output by the temperature recognition model is obtained.

[0035] Furthermore, if it is determined that the blur detection result is that the infrared image is a blurry image, image processing is performed on the infrared image to obtain a target recognition image of the infrared image, wherein the image processing process includes grayscale processing, resolution reconstruction processing and enhancement processing.

[0036] Furthermore, the target recognition image is input into the temperature recognition model, and the target recognition image is recognized by the temperature recognition model to obtain the temperature information of the target power supply equipment output by the temperature recognition model. Among them, the temperature recognition model in the embodiment of the present invention is obtained by training a preset neural network based on the sample image and the temperature label corresponding to the sample image. The preset neural network model can be a feedforward neural network (FNN), a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory network (LSTM), etc.

[0037] Step 40: If the fuzzy detection result shows that the infrared image is a non-fuzzy image, the infrared image is input into a temperature recognition model for recognition, and the temperature information of the target power supply equipment output by the temperature recognition model is obtained.

[0038] Furthermore, if the blur detection result is determined to be a non-blurred image, that is, the infrared image can be identified without processing, the infrared image is input into the temperature recognition model, and the infrared image is identified by the temperature recognition model to obtain the temperature information of the target power supply equipment output by the temperature recognition model.

[0039] The embodiment of the present invention collects infrared images of target power supply equipment, identifies the infrared images, and obtains the temperature information of the target power supply equipment. Since the environment has little influence on the image acquisition process, the detailed features of the collected infrared images are retained intact regardless of whether it is a simple environment or an extreme environment, so the temperature information of the target power supply equipment can be accurately identified. Even in extreme environments, the infrared images processed by image blur detection, grayscale processing, resolution reconstruction processing and enhancement processing can restore the detailed features of the original infrared images, and can accurately identify the temperature information of the target power supply equipment. At the same time, in the whole process, the artificial intelligence models of the fuzzy detection model and the temperature recognition model are also used, thereby improving the monitoring accuracy of the power supply equipment.

[0040] In one embodiment, the blur detection model includes a pixel extraction function, a pixel quantization function and a blur detection function. Therefore, step 20 inputs the infrared image into the blur detection model to perform image blur detection, and obtains a blur detection result output by the blur detection model, including:

[0041] Step 201, inputting the infrared image into the blur detection model, performing pixel extraction on the infrared image based on the pixel extraction function, and obtaining the pixel value of each row of pixels and the pixel value of each column of pixels in the infrared image;

[0042] Step 202, performing pixel quantization on the pixel values ​​of each row of pixels and the pixel values ​​of each column of pixels of the infrared image based on a pixel quantization function to obtain pixel quantization values ​​of the infrared image;

[0043] Step 203 , performing fuzzy detection based on the fuzzy detection function and the pixel quantization value of the infrared image to obtain a fuzzy detection result.

[0044] Optionally, the infrared image is input into the blur detection model, and the infrared image is pixel extracted by a pixel extraction function to obtain the pixel value of each row of pixels and the pixel value of each column of pixels in the infrared image. The pixel extraction function is as follows:

[0045]

[0046] Among them, P i Represents the pixel value of each row of pixels, P j Represents the pixel value of each column of pixels, M i represents the number of pixels in the i-th row, Indicates the Mth row in the i-th row i The pixel value of pixels, N j represents the number of pixels in the jth column, p jN Indicates the jth column, Nth j The pixel value of pixels.

[0047] Furthermore, the pixel values ​​of each row of pixels and each column of pixels in the infrared image are quantized by a pixel quantization function to obtain the pixel quantization value of the infrared image. The pixel quantization function is as follows:

[0048]

[0049] Among them, Q ij Represents the quantized value of the pixel in the i-th row and j-th column in the infrared image.

[0050] Furthermore, fuzzy detection is performed according to the fuzzy detection function combined with the pixel quantization value of the infrared image to obtain a fuzzy detection result. The fuzzy detection function is specifically as follows:

[0051]

[0052] Among them, Result represents the blur detection result, 1 represents a blurry image, 0 represents a non-blurred image, w[Q ij ≤a] represents the pixel quantization value Q ij The number of pixels is less than or equal to the preset pixel value a, and b is the preset number.

[0053] The embodiment of the present invention can quickly determine whether an infrared image is a blurred image through a fuzzy detection model, thereby determining a temperature recognition strategy for the infrared image, and then obtaining temperature information with different recognition strategies, thereby improving the monitoring accuracy of power supply equipment.

[0054] In one embodiment, image processing is performed on an infrared image to obtain a target recognition image of the infrared image, including: step 301 performs grayscale processing on the infrared image to obtain a grayscale image; step 302 performs resolution reconstruction processing on the grayscale image to obtain a resolution reconstructed image; step 303 performs enhancement processing on the resolution reconstructed image to obtain a target recognition image.

[0055] In one embodiment, grayscale processing is performed on the infrared image to obtain a grayscale image, including:

[0056] Step 3011, grayscale processing is performed on the infrared image to obtain the pixel grayscale value of each pixel in the infrared image;

[0057] Step 3012, determining a pixel grayscale mean based on a first preset number of first pixel grayscale values ​​with the largest value and a second preset number of second pixel grayscale values ​​with the smallest value;

[0058] Step 3013, adjusting the pixel grayscale value of each pixel in the infrared image based on the pixel grayscale mean value to obtain a final pixel grayscale value of each pixel in the infrared image.

[0059] Optionally, the infrared image is gray-scaled to obtain the pixel gray-scale value of each pixel in the infrared image. Further, the pixel gray-scale value of each pixel is sorted in descending order to obtain the sorted pixel gray-scale value.

[0060] Furthermore, a first preset number of first pixel grayscale values ​​with the largest value and a second preset number of second pixel grayscale values ​​with the smallest value are obtained, wherein the first preset number and the second preset number are set according to actual conditions. Generally, the first preset number and the second preset number are set to 5, so it can be understood that the first 5 largest first pixel grayscale values ​​and the last 5 smallest second pixel grayscale values ​​are obtained.

[0061] Furthermore, the grayscale mean is calculated based on the grayscale values ​​of the first pixel with the largest value in the first preset number and the grayscale values ​​of the second pixel with the smallest value in the second preset number to determine the pixel grayscale mean. The specific calculation formula of the pixel grayscale mean is as follows:

[0062]

[0063] Among them, G average Represents the pixel grayscale mean, max m represents the first preset quantity, Represents the gray value of the first pixel, Indicates the max m The first pixel gray value, Indicates the gray value of the first and second pixels, minn Indicates the second preset quantity, Indicates the min n The second pixel gray value.

[0064] Furthermore, the pixel grayscale value of each pixel in the infrared image is adjusted according to the pixel grayscale mean to obtain the final pixel grayscale value of each pixel in the infrared image, that is, for pixel grayscale values ​​less than the pixel grayscale mean, it is necessary to increase it and adjust it to be close to the pixel grayscale value, and for pixel grayscale values ​​greater than or equal to the pixel grayscale mean, it is necessary to reduce it and adjust it to be close to the pixel grayscale value, so that the overall pixel grayscale value is close to the pixel grayscale mean, so that the detail features in the image are more uniform, avoiding excessive differences in pixel grayscale values, resulting in the coverage of detail features, and resulting in final recognition errors. Therefore, the image generated according to the final pixel grayscale value of each pixel in the infrared image is the final grayscale image. Among them, the specific formula for grayscale value adjustment is as follows:

[0065]

[0066] Among them, G last Represents the final pixel grayscale value, G(x <G average ) represents the pixel gray value less than the pixel gray mean value, G(x≥G average ) represents the pixel grayscale value that is greater than or equal to the pixel grayscale mean value, L represents the grayscale level, and the grayscale level is generally 256.

[0067] The embodiment of the present invention performs grayscale processing on the infrared image. Even in extreme environments, the processed infrared image can restore the detailed features of the original infrared image, accurately identify the temperature information of the target power supply equipment, and improve the monitoring accuracy of the power supply equipment.

[0068] In one embodiment, the resolution reconstruction network used in the embodiment of the present invention includes an image segmentation layer, a feature extraction layer, a feature optimization layer and a feature fusion layer; the resolution reconstruction network is obtained based on the sample image and the super-resolution network training corresponding to the sample image. Therefore, the grayscale image is subjected to resolution reconstruction processing to obtain a resolution reconstructed image, including:

[0069] Step 3021, input the grayscale image to the image segmentation layer, segment the grayscale image through the image segmentation layer, and obtain multiple sub-images carrying numbering information output by the image segmentation layer.

[0070] Optionally, the grayscale image is input to the image segmentation layer, and the grayscale image is segmented by the image segmentation layer to obtain multiple sub-images with numbering information output by the image segmentation layer, wherein the numbering information is the position information of the sub-image in the grayscale image, so that the sub-images can be quickly fused according to the numbering information during subsequent image fusion. The size of the sub-image can be set according to the size of the original infrared image, such as the size of the infrared image is 256*256, and the size of the sub-image is 16*16, etc.

[0071] Step 3022: input the multiple sub-images into the feature extraction layer, perform feature extraction on the multiple sub-images through the feature extraction layer, and obtain image features of each sub-image output by the feature extraction layer.

[0072] Furthermore, multiple sub-images are input into the feature extraction layer, and features are extracted from each sub-image through the feature extraction layer to obtain image features of each sub-image output by the feature extraction layer, wherein the feature extraction layer includes an upsampling layer, an attention feature extraction layer, and a convolution fusion layer. Therefore, the specific process of feature extraction is as follows: multiple sub-images are input into the feature extraction layer, and each sub-image is upsampled through the upsampling layer to obtain upsampled features of each sub-image. Vector extraction is performed on each sub-image through the embedding vector model of the attention feature extraction layer to obtain the embedding vector of each sub-image, and feature extraction is performed on the embedding vector of each sub-image through the self-attention model of the attention feature extraction layer to obtain the attention features of each sub-image. Among them, the self-attention model can adopt the QKV (Query-Key-Value) mode. The formula of the self-attention model is expressed as:

[0073]

[0074] Among them, Attention(Q, K, V) represents the output of the self-attention model, V represents the input embedding vector, Q and K represent the feature vectors for calculating the attention weights, and K T represents the transpose of K, d represents the dimension of the input vector V, and softamx() represents the normalization function.

[0075] Furthermore, the convolutional fusion layer includes three convolutional layers, namely the first convolutional layer, the second convolutional layer and the third convolutional layer.

[0076] The first convolutional layer has 32 convolutional kernels, a kernel size of 3x3, a stride of 1, padding of Same (keeping the input and output sizes consistent), and an activation function of ReLU. The second convolutional layer has 64 convolutional kernels, a kernel size of 3x3, a stride of 1, padding of Same, and an activation function of ReLU.

[0077] Therefore, the upsampled features and attention features of each sub-image are convolved twice through the first convolutional layer to obtain two 32*32*32 features of each sub-image output by the first convolutional layer, and then the two 32*32*32 features of each sub-image are spliced ​​to obtain 32*32*64 features of each sub-image.

[0078] Furthermore, the upsampled features and attention features of each sub-image are convolved through the second convolutional layer to obtain 32*32*64 features of each sub-image output by the second convolutional layer.

[0079] Furthermore, the 32*32*64 features of each sub-image output by the first convolution layer and the 32*32*64 features of each sub-image output by the second convolution layer are weighted through the third convolution layer to obtain the image features of each sub-image.

[0080] Among them, the loss function of the third convolutional layer is:

[0081]

[0082] Among them, Loss represents the loss function of the third convolutional layer, φ CNN1 represents the first output of the first convolutional layer, φ CNN2 represents the second output of the first convolutional layer, φ CNN3 represents the output of the second convolutional layer, and α, β, and δ represent the preset output weights.

[0083] Step 3023, inputting the image features of each sub-image into the feature optimization layer, optimizing the image features of each sub-image through the feature optimization layer, and obtaining the optimized features of each sub-image output by the feature optimization layer.

[0084] Furthermore, the image features of each sub-image are input into the feature optimization layer, and the image features of each sub-image are optimized by the feature optimization layer to obtain the optimized features of each sub-image output by the feature optimization layer.

[0085] In one embodiment, the feature optimization layer includes a structure reconstruction layer, a texture reconstruction layer, and a sharpness reconstruction layer. Therefore, the process of optimizing the features is specifically as follows: the image features of each sub-image are structurally reconstructed by the structure reconstruction layer to obtain the structurally reconstructed features of each sub-image, wherein the structure reconstruction layer in the embodiment of the present invention includes 3 independent convolutional layers, and the loss function of the structure reconstruction layer can be expressed as:

[0086]

[0087] Where L1 represents the loss function of the structure reconstruction layer, φi(·) represents the output of the i-th independent convolutional layer of the structure reconstruction layer, λ represents the regularization coefficient, and w i represents the weight of the i-th independent convolutional layer, φi(X1) represents the output of the i-th independent convolutional layer after structural reconstruction and feature processing, represents the output of the image feature after processing by the i-th independent convolutional layer, X1 represents the feature after structural reconstruction, Represents image features.

[0088] Furthermore, the texture reconstruction layer is used to perform texture reconstruction on the structural reconstructed features of each sub-image to obtain the texture reconstructed features of each sub-image, wherein the texture reconstruction layer in the embodiment of the present invention includes 4 independent convolutional layers, and the loss function of the texture reconstruction layer can be expressed as:

[0089]

[0090] Among them, L2 represents the loss function of the texture reconstruction layer, ψ j (·) represents the output of the jth independent convolutional layer of the texture reconstruction layer, ψ j (X2) represents the output of the j-th independent convolutional layer after texture reconstruction and feature processing, ψ j (X1) represents the output of the j-th independent convolutional layer after processing the structural reconstruction features, and X2 represents the texture reconstruction features.

[0091] Furthermore, the sharpness reconstruction layer is used to reconstruct the texture reconstructed features of each sub-image to obtain the optimized features of each sub-image, wherein the sharpness reconstruction layer in the embodiment of the present invention includes 5 independent convolutional layers, and the loss function of the sharpness reconstruction layer is:

[0092]

[0093] Among them, L3 represents the loss function of the sharpness reconstruction layer, represents the output of the kth independent convolutional layer of the sharpness reconstruction layer, I target (X) represents the pixel of each sub-image after texture reconstruction, I pred (X) represents the pixel value of each sub-image after sharpness reconstruction, represents the output of the kth independent convolutional layer after processing the optimized features, It represents the output of the k-th independent convolutional layer after processing the texture reconstruction features, and X3 represents the optimized features.

[0094] Step 3024, input the optimized features of each sub-image into the feature fusion layer, and fuse the optimized features of each sub-image according to the number information of each sub-image through the feature fusion layer to obtain the resolution reconstructed image output by the feature fusion layer.

[0095] Furthermore, the optimized features of each sub-image are input into the feature fusion layer, and the optimized features of each sub-image are fused according to the numbering information of each sub-image through the feature fusion layer to obtain the resolution reconstructed image output by the feature fusion layer.

[0096] The embodiment of the present invention performs resolution reconstruction processing on the infrared image. Even in extreme environments, the processed infrared image can restore the detailed features of the infrared image, accurately identify the temperature information of the target power supply equipment, and improve the monitoring accuracy of the power supply equipment.

[0097] In one embodiment, the image after resolution reconstruction is enhanced to obtain a target recognition image, including:

[0098] Step 3031, obtaining the channel complexity of each image channel of the image after resolution reconstruction.

[0099] Optionally, the pixel value of each image channel (such as R, G, B in the RGB channel) of the image after resolution reconstruction is obtained, and the pixel value of each image channel is counted as a histogram to obtain the probability distribution of the grayscale of each image channel. Further, according to the probability distribution of the grayscale of each image channel, the entropy value of each image channel is determined, and the specific formula is as follows:

[0100]

[0101] Among them, H(C) represents the entropy value of image channel C, and p(i) represents the probability distribution of the gray level of image channel C.

[0102] Further, the average value of each image channel and the average value corresponding to the reference channel of each image channel are obtained.

[0103] The average value corresponding to the channel is used to determine the structural similarity index between each image channel and its corresponding reference channel. The specific calculation formula is as follows:

[0104]

[0105] Among them, SSIM (C, C ref ) represents the image channel C and the reference channel C ref The structural similarity index between C represents the average value of image channel C, Indicates reference channel C ref The average value, σ C Represents the standard deviation of image channel C, which is used to characterize the contrast of brightness. Indicates reference channel C ref The standard deviation of Represents the image channel C and the reference channel C ref The covariance between them, C1 and C2 are preset constants.

[0106] Furthermore, according to the entropy value of each image channel and the structural similarity index between each image channel and its corresponding reference channel, the channel complexity of each image channel in the image after resolution reconstruction is determined. The specific formula is as follows:

[0107] Complexity(C)=α*H(C)+β*SSIM(C,C ref )

[0108] Wherein, Complexity(C) represents the channel complexity of the image channel C, and α and β are preset weight coefficients.

[0109] Step 3032: Determine a channel gain value of each image channel based on the channel complexity of each image channel.

[0110] Furthermore, according to the channel complexity of each image channel, the channel gain value of each image channel is determined. The specific formula is as follows:

[0111] Gain(C)=k*Complexity(C)+b

[0112] Wherein, Gain(C) represents the channel gain value of the image channel C, k represents the coefficient of the linear relationship between the control complexity and the gain, and b represents the bias term, which is used to adjust the basic value of the gain so that the gain is within a reasonable range.

[0113] Step 3033: Based on the channel gain value of each image channel, each image channel is enhanced to obtain a target recognition image.

[0114] Further, according to the channel gain value of each image channel, each image channel is enhanced to obtain an image of the enhanced image channel. Further, a target recognition image is generated based on the image of each enhanced image channel. The specific formula of the enhancement process is as follows:

[0115] I enhanced (C) = CLAHE (I (C), Gain (C))

[0116] Among them, I enhanced(C) represents the enhanced image of image channel C, represents the image of original image channel C, and CLAHE() represents the adaptive histogram equalization algorithm.

[0117] The embodiment of the present invention performs enhanced processing on the infrared image. Even in extreme environments, the processed infrared image can restore the detailed features of the original infrared image, accurately identify the temperature information of the target power supply equipment, and improve the monitoring accuracy of the power supply equipment.

[0118] In one embodiment, in order for the user terminal to be able to quickly and accurately receive the infrared image and temperature information sent by the mobile monitoring device, the power supply equipment monitoring method of the present invention also includes: step 50, determining the data transmission method based on the environment in which the target power supply equipment is located; step 60, compressing the infrared image to obtain compressed information; step 70, transmitting the compressed information and temperature information to the user terminal in a data transmission manner.

[0119] Optionally, environmental information of the environment in which the target power supply equipment is located is obtained, and a data transmission method is determined according to the environmental information. Further, the infrared image is compressed to obtain compressed information. Further, the compressed information and temperature information are transmitted to the user terminal by a data communication unit in a data transmission manner. After the user terminal receives the compressed information and temperature information, the compressed information is decompressed to obtain an infrared image, and data verification is further performed based on the infrared image and temperature information.

[0120] On the one hand, the compressed information in the embodiment of the present invention occupies less communication resources and has a faster transmission speed. On the other hand, the compressed information and temperature information are transmitted to the user terminal in different data transmission modes according to different environmental information, thereby reducing the interference of environmental factors on data communication. Therefore, the user terminal can quickly and accurately receive the infrared image and temperature information sent by the mobile monitoring device, thereby improving the speed of data verification.

[0121] In one embodiment, determining the data transmission mode based on the environment of the target power supply device includes:

[0122] Step 501, obtaining environmental information of the environment in which the target power supply equipment is located;

[0123] Step 502, performing spectrum sensing on signals in the environment based on the environment information to obtain interference information existing in the environment; the interference information includes interference type and interference bandwidth;

[0124] Step 503: determine the data transmission mode according to the interference type and the interference bandwidth.

[0125] Optionally, when transmitting data, the mobile monitoring device obtains environmental information of the environment in which the target power supply equipment is located, and performs spectrum sensing on the signals in the environment through a preset spectrum sensing algorithm based on the environmental information to obtain interference information existing in the environment, wherein the spectrum sensing algorithm may include an energy detection method, a matched filter detection method, a cyclostationary detection method, etc., and the interference information includes interference type and interference bandwidth, the interference types include thermal noise, adjacent channel interference and multipath interference, and the interference bandwidth includes narrowband and broadband.

[0126] Furthermore, the interference type and the interference bandwidth are matched in a preset data transmission mode mapping table to obtain a data transmission mode corresponding to the interference type and the interference bandwidth. In one embodiment, the data transmission mode mapping table is shown in Table 1.

[0127] Table 1 Data transmission mode mapping table

[0128]

[0129] The embodiment of the present invention transmits compression information and temperature information to the user terminal in different data transmission modes according to different environmental information, thereby reducing the interference of environmental factors on data communication. Therefore, the user terminal can quickly and accurately receive the infrared image and temperature information sent by the mobile monitoring device, thereby improving the speed of data verification.

[0130] In one embodiment, step 60 compresses the infrared image to obtain compressed information including:

[0131] Step 601, splitting the infrared image into non-overlapping image blocks of a preset size;

[0132] Step 602, performing a two-dimensional discrete cosine transform on each image block to convert the spatial domain data of the infrared image into frequency domain data to obtain DCT coefficients of the infrared image;

[0133] Step 603, quantizing the DCT coefficients of the infrared image to obtain quantized DCT coefficients;

[0134] Step 604: entropy code the quantized DCT coefficients to obtain compressed information.

[0135] Optionally, the infrared image is split into non-overlapping image blocks of a preset size, wherein the preset size may be, for example, 4*4, 8*8, 16*16, etc.

[0136] Furthermore, a two-dimensional discrete cosine transform (DCT) is performed on each image block to convert the spatial domain data of the infrared image into frequency domain data to obtain the DCT coefficient of the infrared image. The specific calculation formula can be expressed as:

[0137]

[0138] Wherein, u and v = 0, 1, 2, ..., 7 represent frequency coordinates, I (x, y) represents the pixel value of each image block, and F (u, v) represents the DCT coefficient of the infrared image.

[0139] Furthermore, the DCT coefficients of the infrared image are quantized to obtain quantized DCT coefficients to reduce the accuracy of high-frequency components and achieve better compression effects. The specific formula is as follows:

[0140]

[0141] Among them, Q(u,v) represents the quantized DCT coefficient, Q u,v represents a predefined quantization matrix, Represents the quantization process.

[0142] Furthermore, the quantized DCT coefficients are entropy encoded to obtain compressed information. The specific formula is as follows:

[0143] C p =EntropyEncode(Q(u,v))

[0144] Among them, C p represents compressed information, and EntropyEncode(·) represents the entropy encoding process.

[0145] Furthermore, after receiving the compressed information, the user terminal first performs entropy decoding on the compressed information. The specific formula is as follows:

[0146] F ′ (u,v)=Q(u,v)×Q u,v

[0147] Among them, F ′ (u,v) represents the inverse DCT coefficients.

[0148] Furthermore, each image block is restored by inverse DCT, specifically:

[0149]

[0150] Among them, I ′ (x,y) represents the restored image block.

[0151] Furthermore, all restored image blocks are reassembled to obtain an infrared image.

[0152] The embodiment of the present invention compresses the infrared image into compressed information that occupies less communication resources, making the data transmission speed faster. Therefore, the user terminal can quickly and accurately receive the infrared image and temperature information sent by the mobile monitoring device, thereby improving the speed of data verification.

[0153] Furthermore, all restored image blocks are reassembled to obtain an infrared image.

[0154] The embodiment of the present invention compresses the infrared image into compressed information that occupies less communication resources, making the data transmission speed faster. Therefore, the user terminal can quickly and accurately receive the infrared image and temperature information sent by the mobile monitoring device, thereby improving the speed of data verification.

[0155] The above disclosure is only the preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A mobile monitoring device based on artificial intelligence, characterized in that: It includes a movable device and a power supply equipment monitoring device, wherein the movable device is connected to the power supply equipment monitoring device; the power supply equipment monitoring device includes a central processing unit, an infrared image acquisition unit, an image blurring unit, an image processing unit and a temperature recognition unit; The infrared image acquisition unit is used to acquire an infrared image of a target power supply equipment; the target power supply equipment is any power supply equipment in the power system; The image processing unit is used to perform image blur detection on the infrared image based on a blur detection model, and output a blur detection result of the infrared image; The central processing unit is used for: If the blur detection result is that the infrared image is a blurry image, the image processing unit is controlled to perform image processing on the infrared image to obtain a target recognition image of the infrared image; the temperature recognition unit is controlled to recognize the target recognition image through a temperature recognition model to obtain temperature information of the target power supply equipment; the image processing process includes grayscale processing, resolution reconstruction processing and enhancement processing; If the blur detection result is that the infrared image is a non-blurred image, the temperature recognition unit is controlled to recognize the infrared image through the temperature recognition model to obtain the temperature information of the target power supply equipment.

2. A method for monitoring power supply equipment, characterized in that: Applied to the artificial intelligence-based mobile monitoring device of claim 1, the method comprises: Collecting an infrared image of the target power supply equipment; inputting the infrared image into a fuzzy detection model to perform image fuzzy detection, and obtaining a fuzzy detection result output by the fuzzy detection model; If the blur detection result is that the infrared image is a blurry image, the infrared image is processed to obtain a target recognition image of the infrared image; the target recognition image is input into a temperature recognition model for recognition to obtain the temperature information of the target power supply equipment output by the temperature recognition model; the image processing process includes grayscale processing, resolution reconstruction processing and enhancement processing; If the blur detection result is that the infrared image is a non-blurred image, the infrared image is input into the temperature recognition model for recognition, and the temperature information of the target power supply equipment output by the temperature recognition model is obtained.

3. The method for monitoring power supply equipment according to claim 2, characterized in that: The blur detection model includes a pixel extraction function, a pixel quantization function and a blur detection function; The step of inputting the infrared image into a fuzzy detection model to perform image fuzzy detection and obtaining a fuzzy detection result output by the fuzzy detection model includes: The infrared image is input into the blur detection model, and pixels of the infrared image are extracted based on the pixel extraction function to obtain the pixel value of each row of pixels and the pixel value of each column of pixels in the infrared image. The pixel extraction function is as follows: Among them, P i Represents the pixel value of each row of pixels, P j Represents the pixel value of each column of pixels, M i represents the number of pixels in the i-th row, Indicates the Mth row in the i-th row i The pixel value of pixels, N j represents the number of pixels in the jth column, p jN Indicates the jth column, Nth j The pixel value of pixels; The pixel values ​​of each row of pixels and each column of pixels of the infrared image are quantized based on the pixel quantization function to obtain the pixel quantization value of the infrared image. The pixel quantization function is specifically as follows: Among them, Q ij Represents the quantized value of the pixel in the i-th row and j-th column in the infrared image; Based on the fuzzy detection function combined with the pixel quantization value of the infrared image, fuzzy detection is performed to obtain the fuzzy detection result. The fuzzy detection function is specifically as follows: Among them, Result represents the blur detection result, 1 represents a blurry image, 0 represents a non-blurred image, w[Q ij ≤a] represents the pixel quantization value Q ij The number of pixels is less than or equal to the preset pixel value a, and b is the preset number.

4. The method for monitoring power supply equipment according to claim 2, characterized in that: The performing image processing on the infrared image to obtain a target recognition image of the infrared image includes: Performing grayscale processing on the infrared image to obtain a grayscale image; Performing resolution reconstruction processing on the grayscale image to obtain a resolution reconstructed image; The resolution reconstructed image is enhanced to obtain the target recognition image.

5. The method for monitoring power supply equipment according to claim 4, characterized in that: The grayscale processing of the infrared image to obtain a grayscale image includes: Performing grayscale processing on the infrared image to obtain a pixel grayscale value of each pixel in the infrared image; Determine a pixel grayscale mean based on a first preset number of first pixel grayscale values ​​with the largest value and a second preset number of second pixel grayscale values ​​with the smallest value; Adjusting the pixel grayscale value of each pixel in the infrared image based on the pixel grayscale mean value to obtain a final pixel grayscale value of each pixel in the infrared image; The image generated based on the final pixel grayscale value of each pixel in the infrared image is the grayscale image; the specific calculation formula of the pixel grayscale mean is as follows: Among them, G average Indicates the mean grayscale value of pixels, max m represents the first preset quantity, Represents the gray value of the first pixel, Indicates the max m The first pixel gray value, Indicates the gray value of the first and second pixels, min n Indicates the second preset quantity, Indicates the min n A second pixel gray value; The specific formula for grayscale value adjustment is as follows: Among them, G last Represents the final pixel grayscale value, G(x <G average ) represents the pixel gray value less than the pixel gray mean value, G(x≥G average ) represents the pixel grayscale value that is greater than or equal to the pixel grayscale mean value, L represents the grayscale level, and the grayscale level is generally 256.

6. The method for monitoring power supply equipment according to claim 4, characterized in that: The resolution reconstruction network includes an image segmentation layer, a feature extraction layer, a feature optimization layer and a feature fusion layer; the resolution reconstruction network is obtained based on the sample image and the super-resolution network training corresponding to the sample image; the resolution reconstruction processing of the grayscale image to obtain the resolution reconstructed image includes: Inputting the grayscale image into the image segmentation layer, segmenting the grayscale image through the image segmentation layer, and obtaining a plurality of sub-images carrying numbering information output by the image segmentation layer; Inputting a plurality of sub-images into the feature extraction layer, performing feature extraction on the plurality of sub-images through the feature extraction layer, and obtaining image features of each sub-image output by the feature extraction layer; Inputting the image features of each sub-image into the feature optimization layer, optimizing the image features of each sub-image through the feature optimization layer, and obtaining the optimized features of each sub-image output by the feature optimization layer; The optimized features of each sub-image are input into the feature fusion layer, and the optimized features of each sub-image are fused according to the numbering information of each sub-image through the feature fusion layer to obtain the resolution reconstructed image output by the feature fusion layer.

7. The method for monitoring power supply equipment according to claim 4, characterized in that: The step of performing enhancement processing on the resolution reconstructed image to obtain the target recognition image includes: Obtaining the channel complexity of each image channel in the image reconstructed at the resolution; Determining a channel gain value of each of the image channels based on the channel complexity of each of the image channels; Based on the channel gain value of each of the image channels, each image channel is enhanced to obtain the target recognition image.

8. The method for monitoring power supply equipment according to any one of claims 2 to 7, characterized in that: The power supply equipment monitoring method further includes: Determining a data transmission mode based on the environment in which the target power supply equipment is located; compressing the infrared image to obtain compressed information; The compression information and the temperature information are transmitted to a user terminal in the data transmission manner.

9. The method for monitoring power supply equipment according to claim 8, characterized in that: The determining of the data transmission mode based on the environment of the target power supply device includes: Obtaining environmental information of the environment in which the target power supply equipment is located; Based on the environmental information, spectrum sensing is performed on the signal in the environment to obtain interference information existing in the environment; the interference information includes interference type and interference bandwidth; The data transmission mode is determined according to the interference type and the interference bandwidth.

10. The method for monitoring power supply equipment according to claim 8, characterized in that: The step of compressing the infrared image to obtain compressed information includes: Splitting the infrared image into non-overlapping image blocks of a preset size; Performing a two-dimensional discrete cosine transform on each of the image blocks to convert the spatial domain data of the infrared image into frequency domain data to obtain DCT coefficients of the infrared image; quantizing the DCT coefficients of the infrared image to obtain quantized DCT coefficients; The quantized DCT coefficients are entropy encoded to obtain the compressed information.

Images