Cable tunnel crack detection method and device and computer equipment

Abstract

The invention relates to a cable tunnel crack detection method and device and computer equipment. The method comprises the following steps: on the basis of obtaining a target infrared thermogram of a cable tunnel, extracting a spatial weight feature map of the target infrared thermogram through a preset tunnel feature extraction model, and fusing the target infrared thermogram and the spatial weight feature map to obtain a weighted feature map; then, fusing the weighted feature map with the channel radiation constraint value of the corresponding channel to obtain a tunnel fusion initial feature map; then, rotating the tunnel fusion initial feature map according to a preset rotation angle set, and fusing image features in the rotated tunnel fusion initial feature map to obtain a tunnel fusion enhanced feature map; and finally, based on the temperature information and the crack information in the tunnel fusion enhancement feature map, performing crack evaluation on the cable tunnel to obtain a crack evaluation result of the cable tunnel. By adopting the method, the accuracy of the crack detection result of the cable tunnel can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent detection, in particular to a cable tunnel crack detection method and device and computer equipment. BACKGROUND

[0002] As an important channel for urban power transmission, the cable tunnel undertakes a large amount of power transmission tasks. Therefore, it is necessary to detect and maintain the cable tunnel, discover and handle potential fault hazards in time, and ensure the reliability of power transmission.

[0003] In the related art, when detecting the cable tunnel, the cable tunnel image is usually collected, and the image processing method or the deep learning model is used to process the cable tunnel image to determine whether there is a crack in the cable tunnel.

[0004] However, the accuracy of the crack detection result of the cable tunnel in the related art is low. SUMMARY

[0005] Therefore, it is necessary to provide a cable tunnel crack detection method, device and computer equipment to improve the accuracy of the crack detection result of the cable tunnel.

[0006] In a first aspect, the present application provides a cable tunnel crack detection method, which comprises:

[0007] Obtaining a target infrared thermal image of a cable tunnel;

[0008] Extracting a spatial weight feature map of the target infrared thermal image through a preset tunnel feature extraction model, and fusing the target infrared thermal image and the spatial weight feature map to obtain a weighted feature map;

[0009] Fusing the weighted feature map and a channel radiation constraint value of a corresponding channel to obtain a tunnel fusion initial feature map;

[0010] Rotating the tunnel fusion initial feature map according to a preset rotation angle set, and fusing image features in the rotated tunnel fusion initial feature map to obtain a tunnel fusion enhanced feature map;

[0011] Based on temperature information and crack information in the tunnel fusion enhanced feature map, performing crack evaluation on the cable tunnel to obtain a crack evaluation result of the cable tunnel.

[0012] In one embodiment, fusing the weighted feature map and the channel radiation constraint value of the corresponding channel to obtain the tunnel fusion initial feature map comprises:

[0013] Calculating a plurality of pixel temperature gradients of the weighted feature map corresponding to each channel;

[0014] respectively, to obtain a channel radiation constraint value of each channel;

[0015] The weighted feature map is fused with the channel radiation constraint value of the corresponding channel to obtain a tunnel fusion initial feature map.

[0016] In one of the embodiments, before the crack evaluation of the cable tunnel is performed to obtain the crack evaluation result of the cable tunnel, the method further comprises:

[0017] The temperature values between the pixel positions in the tunnel fusion enhanced feature map are analyzed for correlation to determine an initial temperature anomaly region set in the tunnel fusion enhanced feature map;

[0018] Regions greater than a preset area threshold are filtered out from the initial temperature anomaly region set as candidate temperature anomaly regions;

[0019] The temperature information is obtained by summarizing each candidate temperature anomaly region, the average temperature of each candidate temperature anomaly region, and the temperature change rate of each candidate temperature anomaly region.

[0020] In one of the embodiments, the temperature information includes at least one temperature anomaly region, and the crack information includes at least one crack region; the crack evaluation of the cable tunnel is performed based on the temperature information and the crack information in the tunnel fusion enhanced feature map to obtain the crack evaluation result of the cable tunnel, which includes:

[0021] At least one target crack is filtered out from the crack set in each crack region;

[0022] For any target crack, the crack region corresponding to the target crack is compared with each temperature anomaly region to determine at least one target temperature anomaly region corresponding to the target crack;

[0023] For any target temperature anomaly region, if the average temperature and the temperature change rate of the target temperature anomaly region meet a preset temperature threshold, an evaluation result is generated; the evaluation result includes the coordinates of the target temperature anomaly region and the crack region coordinates of the target crack.

[0024] In one of the embodiments, at least one target crack is filtered out from the crack set in each crack region, which includes:

[0025] The crack length and the crack complexity index of each crack are obtained;

[0026] The crack whose crack length is greater than a preset length threshold and whose crack complexity is greater than a preset complexity threshold is determined as the target crack.

[0027] In one of the embodiments, the target infrared thermal image of the cable tunnel is obtained, which includes:

[0028] obtain an initial infrared thermal image of the cable tunnel;

[0029] correct the thermal radiation value in the initial infrared thermal image based on a preset environmental temperature compensation quantization value, and perform nonlinear enhancement processing on the pixel value in the corrected infrared thermal image to obtain a tunnel enhanced feature map;

[0030] extract image features of the initial infrared thermal image using a semantic feature extraction model, and fuse the extracted infrared thermal image feature map and the tunnel enhanced feature map to obtain a target infrared thermal image.

[0031] In one of the embodiments, the step of obtaining the environmental temperature compensation quantization value comprises:

[0032] obtain temperature sequences corresponding to multiple positions in the cable tunnel; each temperature sequence includes tunnel temperatures collected at different times at the same position;

[0033] perform mean value calculation on the temperature values of the temperature sequences within a preset time window to obtain an environmental average temperature; the preset time window is determined according to the current collection time and a preset window length;

[0034] process the environmental average temperature based on a preset environmental temperature influence parameter to obtain an actual temperature of the cable tunnel;

[0035] take the difference between the actual temperature and the environmental average temperature as the environmental temperature compensation quantization value.

[0036] In one of the embodiments, the nonlinear enhancement processing on the pixel value in the corrected infrared thermal image comprises:

[0037] for any pixel value, determine a contrast quantization value according to a first difference between the pixel value and a first pixel adjustment parameter, and a second difference between a target pixel value and the first pixel adjustment parameter;

[0038] calculate a pixel value to be enhanced with the contrast quantization value as the base and a second pixel adjustment parameter as the exponent;

[0039] fuse the pixel value to be enhanced, the target pixel value, the second pixel adjustment parameter and a third pixel adjustment parameter to obtain an enhanced pixel value.

[0040] In a second aspect, the present application also provides a cable tunnel crack detection device, comprising:

[0041] an image acquisition module configured to obtain a target infrared thermal image of the cable tunnel;

[0042] The feature extraction module is configured to extract a spatial weight feature map of the target infrared thermal image by using a preset tunnel feature extraction model, fuse the target infrared thermal image and the spatial weight feature map, and obtain a weighted feature map.

[0043] The feature fusion module is configured to fuse the weighted feature map and a channel radiation constraint value of a corresponding channel to obtain a tunnel fusion initial feature map.

[0044] The feature enhancement module is configured to rotate the tunnel fusion initial feature map according to a preset set of rotation angles, fuse image features in the rotated tunnel fusion initial feature map, and obtain a tunnel fusion enhanced feature map.

[0045] The tunnel evaluation module is configured to perform crack evaluation on the cable tunnel based on temperature information and crack information in the tunnel fusion enhanced feature map, and obtain a crack evaluation result of the cable tunnel.

[0046] In a third aspect, the present application also provides a computer device, including a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method in any one of the embodiments of the first aspect when executing the computer program.

[0047] In a fourth aspect, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the method in any one of the embodiments of the first aspect when executed by a processor.

[0048] In a fifth aspect, the present application also provides a computer program product, which includes a computer program, and the computer program implements the steps of the method in any one of the embodiments of the first aspect when executed by a processor.

[0049] The cable tunnel crack detection method, device and computer equipment, based on the obtained target infrared thermal image of the cable tunnel, extracts a spatial weight feature map of the target infrared thermal image through a preset tunnel feature extraction model, fuses the target infrared thermal image and the spatial weight feature map to obtain a weighted feature map, then fuses the weighted feature map and a channel radiation constraint value of a corresponding channel to obtain a tunnel fusion initial feature map, then rotates the tunnel fusion initial feature map according to a preset rotation angle set and fuses image features in the rotated tunnel fusion initial feature map to obtain a tunnel fusion enhanced feature map, and finally, based on temperature information and crack information in the tunnel fusion enhanced feature map, performs crack evaluation on the cable tunnel to obtain a crack evaluation result of the cable tunnel. In the crack detection of the cable tunnel, the spatial weight feature map of the target infrared thermal image is extracted by using the tunnel feature extraction model, the weight of the crack morphology in the target infrared thermal image is quantified, the target infrared thermal image and the spatial weight feature map are fused, the area related to the crack in the target infrared thermal image is highlighted, the interference information is suppressed, and the effectiveness of the weighted feature map is improved, so as to facilitate the subsequent extraction of crack features; then the weighted feature map and the channel radiation constraint value of the corresponding channel are fused, the importance degree quantization value of each position and channel in the weighted feature map is dynamically adjusted, the tunnel fusion initial feature map is obtained, so as to more accurately capture the temperature abnormal information of the crack area; further, considering the uncertainty factor of the crack propagation direction, the tunnel fusion initial feature map is rotated according to the preset rotation angle set, and the image features in the rotated tunnel fusion initial feature map are fused, so that the tunnel fusion enhanced feature map obtained in this way can comprehensively consider the crack features in each direction, and the comprehensiveness and richness of the tunnel fusion enhanced feature map are improved; finally, the crack information in the tunnel fusion enhanced feature map is evaluated from the two dimensions of temperature and crack, and a more accurate crack detection result is obtained. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other related drawings can also be obtained without creative labor.

[0051] Figure 1 It is an internal structure diagram of the computer equipment in an embodiment;

[0052] Figure 2 It is a flowchart of the cable tunnel crack detection method in an embodiment;

[0053] Figure 3A flowchart of a target infrared thermal image acquisition step in an embodiment is shown in FIG. 1.

[0054] Figure 4 A flowchart of a tunnel fusion initial feature map acquisition step in an embodiment is shown in FIG. 2.

[0055] Figure 5 A flowchart of a temperature information acquisition step in an embodiment is shown in FIG. 3.

[0056] Figure 6 A flowchart of a temperature information acquisition step in an embodiment is shown in FIG. 4.

[0057] Figure 7 A flowchart of a cable tunnel crack evaluation step in an embodiment is shown in FIG. 5.

[0058] Figure 8 A flowchart of a target crack determination step in an embodiment is shown in FIG. 6.

[0059] Figure 9 A block diagram of a cable tunnel crack detection device in an embodiment is shown in FIG. 7. DETAILED DESCRIPTION

[0060] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0061] The cable tunnel crack detection method provided by the embodiments of the present application can be applied to a computer device, which can be a server, and the internal structure diagram of the server can be as shown in FIG. 8. Figure 1As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store cable tunnel crack detection data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a cable tunnel crack detection method. The server can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0062] Those skilled in the art can understand that, Figure 1 The skilled in the art can understand that,

[0063] In the current field of intelligent detection of cable tunnels, there are many technical problems that seriously restrict the accuracy and efficiency of detection, including at least the following aspects: (1) The limitations of traditional image processing methods, such as edge detection combined with morphological processing, perform poorly in processing cracks in infrared thermal images. Moreover, such methods cannot associate the morphological features of cracks with temperature anomaly features, resulting in an inability to comprehensively analyze the condition of the cable tunnel. (2) Defects of deep learning models, such as U-Net, DeepLabv3+, etc., have achieved certain results in image detection, but have obvious problems in long-distance crack detection. When faced with continuous cracks longer than 5m, the miss detection rate is high. This is because the model has a feature breakage during long-distance feature extraction, making the detection result inaccurate. (3) Problems of conventional attention mechanisms, such as Convolutional Block Attention Module (CBAM), are easily disturbed by thermal radiation noise in infrared image processing, and the false detection rate increases sharply when the foreground-background temperature difference is greater than the preset temperature difference threshold, which greatly reduces the reliability of detection methods based on these attention mechanisms in complex environments.

[0064] Based on the above problems, the present application provides a cable tunnel crack detection method, device and computer equipment to improve the accuracy of the crack detection result of the cable tunnel. The technical solutions of the present application and how the technical solutions solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0065] In one exemplary embodiment, as shown in Figure 2 A cable tunnel crack detection method is provided, which is described by taking the application to a computer equipment as an example, and can also be applied to a system including a terminal and a server, and realized through the interaction of the terminal and the server. In this embodiment, the method comprises the following steps:

[0066] S201, obtaining a target infrared thermal image of the cable tunnel.

[0067] The cable tunnel is a kind of underground channel facility specially used for laying, installing and maintaining power cables (such as high-voltage transmission cables, urban power grid cables, etc.), which is usually composed of reinforced concrete, brick or metal structure, and has the characteristics of closed or semi-closed space. When detecting the cracks of the cable tunnel, considering safety and detection efficiency, etc., the image of the cable tunnel is usually collected and detected by image processing.

[0068] In the embodiments of the present application, the infrared thermal image of the cable tunnel is taken as the description information of the cable tunnel to realize the crack detection of the cable tunnel. The infrared thermal image refers to the visual image obtained by converting the temperature distribution on the surface of the cable tunnel through infrared thermal imaging technology, which can directly reflect the temperature difference of each part of the cable tunnel surface. Moreover, each pixel point of the infrared thermal image corresponds to a pixel value and a thermal radiation intensity value, the pixel value is used to represent the image features of the cable tunnel, and the thermal radiation intensity value is used to represent the local temperature features of the cable tunnel.

[0069] Optionally, a high-precision infrared thermal imager can be carried on an automatic inspection device, and the infrared thermal image of the cable tunnel can be collected by moving the automatic inspection device in the cable tunnel. In another implementation scenario, the infrared thermal imager can also be fixedly deployed inside the cable tunnel, and the computer equipment communicates with the infrared thermal imager to obtain the infrared thermal image of the cable tunnel sent by the infrared thermal imager.

[0070] After the infrared thermal image of the cable tunnel is acquired, the infrared thermal image collected by the infrared thermal imager can be directly taken as the target infrared thermal image, or the infrared thermal image can be preprocessed, including but not limited to filtering processing, ambient temperature compensation processing, contrast enhancement processing, etc., and the infrared thermal image obtained by the preprocessing is determined as the target infrared thermal image.

[0071] In S202, a spatial weight feature map of the target infrared thermal image is extracted by a preset tunnel feature extraction model, and the target infrared thermal image is fused with the spatial weight feature map to obtain a weighted feature map.

[0072] The tunnel feature extraction model can be a pre-trained convolutional neural network model, which includes, from input to output, two parallel convolutional layers (a first convolutional layer and a second convolutional layer), a feature splicing and fusion layer, a global average pooling layer, a spatial attention weight map generation layer, and a spatial feature weighting layer.

[0073] The target infrared thermal image is input into the two parallel convolutional layers in parallel, the features of the target infrared thermal image are extracted by the two convolutional layers using different numbers and different sizes of convolutional kernels, and two convolutional feature maps are output; the two convolutional feature maps are spliced in the channel dimension by the feature splicing and fusion layer, and the spliced features are fused by a third convolutional layer and input into the global average pooling layer; the input feature map is then processed by the global average pooling layer to obtain a one-dimensional vector; the one-dimensional vector is normalized by the spatial attention weight map generation layer using a normalization function such as a sigmoid function to obtain a spatial weight feature map; and the spatial weight feature map and the target infrared thermal image are multiplied by the principal element by the spatial feature weighting layer to obtain a weighted feature map.

[0074] In S203, the weighted feature map is fused with a channel radiation constraint value of a corresponding channel to obtain a tunnel fusion initial feature map.

[0075] It should be noted that the weighted feature map includes multiple channels, each channel corresponds to a channel radiation constraint value, and the channel radiation constraint value of each channel is determined according to the comprehensive temperature gradient value of each pixel point in the weighted feature map corresponding to the channel.

[0076] The weighted feature map is decomposed into weighted feature maps of multiple channels, then each feature value on the weighted feature map of each channel is multiplied by the channel radiation constraint value of the corresponding channel to obtain a weighted enhanced feature map corresponding to each channel, and finally the weighted enhanced feature maps corresponding to each channel are fused to obtain a tunnel fusion initial feature map.

[0077] S204, rotate the tunnel fusion initial feature map according to the set of preset rotation angles, and fuse the image features in the rotated tunnel fusion initial feature map to obtain a tunnel fusion enhanced feature map.

[0078] Rotate the tunnel fusion initial feature map in different directions, and perform strip convolution operation on each rotated feature map to obtain a rotated convolution image feature, and then fuse each rotated convolution image feature according to a preset rotation weight to obtain a tunnel fusion enhanced feature map.

[0079] Optionally, the tunnel fusion initial feature map is rotated according to four directions {0°, 45°, 90°, 135°}, and then for each rotated tunnel fusion initial feature map, a strip convolution operation with a hole rate of 3 is independently performed to obtain four rotated convolution image features, and then each rotated convolution image feature is fused according to a preset four rotation weights to obtain a tunnel fusion enhanced feature map. Wherein, the hole rate of 3 means that the convolution kernel samples every 3 pixels during convolution operation, which can expand the receptive field of convolution without increasing the size of the convolution kernel, so as to capture more extensive feature information.

[0080] Optionally, the tunnel fusion initial feature map is rotated according to eight directions {0°, 22.5°, 45°, 67.5°, 90°, 120.5°, 135°, 157.5°}, and then for each rotated tunnel fusion initial feature map, a strip convolution operation with a hole rate of 3 is independently performed to obtain eight rotated convolution image features, and then each rotated convolution image feature is fused according to a preset eight rotation weights to obtain a tunnel fusion enhanced feature map.

[0081] S205, based on the temperature information and crack information in the tunnel fusion enhanced feature map, performing crack evaluation on the cable tunnel to obtain a crack evaluation result of the cable tunnel.

[0082] According to the temperature information and crack information in the tunnel fusion enhanced feature map, the cracks in the temperature abnormal area or the temperature value corresponding to the crack area are screened out from the tunnel fusion enhanced feature map, and then in the case that the cracks in the temperature abnormal area or the temperature value corresponding to the crack area meet the preset condition, it is determined that the cable tunnel has cracks, and the position information of the crack area meeting the preset condition is taken as the crack evaluation result of the cable tunnel. In addition, in the case that the cable tunnel has cracks, an alarm information can also be triggered, which carries the position information of the crack area, so as to repair the cracks of the cable tunnel in time and quickly.

[0083] If the temperature abnormal area and the crack area intersect in the tunnel fusion enhanced feature map, and the temperature value corresponding to the crack area in the temperature abnormal area does not satisfy the preset condition, it is determined that the crack evaluation result of the cable tunnel is that there is a potential crack.

[0084] If the temperature abnormal area and the crack area intersect in the tunnel fusion enhanced feature map, and the temperature value corresponding to the crack area in the temperature abnormal area does not satisfy the preset condition, it is determined that the crack evaluation result of the cable tunnel is that there is a potential crack.

[0085] In the embodiment of the present application, on the basis of obtaining the target infrared thermal image of the cable tunnel, the spatial weight feature map of the target infrared thermal image is extracted through the preset tunnel feature extraction model, and the target infrared thermal image and the spatial weight feature map are fused to obtain a weighted feature map. Then, the weighted feature map and the channel radiation constraint value of the corresponding channel are fused to obtain a tunnel fusion initial feature map. Next, the tunnel fusion initial feature map is rotated according to the preset rotation angle set, and the image features in the rotated tunnel fusion initial feature map are fused to obtain a tunnel fusion enhanced feature map. Finally, based on the temperature information and the crack information in the tunnel fusion enhanced feature map, the crack of the cable tunnel is evaluated to obtain the crack evaluation result of the cable tunnel. In the crack detection of the cable tunnel, the spatial weight feature map of the target infrared thermal image is extracted by using the tunnel feature extraction model, the weight of the crack morphology in the target infrared thermal image is quantified, and the target infrared thermal image and the spatial weight feature map are fused to highlight the area related to the crack in the target infrared thermal image and suppress the interference information, thereby improving the effectiveness of the weighted feature map to facilitate the subsequent extraction of crack features. Then, the weighted feature map and the channel radiation constraint value of the corresponding channel are fused to dynamically adjust the importance degree quantization value of each position and channel in the weighted feature map to obtain the tunnel fusion initial feature map, so as to more accurately capture the temperature abnormal information of the crack area. Further, considering the uncertainty factor of the crack propagation direction, the tunnel fusion initial feature map is rotated according to the preset rotation angle set, and the image features in the rotated tunnel fusion initial feature map are fused, so that the tunnel fusion enhanced feature map obtained in this way can comprehensively consider the crack features in various directions, thereby improving the comprehensiveness and richness of the tunnel fusion enhanced feature map. Finally, the crack information in the tunnel fusion enhanced feature map is evaluated from the two dimensions of temperature and crack to obtain a more accurate crack detection result.

[0086] In one exemplary embodiment, as shown in Figure 3 the target infrared thermal image of the cable tunnel is obtained, including:

[0087] S301, obtaining an initial infrared thermal image of a cable tunnel.

[0088] The initial infrared thermal image is an image directly captured by an infrared thermal imager in a cable tunnel. In the initial infrared thermal image, each pixel corresponds to a pixel value and a thermal radiation value. In the embodiment of the present application, the initial infrared thermal image can be obtained by directly communicating with the infrared thermal imager, or the initial infrared thermal image can be read from a storage medium after the initial infrared thermal image is pre-stored in the storage medium.

[0089] S302 , based on a preset ambient temperature compensation quantization value, correct the thermal radiation value in the initial infrared thermal image, and perform nonlinear enhancement processing on the pixel values ​​in the corrected infrared thermal image to obtain a tunnel enhancement feature map.

[0090] The ambient temperature compensation quantified value refers to the quantified value of the temperature interference of the acquisition environment on the acquisition device when the acquisition device acquires images.

[0091] In an exemplary embodiment, the step of obtaining the ambient temperature compensation quantized value includes:

[0092] Acquire temperature sequences corresponding to multiple locations in a cable tunnel; each temperature sequence includes tunnel temperatures collected at different times at the same location; average the temperature values ​​of the temperature sequence within a preset time window to obtain an average ambient temperature; the preset time window is determined based on the current collection time and the preset window length; process the average ambient temperature based on preset ambient temperature influencing parameters to obtain the actual temperature of the cable tunnel; and use the difference between the actual temperature and the average ambient temperature as a quantitative value for ambient temperature compensation.

[0093] The step of determining the preset time window includes: when the time length between the current collection moment and the initial moment is less than the preset window length, the time window between the initial moment and the current collection moment is used as the preset time window; when the time length between the current collection moment and the initial moment is greater than or equal to the preset window length, the current collection moment is used as the end moment, the moment before the current collection moment and the preset window length away from the current collection moment is determined as the start moment, and the time between the start moment and the end moment is used as the preset time window.

[0094] The mean of the temperature sequence in the preset time window is calculated to obtain the ambient average temperature, and the ambient average temperature is mapped according to the first ambient temperature influencing parameter and the second ambient temperature influencing parameter to obtain the actual temperature. The difference between the actual temperature and the ambient average temperature is used as the ambient temperature compensation quantization value.

[0095] For example, deployment at different locations in the cable tunnel A temperature sensor, Time point, The temperature value collected by the sensor is recorded as , ( is the total number of time points), .

[0096] The sliding average filter method is used to process the ambient temperature data, and the preset window length is set to In the Time point, average temperature The calculation formula is:

[0097]

[0098] in, Indicates the The average ambient temperature at a time point, is the number of sensors, is the size of the sliding window, It is Time point The temperature value collected by the sensor.

[0099] The actual temperature is calculated as:

[0100]

[0101] in, Indicates the target infrared thermal image Rank The actual temperature value corresponding to the column pixel, and These are coefficients determined through calibration experiments, namely, the first ambient temperature influencing parameter and the second ambient temperature influencing parameter.

[0102] The difference between the actual temperature and the average ambient temperature is used as the ambient temperature compensation quantification value.

[0103] In an embodiment of the present application, the average temperature is determined based on the temperature sequences corresponding to multiple positions in the cable tunnel in combination with a preset time window, so that the collected temperature of the cable tunnel can be more comprehensively and realistically characterized; based on the preset ambient temperature influencing parameters, the ambient average temperature is mapped to obtain the actual temperature of the cable tunnel, thereby improving the speed and authenticity of obtaining the actual temperature; then the difference between the actual temperature and the ambient average temperature is used as a quantitative value for ambient temperature compensation, providing a reliable quantitative reference basis for temperature compensation of the initial infrared thermal image.

[0104] The thermal radiation value in the initial infrared thermal image is corrected according to the ambient temperature compensation quantization value. This means that the current thermal radiation value of each pixel in the initial infrared thermal image is subtracted from the ambient temperature compensation quantization value to eliminate the influence of the environment on the temperature fluctuation of the infrared thermal image. The infrared thermal image obtained after thermal radiation correction can more accurately reflect the actual temperature conditions inside the cable tunnel.

[0105]

[0106] in, It is Rank The compensated temperature value of the column pixel, For the Rank The raw thermal radiation value of the column pixels.

[0107] After thermal radiation correction, the temperature information of the thermal image is more accurate, but there may be a problem of insufficient contrast, that is, the difference between different temperature areas is not obvious enough, and some potential temperature anomaly areas may be difficult to clearly identify. In an exemplary embodiment, the pixel values ​​in the corrected infrared thermal image are subjected to nonlinear enhancement processing, including:

[0108] For any pixel value, a contrast quantization value is determined based on a first difference between the pixel value and a first pixel adjustment parameter, and a second difference between the target pixel value and the first pixel adjustment parameter; the pixel value to be enhanced is calculated using the contrast quantization value as a base and the second pixel adjustment parameter as an exponent; the pixel value to be enhanced, the target pixel value, the second pixel adjustment parameter, and the third pixel adjustment parameter are fused to obtain the pixel value after enhancement processing.

[0109] Assuming that the first pixel adjustment parameter is a, the second pixel adjustment parameter is c, and the third pixel adjustment parameter is b, the expression of the nonlinear enhancement process is as follows:

[0110]

[0111] in, is the coordinate in the infrared thermal image before pixel enhancement The pixel value at It is the coordinate in the infrared thermal image after pixel enhancement The pixel value at .

[0112] In the embodiments of the present application, the infrared thermal image is subjected to nonlinear enhancement, and the contrast of the thermal image is optimized under the premise of maintaining the temperature gradient. In the detection of cable tunnels, this optimization can make the temperature difference more obvious and highlight the potential temperature anomaly area, thereby providing clearer image data for subsequent crack detection. For example, in some areas with small temperature differences, after the above-mentioned nonlinear enhancement processing, the slight temperature changes can be more clearly exhibited, which helps to find early signs of failure.

[0113] After the above processing, the enhanced target infrared thermal image is output, and the temperature resolution of ±0.5°C is retained. This temperature resolution can meet the requirements of temperature accuracy in cable tunnel detection, and ensures that important temperature information is not missed, thereby providing a reliable data basis for subsequent feature extraction and analysis.

[0114] In S303, an image feature of the initial infrared thermal image is extracted by using a semantic feature extraction model, and the extracted infrared thermal image feature map and the tunnel enhanced feature map are fused to obtain a target infrared thermal image.

[0115] In the embodiments of the present application, the semantic feature extraction model is a deep convolutional neural network model, which is used to extract rich semantic and structural features in the initial infrared thermal image to obtain an infrared thermal image feature map. Then, the infrared thermal image feature map and the tunnel enhanced feature map are fused to obtain a target infrared thermal image.

[0116] In the embodiments of the present application, for the initial infrared thermal image, the initial infrared thermal image is subjected to thermal radiation correction and pixel non-enhancement processing to obtain a tunnel enhanced feature map, and the semantic features of the initial infrared thermal image are extracted to obtain an infrared thermal image feature map. Then, the infrared thermal image feature map and the tunnel enhanced feature map are fused to obtain a target infrared thermal image. Since the tunnel enhanced feature map includes temperature-related information and the infrared thermal image feature map includes rich semantic and structural features, the target infrared thermal image obtained by combining the two can more comprehensively and accurately represent the features of the cable tunnel.

[0117] In one exemplary embodiment, as shown in FIG. 4, Figure 4 the foregoing S204 “fusing the weighted feature map with the channel radiation constraint value of the corresponding channel to obtain a tunnel fused initial feature map” includes the following steps:

[0118] In S401, the temperature gradient of the weighted feature map at each channel corresponding to a plurality of pixels is calculated.

[0119] The weighted feature map is decomposed according to the number of channels to obtain a weighted feature sub-map corresponding to each channel. For each weighted feature sub-map, the horizontal temperature gradient, the vertical temperature gradient of each pixel point in the weighted feature sub-map are calculated, and the comprehensive temperature gradient, i.e., the pixel temperature gradient of each pixel point, is obtained by combining the horizontal gradient and the vertical gradient.

[0120] S402, respectively, the mean value of the plurality of pixel temperature gradients corresponding to each channel is processed to obtain the channel radiation constraint value of each channel.

[0121] For any channel weighted feature sub-map, the pixel temperature gradient corresponding to each pixel point is obtained according to the channel value, the row and column index value, and the temperature gradient weight coefficient, and then the mean value of the pixel temperature gradient of each pixel point of the weighted feature sub-map is obtained. The channel radiation constraint value of the channel.

[0122] S403, the weighted feature map and the channel radiation constraint value of the corresponding channel are fused to obtain a tunnel fusion initial feature map.

[0123] The thermal radiation value corresponding to each pixel point of the weighted feature map is multiplied by the channel radiation constraint value of the corresponding channel to obtain a tunnel fusion initial feature map. In the embodiment of the application, based on the channel radiation constraint value of each channel, the weighted feature map is fused and processed to suppress thermal radiation noise interference, so that the effectiveness of the tunnel fusion initial feature is ensured.

[0124] In an exemplary embodiment, as shown in Figure 5 Before the crack evaluation of the cable tunnel is performed to obtain the crack evaluation result of the cable tunnel, the method further includes:

[0125] S501, the correlation between the temperature values of the pixel points in the tunnel fusion enhanced feature map is analyzed to determine an initial temperature anomaly region set in the tunnel fusion enhanced feature map.

[0126] For each pixel point, the temperature gradient parameter of each pixel point is calculated, and the pixel points greater than the temperature gradient threshold value are filtered out from the tunnel fusion enhanced feature map, and then the filtered pixel points are subjected to connected domain analysis and marked as initial temperature anomaly regions to obtain the initial temperature anomaly region set.

[0127] S502, the region greater than the preset area threshold value is filtered out from the initial temperature anomaly region set as a candidate temperature anomaly region.

[0128] The preset area threshold value is a lower limit value of the connected domain area set according to the empirical value, which can be 3x3 pixels. The regions smaller than the preset area threshold value are removed from the initial temperature anomaly region to obtain the candidate temperature anomaly region.

[0129] S503, aggregate each candidate temperature abnormal region, the average temperature of each candidate temperature abnormal region, and the temperature change rate of each candidate temperature abnormal region to obtain temperature information.

[0130] For each candidate temperature abnormal region, the average value of the temperature of each pixel point in the candidate temperature abnormal region, i.e., the average temperature, is calculated, and the ratio of the temperature difference of the candidate temperature abnormal region at adjacent time points to the time interval, i.e., the temperature change rate, is calculated. The average temperature and the temperature change rate of each candidate temperature abnormal region are taken as a set of temperature information.

[0131] In the embodiments of the present application, considering the influence of temperature abnormalities in the cable tunnel on the cable tunnel fault, the regions of temperature abnormalities in the cable tunnel are extracted, thereby providing a reference range for troubleshooting of abnormal conditions of the cable tunnel.

[0132] In one exemplary embodiment, as shown in Figure 6 The temperature information includes at least one temperature abnormal region, and the crack information includes at least one crack region. Based on the temperature information and the crack information in the tunnel fusion enhanced feature map, crack evaluation of the cable tunnel is performed to obtain a crack evaluation result of the cable tunnel, including:

[0133] S601, from the crack set in each of the crack regions, at least one target crack is screened.

[0134] According to the complexity of each crack in the crack set and / or the crack length, each crack is evaluated, and the crack that meets the preset evaluation standard is determined as the target crack.

[0135] S602, for any target crack, the crack region corresponding to the target crack is compared with each temperature abnormal region to determine at least one target temperature abnormal region corresponding to the target crack.

[0136] For each crack region corresponding to each target crack, the crack region is compared with each temperature abnormal region to determine a temperature abnormal region that intersects the crack region and has a region intersection ratio greater than a preset determination threshold, i.e., a target temperature abnormal region.

[0137] It should be noted that the number of target temperature abnormal regions can be 0, 1, or multiple. In the embodiments, the determination step of the crack evaluation result is described in the scenario where the number of target temperature abnormal regions includes at least one.

[0138] S603, for any target temperature abnormal region, if the average temperature and the temperature change rate of the target temperature abnormal region meet a preset temperature threshold, an evaluation result is generated; the evaluation result includes the coordinates of the target temperature abnormal region and the crack region coordinates of the target crack.

[0139] The preset temperature threshold includes a mean temperature lower limit value and a temperature change rate lower limit value. For any target temperature abnormal region, if the mean temperature of the target temperature abnormal region is greater than the mean temperature lower limit value and the temperature change rate is greater than the temperature change rate lower limit value, an evaluation result is generated. In addition, alarm information can also be generated to further improve the crack maintenance efficiency of the cable tunnel.

[0140] In the embodiments of the present application, from two dimensions of the crack attribute (crack length and crack complexity index) of the target crack and the temperature attribute (mean temperature and temperature change rate) of the target temperature region, double-threshold evaluation is performed on each target temperature abnormal region corresponding to the target crack, to improve the reliability of the evaluation result. Further, the coordinates of the target temperature abnormal region and the target crack coordinates are carried in the evaluation result to indicate accurate maintenance of the cable tunnel.

[0141] In one exemplary embodiment, as shown in Figure 7 The method comprises the following steps:

[0142] S701, obtaining the crack length and the crack complexity index of each crack.

[0143] For any crack, the crack is processed by a crack length calculation model to obtain the crack length of the crack, and the crack is processed by a morphological processing model and a branch statistical model to obtain the morphological complexity and the number of branches of the target crack. Then, the two are normalized and weighted to obtain the crack complexity index of the crack.

[0144] S702, determining the crack as a target crack if the crack length of the crack is greater than a preset length threshold and the crack complexity of the crack is greater than a preset complexity threshold.

[0145] For any crack, if the crack length of the crack is greater than a preset crack length and the crack complexity index of the crack is greater than a preset crack complexity index, the crack is determined as a target crack. In the embodiments of the present application, the crack is evaluated from two dimensions of the crack length and the crack complexity index to quickly and accurately determine the target crack.

[0146] In one exemplary embodiment, as shown in Figure 8 The method comprises the following steps:

[0147] S801, obtaining multi-modal data, the multi-modal data comprising: an initial infrared thermal image and environmental temperature sensor data.

[0148] S802, inputting the initial infrared thermal image into a semantic feature extraction model to obtain a backbone feature map.

[0149] The semantic feature extraction model can be a fourth convolutional block of a VGG16 model.

[0150] S803, determining an ambient temperature compensation quantization value according to the mean value of the ambient temperature sensor data in the preset time window and the predicted actual value corresponding to the mean value.

[0151] S804, performing temperature compensation on the thermal radiation value of the initial infrared thermal image according to the ambient temperature compensation quantization value.

[0152] S805, performing nonlinear enhancement on the initial infrared thermal image after temperature compensation.

[0153] S806, fusing the backbone feature map and the infrared thermal image after nonlinear enhancement to obtain a target infrared thermal image.

[0154] S807, inputting the target infrared thermal image into a feedback attention memory (FAM) model, weighting and fusing the extracted spatial attention weight map and the initial infrared thermal image to obtain a first feature map.

[0155] ① Feature map convolution processing.

[0156] Let the input VGG16 fourth convolutional block output backbone feature map be , wherein is the height of the feature map, is the width of the feature map, is the number of channels of the feature map. Let be input into two parallel convolutional layers respectively:

[0157] The first convolutional layer uses convolutional kernels, the number of convolutional kernels is (e.g. ), the convolution operation is denoted as , and the feature map is obtained. The second convolutional layer uses convolutional kernels, the number of convolutional kernels is (e.g. ), the convolution operation is denoted as , and the feature map is obtained.

[0158] ② Feature splicing and fusion.

[0159] Splice and in the channel dimension to obtain . Then pass through a convolutional layer (the number of convolutional kernels is , for example fusion is performed, denoted as , to obtain the fused feature map .

[0160] ③Global average pooling.

[0161] The global average pooling operation is performed on , denoted as , to obtain a one-dimensional vector : wherein, , represents the element in the i-th channel of the j-th row and the k-th column in .

[0162] ④Generating spatial attention weight map.

[0163] The one-dimensional vector is input into a first fully connected layer with an output dimension of to obtain an intermediate vector . Then is input into a second fully connected layer with an output dimension of , and is normalized by a Sigmoid function to obtain a spatial attention weight map :

[0164] ⑤Feature map weighting.

[0165] The spatial attention weight map is expanded to the same dimension as the backbone feature map , denoted as (the weight map of each channel is the same), and then element-wise multiplied with the original backbone feature map to obtain a weighted feature map : wherein, represents element-wise multiplication.

[0166] S808, based on the first feature map, calculates a temperature gradient and substitutes the temperature gradient into a radiation constraint formula to obtain a radiation constraint value for each channel, and then applies the radiation constraint values of the channels to the first feature map to obtain a second feature map with radiation constraint.

[0167] The spatial attention mechanism finally outputs the weighted feature map , which is taken as the input basis of the radiation constraint link. Since ​​​Critical areas associated with cracks have been highlighted, which can provide more targeted characterization information for radiation constraints.

[0168] Next, combine the feature map output by spatial attention To redefine the local temperature gradient .for Each channel in , calculate the local temperature gradient of each pixel in the image on the channel in space. For example, the finite difference method is used to approximate the local temperature gradient, where the horizontal gradient and the local gradient calculation formula are as follows:

[0169]

[0170] In the vertical direction, the local gradient calculation formula is as follows:

[0171]

[0172] The comprehensive gradient and local gradient calculation formula are as follows:

[0173]

[0174] Combined with the redefined local temperature gradient, a new radiation constraint formulation is generated:

[0175]

[0176] in express Middle Row, No. Column, No. The temperature value corresponding to each channel, is the average temperature of the channel, is the temperature variance of the channel, is a constant, is the temperature gradient weight coefficient. It can be defined according to actual conditions, for example, to represent a small temperature fluctuation threshold.

[0177] In order to obtain the comprehensive radiation constraint value of each channel, all pixels in each channel are Perform an average calculation: Finally, the calculated channel average radiation constraint value Feature map applied to the output of spatial attention For Each element in , multiply it by the corresponding channel radiation constraint value , and obtain the characteristic map after radiation constraint processing :

[0178]

[0179] Through the above steps, the radiation constraint link is regenerated on the basis of the spatial attention mechanism. The new radiation constraint combines the features of the spatial attention output, which can more accurately capture the temperature anomaly information of the crack area, thereby improving the accuracy and reliability of the cable tunnel detection.

[0180] After the spatial attention and radiation constraint processing, the output feature map with radiation constraint is obtained, and the size remains unchanged, but the channel number is compressed to 256. The compression of the channel number is to reduce the data amount and improve the calculation efficiency, while retaining the key feature information. This output feature serves as the initialization input of the four-way strip convolution, laying the foundation for further extracting multi-directional crack features.

[0181] S809, the second feature map is directionally expanded.

[0182] The input feature map is rotated in four directions {0°, 45°, 90°, 135°}. In this way, the feature map can be analyzed from different angles to fully capture the expansion characteristics of the crack in each direction. During the rotation process, methods such as bilinear interpolation are used to resample the pixel values to ensure the quality of the image and the integrity of the feature information.

[0183] S810, the expanded image is subjected to strip convolution operation.

[0184] In each rotated direction, strip convolution operation with a hole rate of 3 is independently performed. A hole rate of 3 means that the convolution kernel samples every 3 pixels during convolution operation, which can expand the receptive field of convolution without increasing the size of the convolution kernel, thereby capturing more extensive feature information. The strip convolution operation formula is:

[0185]

[0186] wherein represents the feature map after rotation , is the center coordinate of the current convolution kernel, and are the sampling intervals of the convolution kernel in the horizontal and vertical directions, respectively, is the weight coefficient of the convolution kernel. Through this operation, the crack features in each direction can be extracted. In this way, we can obtain four directional outputs , , and .

[0187] S811, fuse the image obtained by the strip convolution operation to obtain a multi-directional enhanced feature map.

[0188] In the feature fusion stage, the outputs of the four directional strip convolution operations are merged by weighted summation to obtain a multi-directional enhanced feature map .

[0189] Let the weights assigned to the outputs of each direction be , and satisfy , . For each position in the multi-directional enhanced feature map , the height direction index , the width direction index , and the channel direction index , the value is obtained by weighted summation of the values of the corresponding positions of the four directional strip convolution outputs, and the formula is as follows:

[0190]

[0191] The above fusion method can integrate the crack features of each direction, so that the feature map contains more comprehensive and rich information. The size of the final output multi-directional enhanced feature map is 80x60x128, and this size of feature map can provide more effective feature information for the subsequent temperature branch and crack branch.

[0192] Output and connection: the multi-directional enhanced feature map is input into the temperature branch and the crack branch at the same time, providing data support for subsequent temperature anomaly judgment and crack length prediction.

[0193] S812, according to the temperature anomaly area and the crack area in the multi-directional enhanced feature map, determine the temperature anomaly area corresponding to each crack.

[0194] After four-way mixed strip convolution processing, the multi-directional enhanced feature map (size is 80x60x128) is input into the temperature branch and the crack branch at the same time. The temperature anomaly area in the multi-directional enhanced feature map is extracted by the temperature branch, and the crack area in the multi-directional enhanced feature map is determined by the crack branch.

[0195] Through the spatial correlation detection module, determine the temperature anomaly area to which each crack belongs.

[0196] First, the regional boundary definition is performed: ① Temperature anomaly region: the boundary box of each region is obtained by connected component analysis, which is represented as a rectangular coordinate range (x_min, y_min, x_max, y_max). ② Crack region: the minimum circumscribed rectangle of each crack is calculated, which is also represented as a boundary box. ③ Overlap detection algorithm: the intersection over union (IoU) is used to determine whether the crack is associated with the temperature region.

[0197]

[0198] Determination threshold: if IoU ≥ 0.3 (i.e., the overlap area ratio ≥ 30%), it is considered that the crack belongs to the temperature region. Multi-region association: a crack may be associated with multiple temperature regions, and all possible combinations need to be recorded.

[0199] S813, for any crack, if the crack corresponds to a temperature anomaly region, it is determined whether the crack length, crack complexity index, average temperature of the temperature anomaly region, and temperature change rate of the temperature anomaly region all satisfy the corresponding threshold conditions. If yes, the evaluation result and alarm information are generated, and the evaluation result includes the position information of the crack.

[0200] The judgment conditions of the alarm information are as follows:

[0201]

[0202] wherein, Alarm = 1 is an alarm; Alarm = 0 is no alarm. is the crack length of the crack, is the crack complexity index of the crack. is the temperature change rate of the temperature anomaly region, is the average temperature of the temperature anomaly region. is the average temperature alarm threshold (the value range is 55 - 65℃, which can be adjusted according to the actual cable tunnel operating environment and equipment characteristics), is the temperature change rate alarm threshold (obtained by statistical analysis of a large number of cable tunnel temperature change data under normal and fault conditions), is the crack length alarm threshold (set to be greater than or equal to 1.2m, which can be adjusted according to the cable tunnel structure safety standard), is the crack complexity index alarm threshold (determined based on the shape research of different severity cracks).

[0203] S814, for any crack, if the crack corresponds to multiple temperature anomaly regions, if the crack length, crack complexity index, average temperature and temperature change rate of any temperature anomaly region of the crack meet the corresponding threshold conditions, generate evaluation results and alarm information, the evaluation results include the position information of the crack.

[0204] For many-to-many scene processing, it can be divided into the following two cases:

[0205] One-to-many scene: if a crack is associated with multiple temperature regions, as long as any temperature region meets the condition, an alarm is triggered.

[0206] Many-to-one scene: if multiple cracks are associated with the same temperature region, each crack needs to be evaluated to see if it meets the condition.

[0207] Alarm signal logic triggers an alarm (Alarm = 1) as long as there is at least one crack-temperature region combination that meets the condition. If no combination meets the condition, no alarm is given (Alarm = 0). The positioning information generation rule is as follows: temperature hotspot coordinates: take the centroid of the temperature region that meets the condition (calculate by weighted average of pixel coordinates). Crack coordinates: take the geometric center of the crack that meets the condition (calculate by the center point of the minimum bounding rectangle). If multiple combinations meet the condition, multiple combinations are output.

[0208] The cable tunnel crack evaluation method in the embodiments of the present application has at least the following advantages:

[0209] (1) Thermal radiation features are preserved and information is accurate.

[0210] To solve the problems of loss of thermal radiation features and distortion of crack edge temperature gradient information, by collecting and analyzing environmental temperature data, the interference of environmental temperature fluctuations on thermal images is eliminated to ensure accurate temperature information; nonlinear enhancement optimizes image contrast while preserving temperature gradient, making small temperature changes clear. After this series of processing, the thermal radiation features are effectively preserved, and the distortion problem of crack edge temperature gradient information is significantly improved, providing a high-quality data basis for subsequent detection and analysis.

[0211] (2) Joint modeling is realized to improve comprehensive analysis and hazard judgment ability.

[0212] Multi-modal fusion analysis: In view of the problem in the prior art that the structural crack morphology feature and the temperature anomaly radiation feature cannot be jointly modeled, and the cable tunnel health status is difficult to comprehensively evaluate from multiple dimensions, the VGG16 fourth convolution block extracted structural feature is deeply fused with the preprocessed thermal radiation data through the fusion radiation constraint FAM mechanism. In the spatial attention link, based on the powerful feature extraction capability of the convolutional neural network, the crack related area is accurately located; the temperature gradient prior information is ingeniously introduced in the radiation constraint link, so that the model can simultaneously consider the crack morphology and temperature anomaly features, realize the joint modeling of the two key features, and comprehensively analyze the cable tunnel status from multiple dimensions.

[0213] (3) Optimize feature capture and information processing to reduce distortion.

[0214] Multi-directional crack feature capture, four-way mixed strip convolution technology is adopted, the input feature map is rotated in {0°, 45°, 90°, 135°} four directions, and strip convolution operation with a hole rate of 3 is independently performed in each direction. Through direction expansion and strip convolution operation, crack propagation features in different directions can be comprehensively captured, effectively solving the defects in the prior art in the detection of diagonal cracks, making the feature capture more comprehensive and accurate.

[0215] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the direction of the arrow, these steps are not necessarily executed in sequence according to the direction of the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or steps or stages in other steps.

[0216] Based on the same inventive concept, the present embodiment also provides a cable tunnel crack detection device for implementing the above-mentioned cable tunnel crack detection method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more cable tunnel crack detection device embodiments provided below can refer to the limitations of the cable tunnel crack detection method in the foregoing, which will not be repeated here.

[0217] In one exemplary embodiment, as Figure 9As shown, a cable tunnel crack detection device is provided, comprising: an image acquisition module 901, a feature extraction module 902, a feature fusion module 903, a feature enhancement module 904 and a tunnel assessment module 905, wherein:

[0218] Image acquisition module 901, used to acquire infrared thermal image of the target in the cable tunnel;

[0219] The feature extraction module 902 is used to extract a spatial weight feature map of the target infrared thermal image using a preset tunnel feature extraction model, and fuse the target infrared thermal image with the spatial weight feature map to obtain a weighted feature map;

[0220] The feature fusion module 903 is used to fuse the weighted feature map with the channel radiation constraint value of the corresponding channel to obtain a tunnel fusion initial feature map;

[0221] A feature enhancement module 904 is configured to rotate the initial tunnel fusion feature map according to a preset rotation angle set, and fuse the image features in the rotated initial tunnel fusion feature map to obtain an enhanced tunnel fusion feature map;

[0222] The tunnel assessment module 905 is used to perform crack assessment on the cable tunnel based on the temperature information and crack information in the tunnel fusion enhancement feature map to obtain a crack assessment result of the cable tunnel.

[0223] In an exemplary embodiment, the feature fusion module 903 includes: a gradient calculation unit, a constraint value calculation unit, and a radiation constraint unit, wherein:

[0224] A gradient calculation unit is used to calculate the temperature gradients of multiple pixels corresponding to each channel of the weighted feature map;

[0225] A constraint value calculation unit is used to perform mean processing on the temperature gradients of multiple pixels corresponding to each channel to obtain the channel radiation constraint value of each channel;

[0226] The radiation constraint unit is used to fuse the weighted feature map with the channel radiation constraint value of the corresponding channel to obtain the tunnel fusion initial feature map.

[0227] In an exemplary embodiment, the tunnel assessment module 905 includes: a correlation analysis unit, a region screening unit, and a temperature summary unit, wherein:

[0228] A correlation analysis unit is used to perform correlation analysis on the temperature values ​​between the positions of each pixel in the tunnel fusion enhancement feature map to determine the initial temperature anomaly area set in the tunnel fusion enhancement feature map;

[0229] A region screening unit is used to screen out regions larger than a preset area threshold from the initial temperature anomaly region set as candidate temperature anomaly regions;

[0230] The temperature summarizing unit is used to summarize each candidate temperature anomaly region, the average temperature of each candidate temperature anomaly region, and the temperature change rate of each candidate temperature anomaly region to obtain temperature information.

[0231] In an exemplary embodiment, the temperature information includes at least one temperature anomaly region, and the crack information includes at least one crack region; the tunnel assessment module 905 further includes: a crack screening unit, a region comparison unit, and a crack assessment unit, wherein:

[0232] a crack screening unit, configured to screen at least one target crack from a set of cracks in each crack region;

[0233] A region comparison unit is used to compare, for any target crack, a crack region corresponding to the target crack with each temperature anomaly region, and determine at least one target temperature anomaly region corresponding to the target crack;

[0234] The crack assessment unit is used to generate an assessment result for any target temperature anomaly area if the average temperature and temperature change rate of the target temperature anomaly area meet the preset temperature threshold; the assessment result includes the coordinates of the target temperature anomaly area and the crack area coordinates of the target crack.

[0235] In an exemplary embodiment, the crack screening unit is further configured to obtain the crack length and crack complexity index of each crack; and determine a crack having a crack length greater than a preset length threshold and a crack complexity greater than a preset complexity threshold as a target crack.

[0236] In an exemplary embodiment, the image acquisition module includes: an initial image acquisition unit, an image correction unit, and a target image determination unit, wherein:

[0237] An initial image acquisition unit, used for acquiring an initial infrared thermal image of the cable tunnel;

[0238] An image correction unit is used to correct the thermal radiation value in the initial infrared thermal image based on a preset ambient temperature compensation quantization value, and perform nonlinear enhancement processing on the pixel values ​​in the corrected infrared thermal image to obtain a tunnel enhancement feature map;

[0239] The target image determination unit is used to extract the image features of the initial infrared thermal image using a semantic feature extraction model, and fuse the extracted infrared thermal image feature map with the tunnel enhancement feature map to obtain a target infrared thermal image.

[0240] In an example embodiment, the image acquisition module further comprises: a sequence acquisition unit, a temperature calculation unit, a temperature processing unit, and a compensation temperature determination unit, wherein:

[0241] The sequence acquisition unit is configured to acquire temperature sequences corresponding to a plurality of positions in the cable tunnel; each temperature sequence comprises tunnel temperatures collected at different time instants at the same position;

[0242] The temperature calculation unit is configured to perform mean value calculation on temperature values of the temperature sequences within a preset time window to obtain an environmental average temperature; the preset time window is determined according to a current collection time instant and a preset window length;

[0243] The temperature processing unit is configured to process the environmental average temperature based on a preset environmental temperature influence parameter to obtain an actual temperature of the cable tunnel;

[0244] The compensation temperature determination unit is configured to take a difference between the actual temperature and the environmental average temperature as an environmental temperature compensation quantitative value.

[0245] In an example embodiment, the image correction unit is further configured to, for any pixel value, determine a contrast quantitative value according to a first difference between the pixel value and the first pixel adjustment parameter, and a second difference between the target pixel value and the first pixel adjustment parameter; calculate an enhanced pixel value with the contrast quantitative value as a base number and the second pixel adjustment parameter as an index; and fuse the enhanced pixel value, the target pixel value, the second pixel adjustment parameter, and the third pixel adjustment parameter to obtain an enhanced processed pixel value.

[0246] The above-mentioned modules in the cable tunnel crack detection device can be realized by software, hardware, or a combination thereof, in whole or in part. The above-mentioned modules can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform operations corresponding to the above-mentioned modules.

[0247] In an example embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above-mentioned method embodiments when executing the computer program.

[0248] In an embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program is executed by a processor to implement the steps in the above-mentioned method embodiments.

[0249] In an embodiment, a computer program product is provided, comprising a computer program, and the computer program is executed by a processor to implement the steps in the above-mentioned method embodiments.

[0250] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0251] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present application.

[0252] The above-described embodiments are merely illustrative of several embodiments of the present application, which are described in more detail and in a specific manner, but should not be construed as limiting the scope of the patent of the present application. It should be noted that, for those of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method of detecting cracks in a cable tunnel, characterized by, The method comprises: acquire a target infrared thermal image of a cable tunnel; extract a spatial weight feature map of the target infrared thermal image through a preset tunnel feature extraction model, and fuse the target infrared thermal image and the spatial weight feature map to obtain a weighted feature map; fuse the weighted feature map and a channel radiation constraint value of a corresponding channel to obtain a tunnel fusion initial feature map; rotate the tunnel fusion initial feature map according to a preset rotation angle set, and fuse image features in the rotated tunnel fusion initial feature map to obtain a tunnel fusion enhanced feature map; based on temperature information and crack information in the tunnel fusion enhanced feature map, perform crack evaluation on the cable tunnel to obtain a crack evaluation result of the cable tunnel.

2. The method of claim 1, wherein, The fusion of the weighted feature map and the channel radiation constraint value of the corresponding channel to obtain the tunnel fusion initial feature map comprises: calculate a plurality of pixel temperature gradients of the weighted feature map corresponding to each channel; respectively, the plurality of pixel temperature gradients corresponding to each channel are subjected to mean value processing to obtain the channel radiation constraint value of each channel; fuse the weighted feature map and the channel radiation constraint value of the corresponding channel to obtain the tunnel fusion initial feature map.

3. The method of claim 1, wherein, Before the crack evaluation on the cable tunnel to obtain the crack evaluation result of the cable tunnel, the method further comprises: correlation analysis is performed on temperature values between pixel point positions in the tunnel fusion enhanced feature map to determine an initial temperature abnormal region set in the tunnel fusion enhanced feature map; regions greater than a preset area threshold are screened out from the initial temperature abnormal region set as candidate temperature abnormal regions; the candidate temperature abnormal regions, the average temperature of each candidate temperature abnormal region, and the temperature change rate of each candidate temperature abnormal region are summarized to obtain the temperature information.

4. The method according to any one of claims 1 to 3, characterized in that, The temperature information comprises at least one temperature abnormal region, and the crack information comprises at least one crack region; the crack evaluation on the cable tunnel based on the temperature information and the crack information in the tunnel fusion enhanced feature map to obtain the crack evaluation result of the cable tunnel comprises: at least one target crack is screened from a crack set in each crack region; for any target crack, the crack region corresponding to the target crack is compared with each temperature abnormal region to determine at least one target temperature abnormal region corresponding to the target crack; for any target temperature abnormal region, if the average temperature and the temperature change rate of the target temperature abnormal region meet a preset temperature threshold, an evaluation result is generated; the evaluation result comprises coordinates of the target temperature abnormal region and crack region coordinates of the target crack.

5. The method of claim 4, wherein, The screening of at least one target crack from a crack set in each crack region comprises: acquire the crack length and crack complexity index of each crack; cracks with a crack length greater than a preset length threshold and a crack complexity greater than a preset complexity threshold are determined as the target crack.

6. The method according to any one of claims 1 to 3, characterized in that, The acquisition of the target infrared thermal image of the cable tunnel comprises: acquire an initial infrared thermal image of the cable tunnel; correct a thermal radiation value in the initial infrared thermal image based on a preset environmental temperature compensation quantization value, and perform nonlinear enhancement processing on a pixel value in the corrected infrared thermal image to obtain a tunnel enhanced feature map; extract image features of the initial infrared thermal image by using a semantic feature extraction model, and fuse the extracted infrared thermal feature map and the tunnel enhanced feature map to obtain the target infrared thermal image.

7. The method of claim 6, wherein, The step of acquiring the environmental temperature compensation quantization value comprises: acquiring temperature sequences corresponding to multiple positions in the cable tunnel; each temperature sequence comprises tunnel temperatures collected at different time instants at the same position; performing mean value calculation on temperature values of the temperature sequences within a preset time window to obtain an environmental average temperature; the preset time window is determined according to a current collection time instant and a preset window length; processing the environmental average temperature based on a preset environmental temperature influence parameter to obtain an actual temperature of the cable tunnel; taking a difference value between the actual temperature and the environmental average temperature as the environmental temperature compensation quantization value.

8. The method of claim 6, wherein, The nonlinear enhancement processing on the pixel value in the corrected infrared thermal image comprises: for any pixel value, determining a contrast quantization value according to a first difference value between the pixel value and a first pixel adjustment parameter, and a second difference value between a target pixel value and the first pixel adjustment parameter; calculating a pixel value to be enhanced with the contrast quantization value as a base and a second pixel adjustment parameter as an index; fusing the pixel value to be enhanced, the target pixel value, the second pixel adjustment parameter, and a third pixel adjustment parameter to obtain an enhanced pixel value.

9. A cable tunnel crack detection apparatus characterized by comprising: The device comprises: an image acquisition module configured to acquire a target infrared thermal image of a cable tunnel; a feature extraction module configured to extract a spatial weight feature map of the target infrared thermal image by using a preset tunnel feature extraction model, and fuse the target infrared thermal image and the spatial weight feature map to obtain a weighted feature map; a feature fusion module configured to fuse the weighted feature map and a channel radiation constraint value of a corresponding channel to obtain a tunnel fusion initial feature map; a feature enhancement module configured to rotate the tunnel fusion initial feature map according to a preset set of rotation angles, and fuse image features in the rotated tunnel fusion initial feature map to obtain a tunnel fusion enhanced feature map; a tunnel evaluation module configured to perform crack evaluation on the cable tunnel based on temperature information and crack information in the tunnel fusion enhanced feature map to obtain a crack evaluation result of the cable tunnel. 10.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-9. The processor implements the steps of the method of any one of claims 1 to 8 when executing the computer program.

Images