Insulator material defect detection method, system, device, medium and program

Abstract

The invention relates to an insulator material defect detection method, system, device, medium and program, and the method comprises the steps: carrying out the denoising filtering and image standardization operation of an insulator image obtained in advance, and obtaining a standard insulator image; performing multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set; performing global attention coding on the insulator image feature set to obtain an attention image feature set; extracting global channel image features from the insulator image feature set, and performing channel fusion and spatial fusion on the global channel image features to obtain perception image features; and performing layer-by-layer feature decoding on the attention image feature set by using the perception image features to obtain a defect detection result, thereby improving the insulator defect detection efficiency.

Description

Technical Field

[0001] The present disclosure relates to the field of defect detection technology, and in particular to an insulator material defect detection method, system, device, medium and program. Background Art

[0002] Electric locomotives obtain high-voltage electricity from overhead lines through pantographs, and use insulators on the roof to ensure that the high-voltage current will not be transmitted to the car body, protecting the safety of equipment and personnel. However, the working environment of insulators is exposed to harsh climatic conditions all year round, and they are prone to aging, damage or pollution, which may lead to a decrease in insulation performance. In order to ensure the safety of locomotives, defect detection of insulator materials is required.

[0003] Traditional defect detection technology is a defect detection method based on ultrasonic detection, that is, by emitting sound waves and receiving reflected waves, detecting anomalies in the sound wave propagation process to identify defects inside the material. In practical applications, ultrasonic detection requires a contact probe, which makes it difficult to achieve online detection of high-altitude roof insulators. In addition, ultrasonic detection is not effective for insulators with complex shapes, which may lead to low efficiency in defect detection. Summary of the invention

[0004] The present invention discloses an insulator material defect detection method, system, device, medium and program, which improves the U-type network with a residual network, improves the scale of image features, improves the details of image features through an adaptive image convolution method, and improves the accuracy of defect feature extraction by combining a channel attention mechanism. While improving the accuracy, the network parameters are greatly reduced, thereby improving the efficiency of defect detection.

[0005] In a first aspect, the present disclosure provides an insulator material defect detection method, comprising: performing denoising filtering and image standardization operations on a pre-acquired insulator image to obtain a standard insulator image; performing multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set; performing global attention encoding on the insulator image feature set to obtain an attention image feature set; extracting global channel image features from the insulator image feature set, and performing channel fusion and spatial fusion on the global channel image features to obtain perceived image features; and performing layer-by-layer feature decoding on the attention image feature set using the perceived image features to obtain a defect detection result.

[0006] In some embodiments, the denoising filtering and image standardization operations are performed on the pre-acquired insulator image to obtain a standard insulator image, including: bilateral filtering the pre-acquired insulator image to obtain a denoised insulator image; cropping and size stretching the denoised insulator image to obtain a scaled insulator image; performing color space conversion on the scaled insulator image to obtain an insulator color space; performing brightness equalization on the insulator color space to obtain a balanced color space; performing image conversion on the balanced color space to obtain a balanced insulator image; and performing edge detection and image segmentation operations on the balanced insulator image to obtain a standard insulator image.

[0007] In some embodiments, performing multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set includes: performing primary convolution on the standard insulator image to obtain primary insulator features; performing self-attention convolution on the primary insulator features to obtain self-attention insulator features; performing secondary convolution on the self-attention insulator features to obtain secondary insulator features; performing residual connection on the secondary insulator features using the standard insulator image to obtain insulator image features; and performing iterative convolution on the insulator image features to obtain an insulator image feature set.

[0008] In some embodiments, the global attention encoding of the insulator image feature set to obtain the attention image feature set includes: selecting the insulator image features in the insulator image feature set one by one as the target image features, performing channel convolution on the target image features to obtain convolution image features; performing probability normalization calculation on the convolution image features to obtain feature probability distribution, and performing attention weighted operations on the target image features according to the feature probability distribution to obtain weighted image features; performing feature regularization and feature activation operations on the weighted image features to obtain activated image features; performing residual connection on the activated image features using the target image features to obtain attention image features; and aggregating the attention image features of all target image features in the insulator image feature set into an attention image feature set.

[0009] In some embodiments, the channel fusion and spatial fusion of the global channel image features to obtain perceived image features include: feature slicing the global channel image features to obtain a sliced ​​image feature group; extracting image channel weights from the sliced ​​image feature group; linearly transforming the image channel feature group using the image channel weights to obtain a channel fusion feature group; extracting spatial dimension weights from the channel fusion feature group; linearly transforming the channel fusion feature group using the spatial dimension weights to obtain a spatial fusion feature group; and feature splicing the spatial fusion feature group to obtain perceived image features.

[0010] In some embodiments, the use of the perceived image features to perform layer-by-layer feature decoding on the attention image feature set to obtain a defect detection result includes: sorting the attention image feature set in order of feature size from small to large to obtain an attention image feature sequence; selecting the attention image features in the attention image feature sequence one by one as the target attention feature, and using the feature size of the target attention feature as the target feature size; upsampling the perceived image features according to the target feature size to obtain a target upsampled feature; performing feature fusion on the target attention feature and the target upsampled feature to obtain a target defect feature; determining whether the target attention feature is the last attention image feature in the attention image feature set; if not, updating the perceived image feature using the target defect feature, and returning to the step of selecting the attention image features in the attention image feature sequence one by one as the target attention feature; if so, performing feature decoding and feature activation on the target defect feature to obtain a defect detection result.

[0011] In a second aspect, the present disclosure provides an insulator material defect detection system, comprising: a denoising and filtering module, used to perform denoising and filtering and image standardization operations on a pre-acquired insulator image to obtain a standard insulator image; a feature extraction module, used to perform multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set; an attention encoding module, used to perform global attention encoding on the insulator image feature set to obtain an attention image feature set; a feature fusion module, used to extract global channel image features from the insulator image feature set, and perform channel fusion and spatial fusion on the global channel image features to obtain perceived image features; a layer-by-layer decoding module, used to perform layer-by-layer feature decoding on the attention image feature set using the perceived image features to obtain a defect detection result.

[0012] In a third aspect, the present disclosure provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of an insulator material defect detection method described in the above aspect.

[0013] In a fourth aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of an insulator material defect detection method described in the above aspect.

[0014] In a fifth aspect, the present disclosure provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of an insulator material defect detection method described in the above aspect.

[0015] The present invention provides an insulator material defect detection method, system, device, medium and program. By performing noise removal filtering and image standardization operations, the noise of the insulator image can be reduced, and the image can be standardized to a preset input standard, so as to facilitate subsequent defect recognition operations through the model. By performing multi-channel convolution and residual connection, the image features of insulators of different sizes and channels can be flexibly captured, and the information flow can be maintained through residual connection to enhance the details of the image features.

[0016] By performing global attention encoding on the insulator image feature set, global feature modeling can be achieved, the relationship between each position in the image and the global information can be captured, and image features with higher robust performance can be obtained. The steps of self-attention operation are reduced, and the efficiency of feature extraction is improved. By performing channel fusion and spatial fusion, long-distance dependencies in image features can be captured, thereby enhancing the feature details and global information of image features and improving the accuracy of subsequent defect detection.

[0017] By performing layer-by-layer feature decoding, the defect features can be upsampled layer by layer, and the attention mechanism can be used for feature fusion to improve the accuracy of defect detection, thereby improving the efficiency of defect detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present disclosure will be described in more detail below based on embodiments and with reference to the accompanying drawings:

[0019] Figure 1 A schematic flow chart of an insulator material defect detection method according to a first embodiment of the present disclosure is shown.

[0020] Figure 2 A schematic diagram of the network structure for performing multi-channel convolution and residual connection in the first embodiment of the present disclosure is shown.

[0021] Figure 3 A schematic diagram of the network structure for global attention encoding in the first embodiment of the present disclosure is shown.

[0022] Figure 4 A schematic flow chart of an insulator material defect detection method according to a second embodiment of the present disclosure is shown.

[0023] Figure 5 A functional module diagram of an insulator material defect detection system according to a third embodiment of the present disclosure is shown.

[0024] In the drawings, the same reference numerals are used for the same components, and the drawings are not drawn to scale. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the technical solution of the present disclosure, and to fully understand and implement how the present disclosure applies technical means to solve technical problems and achieve the corresponding technical effects, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only embodiments of a part of the present disclosure, not all of the embodiments. The embodiments of the present disclosure and the various features in the embodiments can be combined with each other without conflict, and the technical solutions formed are all within the scope of protection of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present disclosure.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0027] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] Embodiment 1

[0029] Figure 1 A schematic flow chart of a method for detecting defects in insulator materials provided in an embodiment of the present disclosure.

[0030] like Figure 1 As shown, a method for detecting defects in insulator materials comprises:

[0031] S1. Perform denoising and filtering and image standardization operations on the pre-acquired insulator image to obtain a standard insulator image.

[0032] In detail, the insulator image is an image obtained by using an image acquisition device to capture the image of the insulator on the locomotive. The insulator is a component used in electrical systems, usually used to isolate the current conduction between a high-voltage power supply and the ground. For the roof of the locomotive, the insulator is located between the pantograph and the roof of the electric locomotive, playing a key insulating and supporting role.

[0033] In an embodiment of the present invention, the denoising filtering and image standardization operations are performed on the pre-acquired insulator image to obtain a standard insulator image, including: bilateral filtering is performed on the pre-acquired insulator image to obtain a denoised insulator image; image cropping and size stretching are performed on the denoised insulator image to obtain a scaled insulator image; color space conversion is performed on the scaled insulator image to obtain an insulator color space; brightness equalization is performed on the insulator color space to obtain a balanced color space; image conversion is performed on the balanced color space to obtain a balanced insulator image; edge detection and image segmentation are performed on the balanced insulator image to obtain a standard insulator image.

[0034] In detail, the image cropping refers to cropping out irrelevant areas in the image, and the size stretching refers to stretching the size of the denoised insulator image after image cropping to a preset model input size. The size stretching method can be GIMP or a CSS-based image stretching method, and the size can be (512﹡512﹡3), where 512 refers to the pixel length and pixel width of the image, and 3 refers to the number of channels of the image.

[0035] Specifically, the Lab color space conversion algorithm or the HSV color space conversion algorithm can be used to perform color space conversion, the histogram equalization method can be used to perform brightness equalization, the image conversion method is the inverse method of the color space conversion, the sobel or canny operator can be used for edge detection, a mask can be generated according to the edge obtained by the edge detection, and the mask is used to perform image segmentation operations to obtain a standard insulator image.

[0036] In the embodiment of the present invention, by performing noise removal filtering and image standardization operations, the noise of the insulator image can be reduced, and the image can be standardized to a preset input standard, so as to facilitate subsequent defect recognition operations through the model.

[0037] S2. Perform multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set.

[0038] In detail, the insulator image feature set is a set consisting of feature maps of the standard insulator images with different channel sizes.

[0039] In an embodiment of the present invention, performing multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set includes: performing primary convolution on the standard insulator image to obtain primary insulator features; performing self-attention convolution on the primary insulator features to obtain self-attention insulator features; performing secondary convolution on the self-attention insulator features to obtain secondary insulator features; performing residual connection on the secondary insulator features using the standard insulator image to obtain insulator image features; and performing iterative convolution on the insulator image features to obtain an insulator image feature set.

[0040] In detail, the primary convolution includes performing a convolution operation on the standard insulator image to obtain insulator features; performing a batch normalization operation on the insulator features to obtain normalized insulator features; and performing feature activation on the normalized insulator features to obtain primary insulator features, wherein the feature activation can be performed using a ReLU activation function.

[0041] Specifically, the self-attention convolution includes convolving the primary insulator feature using a self-attention mechanism to obtain a self-attention insulator feature, performing a batch normalization operation on the self-attention insulator feature to obtain a self-attention normalized insulator feature; performing feature activation on the self-attention normalized insulator feature to obtain a self-attention insulator feature;

[0042] In detail, the secondary convolution method is consistent with the primary convolution method and will not be repeated here. The residual connection refers to adding the stimulus insulator feature to the standard insulator image, thereby retaining the original information of the input and solving the problem of gradient explosion in the deep network.

[0043] Specifically, refer to Figure 2 As shown, it is a schematic diagram of the network structure of the multi-channel convolution and residual connection, including the primary convolution, the self-attention convolution, the secondary convolution and the residual connection, wherein conv refers to the volume operation, BN refers to the batch normalization operation, Relu refers to the ReLU activation function, Involution refers to the self-attention convolution, and the arrow jumping from the input to the output refers to the residual connection.

[0044] In detail, the iterative convolution refers to updating the standard insulator image using the insulator image feature, returning to the step of performing primary convolution on the standard insulator image to obtain primary insulator features, until a preset number of iterations is reached, and then aggregating the insulator image features obtained each time into an insulator image feature set, wherein the number of channels of each insulator image feature in the insulator image feature set gradually increases, for example, the size of the first insulator image feature is (256﹡256﹡32), the size of the second insulator image feature is (128﹡128﹡64), and the size of the third insulator image feature is (64﹡64﹡128).

[0045] In the embodiment of the present invention, by performing multi-channel convolution and residual connection, the image features of insulators of different sizes and channels can be flexibly captured, and the information flow is maintained through the residual connection to enhance the details of the image features.

[0046] S3. Perform global attention encoding on the insulator image feature set to obtain an attention image feature set.

[0047] In an embodiment of the present invention, the global attention encoding is performed on the insulator image feature set to obtain the attention image feature set, including: selecting the insulator image features in the insulator image feature set one by one as the target image features, performing channel convolution on the target image features to obtain convolution image features; performing probability normalization calculation on the convolution image features to obtain feature probability distribution, and performing attention weighted operation on the target image features according to the feature probability distribution to obtain weighted image features; performing feature regularization and feature activation operations on the weighted image features to obtain activated image features; performing residual connection on the activated image features using the target image features to obtain attention image features; and aggregating the attention image features of all target image features in the insulator image feature set into an attention image feature set.

[0048] In detail, the channel convolution refers to adjusting the number of channels of the target image features using a 1 by 1 convolution layer, and the softmax algorithm can be used to perform probability normalization operations. The attention weighted operation refers to inputting the feature probability distribution into a fully connected layer or a convolution layer to obtain a set of attention weights, and using the obtained attention weights to perform feature weighted operations on the insulator image features to obtain weighted image features.

[0049] Specifically, the LayerNorm normalization layer can be used to perform feature value regularization operations, and the Relu activation function can be used to perform feature activation operations. The residual connection refers to feature fusion of the target image features and the activated image features.

[0050] For details, refer to Figure 3 As shown, it is a schematic diagram of the network structure for global attention encoding, wherein conv(1×1) refers to channel convolution, softmax refers to the probability normalization calculation, the × symbol in the figure refers to the attention weighted operation, LayerNorm refers to the eigenvalue regularization operation, Relu refers to the ReLU activation function, and the + symbol in the figure refers to the residual connection.

[0051] In an embodiment of the present invention, by performing global attention encoding on the insulator image feature set, global feature modeling can be achieved, the relationship between each position in the image and the global information can be captured, image features with higher robust performance can be obtained, the steps of self-attention operations can be reduced, and the efficiency of feature extraction can be improved.

[0052] S4. Extracting global channel image features from the insulator image feature set, and performing channel fusion and spatial fusion on the global channel image features to obtain perceived image features.

[0053] In detail, the global channel image feature is the insulator image feature with the largest number of channels in the insulator image feature set. For example, when the sizes of the insulator image features in the insulator image feature set are (256﹡256﹡32), (128﹡128﹡64) and (64﹡64﹡128), respectively, the size of the global channel image feature is (64﹡64﹡128).

[0054] In an embodiment of the present invention, the channel fusion and spatial fusion of the global channel image features to obtain perceived image features include: feature slicing the global channel image features to obtain a sliced ​​image feature group; extracting image channel weights from the sliced ​​image feature group; linearly transforming the image channel feature group using the image channel weights to obtain a channel fusion feature group; extracting spatial dimension weights from the channel fusion feature group; linearly transforming the channel fusion feature group using the spatial dimension weights to obtain a spatial fusion feature group; and feature splicing the spatial fusion feature group to obtain perceived image features.

[0055] In detail, the feature slicing refers to slicing the global channel image features into a sliced ​​image feature group consisting of multiple image features with fixed dimensions, and the feature splicing method is the inverse step of the feature slicing method.

[0056] Specifically, the image channel weight refers to the feature weight of each channel after global perception, and the spatial dimension weight refers to the feature weight of the image features in each spatial dimension. The image channel weight and spatial dimension weight can be extracted using the fully connected layer of the multi-layer perceptron, and the fully connected layer can be used for linear transformation.

[0057] In the embodiment of the present invention, by performing channel fusion and spatial fusion, long-distance dependencies in image features can be captured, thereby enhancing the feature details and global information of the image features and improving the accuracy of subsequent defect detection.

[0058] S5. Use the perceived image features to perform layer-by-layer feature decoding on the attention image feature set to obtain a defect detection result.

[0059] In detail, the defect detection result refers to the specific position of the corresponding defect in the insulator image and the type of the defect.

[0060] In an embodiment of the present invention, the method of using the perceived image features to perform layer-by-layer feature decoding on the attention image feature set to obtain a defect detection result includes: sorting the attention image feature set in order of feature size from small to large to obtain an attention image feature sequence; selecting the attention image features in the attention image feature sequence one by one as target attention features, and using the feature size of the target attention features as the target feature size; upsampling the perceived image features according to the target feature size to obtain target upsampled features; performing feature fusion on the target attention features and the target upsampled features to obtain target defect features; determining whether the target attention feature is the last attention image feature in the attention image feature set; if not, updating the perceived image features using the target defect feature, and returning to the step of selecting the attention image features in the attention image feature sequence one by one as the target attention features; if so, performing feature decoding and feature activation on the target defect feature to obtain a defect detection result.

[0061] In detail, the upsampling refers to enlarging the feature size of the perceived image feature to the target feature size, and the upsampling can be performed using nearest neighbor interpolation, bilinear interpolation or deconvolution methods, and the feature fusion method can be feature quantity addition.

[0062] Specifically, the convolutional layer of the decoder of the pre-trained UNeXt neural network can be used for feature decoding. UNeXt is an improved neural network based on the classic UNet network, focusing on efficient feature extraction and reconstruction in image processing tasks. The decoder part of UNeXt is the key stage for upsampling and feature fusion of input features. The decoder is responsible for restoring the low-resolution deep features extracted by the encoder to the high-resolution original image shape, and can use ReLU activation or sigmoid activation function for feature activation.

[0063] In the embodiment of the present invention, by performing layer-by-layer feature decoding, it is possible to achieve upsampling of defect features layer by layer, and combine feature fusion with the attention mechanism to improve the accuracy of defect detection, thereby improving the efficiency of defect detection.

[0064] Embodiment 2

[0065] On the basis of the above embodiments, in order to more clearly understand the present disclosure, refer to Figure 4 As shown, a second embodiment is used below to further explain the situation in which a more accurate defect recognition result is required according to the first embodiment of the present disclosure.

[0066] Figure 4 A schematic flow chart of a method for detecting defects in insulator materials provided in an embodiment of the present disclosure.

[0067] like Figure 4 As shown, a method for detecting defects in insulator materials comprises:

[0068] S41, performing noise removal filtering and image standardization operations on the pre-acquired insulator image to obtain a standard insulator image.

[0069] S42, performing multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set.

[0070] S43. Perform global attention encoding on the insulator image feature set to obtain an attention image feature set.

[0071] S44, extracting global channel image features from the insulator image feature set, and performing global attention encoding and defect feature decoding on the global channel image features to obtain attention defect image features.

[0072] In detail, the global channel image feature is the insulator image feature with the largest number of channels in the insulator image feature set. For example, when the sizes of the insulator image features in the insulator image feature set are (256﹡256﹡32), (128﹡128﹡64) and (64﹡64﹡128), respectively, the size of the global channel image feature is (64﹡64﹡128).

[0073] In an embodiment of the present invention, the global channel image features are globally attention encoded and defect features are decoded to obtain attention defect image features, including: feature slicing the global channel image features to obtain a slice image feature group; position encoding the slice image feature group to obtain a slice feature code; using a self-attention mechanism to calculate the self-attention weight of the slice feature code; using the self-attention weight to feature encode the slice feature code to obtain an encoded defect feature; and feedforward activation and residual connection are performed on the encoded defect feature to obtain an attention defect image feature.

[0074] In detail, the feature slicing refers to slicing the global channel image features into a sliced ​​image feature group consisting of multiple image features with fixed dimensions, and the feature splicing method is the inverse step of the feature slicing method.

[0075] Specifically, the position encoding refers to feature encoding according to the position of each feature in the slice image feature group among the global channel image features. The encoding layer of a pre-trained Transformer model can be used for feature encoding, and the decoding layer of the Transformer model can be used for feedforward activation and residual connection.

[0076] In the embodiment of the present invention, by performing global attention encoding and defect feature decoding, the defect features can be identified in combination with the self-attention mechanism, thereby improving the accuracy of defect detection.

[0077] S45. Use the attention-deficit image features to perform layer-by-layer feature decoding on the attention image feature set to obtain a defect detection result.

[0078] In detail, the defect detection result refers to the specific position of the corresponding defect in the insulator image and the type of the defect.

[0079] In an embodiment of the present invention, the method of using the attention defect image feature to perform layer-by-layer feature decoding on the attention image feature set to obtain a defect detection result includes: sorting the attention image feature set in order of feature size from small to large to obtain an attention image feature sequence; selecting the attention image features in the attention image feature sequence one by one as the target attention feature, and using the feature size of the target attention feature as the target feature size; upsampling the attention defect image feature according to the target feature size to obtain a target upsampled feature; performing feature fusion on the target attention feature and the target upsampled feature to obtain a target defect feature; determining whether the target attention feature is the last attention image feature in the attention image feature set; if not, updating the attention defect image feature using the target defect feature, and returning to the step of selecting the attention image features in the attention image feature sequence one by one as the target attention feature; if yes, performing feature decoding and feature activation on the target defect feature to obtain a defect detection result.

[0080] In detail, the upsampling refers to enlarging the feature size of the attention deficit image feature to the target feature size. The upsampling can be performed using nearest neighbor interpolation, bilinear interpolation or deconvolution methods, and the feature fusion method can be feature quantity addition.

[0081] Specifically, the convolutional layer of the decoder of the pre-trained UNeXt neural network can be used for feature decoding. UNeXt is an improved neural network based on the classic UNet network, focusing on efficient feature extraction and reconstruction in image processing tasks. The decoder part of UNeXt is the key stage for upsampling and feature fusion of input features. The decoder is responsible for restoring the low-resolution deep features extracted by the encoder to the high-resolution original image shape, and can use ReLU activation or sigmoid activation function for feature activation.

[0082] In the embodiment of the present invention, by performing layer-by-layer feature decoding, it is possible to achieve upsampling of defect features layer by layer, and combine feature fusion with the attention mechanism to improve the accuracy of defect detection, thereby improving the efficiency of defect detection.

[0083] Embodiment 3

[0084] Based on the above embodiments, Figure 5 This is a functional module diagram of an insulator material defect detection system provided by an embodiment of the present disclosure. Figure 5 As shown, an insulator material defect detection system comprises:

[0085] The insulator material defect detection system 500 described in this embodiment can be installed in an electronic device. According to the functions implemented, the insulator material defect detection system 500 may include a noise removal and filtering module 501, a feature extraction module 502, an attention encoding module 503, a feature fusion module 504 and a layer-by-layer decoding module 505. The module described in the present disclosure may also be referred to as a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0086] In this embodiment, the functions of each module / unit are as follows:

[0087] The noise removal and filtering module 501 is used to perform noise removal and image standardization operations on the pre-acquired insulator image to obtain a standard insulator image;

[0088] The feature extraction module 502 is used to perform multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set;

[0089] The attention encoding module 503 is used to perform global attention encoding on the insulator image feature set to obtain an attention image feature set;

[0090] The feature fusion module 504 is used to extract global channel image features from the insulator image feature set, and perform channel fusion and spatial fusion on the global channel image features to obtain perceived image features;

[0091] The layer-by-layer decoding module 504 is used to perform layer-by-layer feature decoding on the attention image feature set using the perceived image features to obtain a defect detection result.

[0092] In detail, each module described in the insulator material defect detection system 500 described in the embodiment of the present disclosure adopts the same technical means as the insulator material defect detection method described in Example 1 when used, and can produce the same technical effects, which will not be repeated here.

[0093] Embodiment 4

[0094] On the basis of the above embodiments, this embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the above embodiments.

[0095] In some implementations of this embodiment, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the computer program is executed by a processor, the steps of the method described in the above embodiment are implemented.

[0096] In some implementations of this embodiment, a computer program product is provided, including a computer program, characterized in that when the computer program is executed by a processor, the steps of the method described in the above embodiment are implemented.

[0097] The processor may include, but is not limited to, one or more processors or microprocessors, etc. Each processor may be an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller, a microprocessor or other electronic components to execute the methods in the above embodiments.

[0098] The computer-readable storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, and the computer-readable storage medium may include but is not limited to, for example, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, computer storage media (e.g., hard disk, floppy disk, solid-state drive, removable disk, CD ROM, DVD ROM, Blu-ray disc, etc.).

[0099] The computer-readable storage medium may also store at least one computer executable program, such as a computer-readable instruction. The computer-readable storage medium includes, but is not limited to, for example, a volatile memory and / or a non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The computer-readable storage medium may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device runs the computer-readable instructions stored on the computer-readable storage medium, the various methods described above may be performed.

[0100] In addition, the computer device may also include (but not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (eg, keyboard, mouse, speaker, etc.), etc.

[0101] The processor may communicate with external devices via an I / O bus via a wired or wireless network.

[0102] In one embodiment, the at least one computer executable instruction may also be compiled into or constitute a software product / computer program product, wherein one or more computer executable instructions are executed by a processor to perform the various functions and / or method steps in the embodiments described in the present technology.

[0103] In the embodiments provided in the present disclosure, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to the multiple embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the above-mentioned module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0104] It should be noted that in the present disclosure, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element limited by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0105] Although the embodiments disclosed in the present disclosure are as above, the above contents are only embodiments adopted for facilitating the understanding of the present disclosure and are not intended to limit the present disclosure. Any technician in the technical field to which the present disclosure belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present disclosure, but the scope of patent protection of the present disclosure shall still be subject to the scope defined in the attached claims.

Claims

1. A method for detecting defects in insulator materials, characterized in that: include: Performing noise removal and image standardization operations on the pre-acquired insulator image to obtain a standard insulator image; Performing multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set; Performing global attention encoding on the insulator image feature set to obtain an attention image feature set; Extracting global channel image features from the insulator image feature set, and performing channel fusion and spatial fusion on the global channel image features to obtain perceived image features; The perceived image features are used to perform layer-by-layer feature decoding on the attention image feature set to obtain a defect detection result.

2. The insulator material defect detection method according to claim 1, characterized in that: The step of performing noise removal, filtering and image standardization operations on the pre-acquired insulator image to obtain a standard insulator image includes: Performing bilateral filtering on the pre-acquired insulator image to obtain a denoised insulator image; Performing image cropping and size stretching on the noise-removed insulator image to obtain a scaled insulator image; Performing color space conversion on the scaled insulator image to obtain an insulator color space; Performing brightness equalization on the insulator color space to obtain a balanced color space; Performing image conversion on the balanced color space to obtain a balanced insulator image; Edge detection and image segmentation operations are performed on the balanced insulator image to obtain a standard insulator image.

3. The insulator material defect detection method according to claim 1, characterized in that: The method of performing multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set includes: Performing primary convolution on the standard insulator image to obtain primary insulator features; Performing self-attention convolution on the primary insulator feature to obtain a self-attention insulator feature; Performing secondary convolution on the self-attention insulator feature to obtain a secondary insulator feature; Performing residual connection on the secondary insulator features using the standard insulator image to obtain insulator image features; The insulator image features are iteratively convolved to obtain an insulator image feature set.

4. The insulator material defect detection method according to claim 1, characterized in that: The step of performing global attention encoding on the insulator image feature set to obtain an attention image feature set includes: Selecting insulator image features in the insulator image feature set one by one as target image features, performing channel convolution on the target image features, and obtaining convolution image features; Performing probability normalization calculation on the convolution image features to obtain feature probability distribution, and performing attention weighting operation on the target image features according to the feature probability distribution to obtain weighted image features; Performing feature regularization and feature activation operations on the weighted image features to obtain activated image features; Performing a residual connection on the activation image feature using the target image feature to obtain an attention image feature; The attention image features of all target image features in the insulator image feature set are aggregated into an attention image feature set.

5. The insulator material defect detection method according to claim 1, characterized in that: The performing channel fusion and spatial fusion on the global channel image features to obtain the perceived image features includes: Performing feature slicing on the global channel image features to obtain a sliced ​​image feature group; Extracting image channel weights from the slice image feature group; Performing a linear transformation on the image channel feature group using the image channel weights to obtain a channel fusion feature group; Extracting spatial dimension weights from the channel fusion feature group; Performing a linear transformation on the channel fusion feature group using the spatial dimension weight to obtain a spatial fusion feature group; Feature splicing is performed on the spatial fusion feature group to obtain perceived image features.

6. The insulator material defect detection method according to claim 1, characterized in that: The step of using the perceived image features to perform layer-by-layer feature decoding on the attention image feature set to obtain a defect detection result includes: Sort the attention image feature set in ascending order of feature size to obtain an attention image feature sequence; Selecting the attention image features in the attention image feature sequence one by one as the target attention feature, and taking the feature size of the target attention feature as the target feature size; Upsampling the perceived image feature according to the target feature size to obtain a target upsampled feature; Performing feature fusion on the target attention feature and the target upsampling feature to obtain a target defect feature; Determining whether the target attention feature is the last attention image feature in the attention image feature set; If not, the perceived image feature is updated using the target defect feature, and the process returns to the step of selecting the attention image features in the attention image feature sequence one by one as the target attention feature; If so, feature decoding and feature activation are performed on the target defect feature to obtain a defect detection result.

7. An insulator material defect detection system, characterized in that: include: A noise removal and filtering module is used to perform noise removal and image standardization operations on the pre-acquired insulator image to obtain a standard insulator image; A feature extraction module, used for performing multi-channel convolution and residual connection on the standard insulator image to obtain an insulator image feature set; An attention encoding module, used for performing global attention encoding on the insulator image feature set to obtain an attention image feature set; A feature fusion module, used to extract global channel image features from the insulator image feature set, and perform channel fusion and spatial fusion on the global channel image features to obtain perceived image features; The layer-by-layer decoding module is used to use the perceived image features to perform layer-by-layer feature decoding on the attention image feature set to obtain a defect detection result.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the insulator material defect detection method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the insulator material defect detection method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the insulator material defect detection method according to any one of claims 1 to 6 are implemented.

Images