Multi-weather robust target detection method and device for power inspection and storage medium

Abstract

The application discloses a kind of multi-weather robust target detection method, device and storage medium of electric power inspection, it is related to image processing technical field, the method includes: obtaining the target image of electric power system;The target image is input into the target detection model of pre-training, and the detection result output by the target detection model is obtained;Wherein, the target detection model is: using pre-labeled sample target image to train the preset neural network model and obtain the model used for target detection, wherein the neural network model processes the sample target image, obtains global style feature and local semantic feature, and is trained by minimizing the orthogonal loss of global style feature and local semantic feature, to improve target detection precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a multi-weather robust target detection method, device and storage medium for power inspection. BACKGROUND

[0002] In the power system, target detection as the core basic task in power inspection aims to realize accurate positioning and identification of key targets in natural images. Thanks to the rapid development of deep learning technology, significant progress has been made in target detection in the past few decades. However, existing deep learning methods usually rely on the assumption that the distribution of training data and test data is consistent, which is often difficult to meet in practical applications. In the changing real environment, the image distribution will change significantly due to weather conditions, lighting conditions, imaging device differences and target appearance changes, etc., which will cause the performance of the deep learning-based detector to degrade seriously in the cross-domain scene, especially in the safety-critical scene, which limits its actual deployment and application.

[0003] Most of the generalization methods focus on aligning the feature space or extracting domain-invariant features, but visual features are extremely sensitive to low-level information such as illumination, texture, and resolution. When encountering extreme weather (such as rain, fog, snow) or severe sensor noise interference, the performance of the model is still difficult to guarantee, especially in real inspection scenarios, the degradation factors often preferentially affect the edge structure and low-frequency components of the image, thereby significantly weakening the target detection accuracy. SUMMARY

[0004] The present application aims to at least partially solve one of the technical problems in the related art. To this end, one object of the present application is to provide a multi-weather robust target detection method, device and storage medium for power inspection to improve target detection accuracy.

[0005] According to a first aspect of an embodiment of the present application, a multi-weather robust target detection method for power inspection is provided, the method comprising:

[0006] obtaining a target image of a power system;

[0007] inputting the target image into a pre-trained target detection model to obtain a detection result output by the target detection model;

[0008] wherein the target detection model is a model for target detection obtained by training a pre-set neural network model using pre-labeled sample target images, wherein the neural network model processes the sample target images to obtain global style features and local semantic features, and is trained by minimizing the orthogonal loss of the global style features and the local semantic features.

[0009] Optionally, before the neural network model processes the sample target image, the method further comprises:

[0010] performing image segmentation on the sample target image to obtain a target region and a background region of the sample target image;

[0011] performing local enhancement transformation on the target region and global transformation on the background region to obtain an enhanced sample target image.

[0012] Optionally, the neural network model is trained in the following manner:

[0013] determining a total loss based on the orthogonal loss and a detection loss of a detection result output by the neural network model;

[0014] adjusting model parameters of the neural network model according to the total loss until a preset training completion condition is reached.

[0015] Optionally, the orthogonal loss of the global style feature and the local semantic feature is generated in the following manner :

[0016]

[0017] wherein, denotes a batch size of the sample target image; and denote a global style feature and a local semantic feature of an i-th sample target image, respectively.

[0018] Optionally, the detection loss comprises a classification loss and a bounding box regression loss; the classification loss and the bounding box regression loss are obtained based on a detection result for the local semantic feature.

[0019] Optionally, the total loss further comprises a cross-scale consistency loss, which is obtained according to a bounding box prediction result of a multi-scale feature map of the local semantic feature.

[0020] Optionally, the total loss further comprises a cross-style consistency loss, which is obtained based on a class probability distribution of a different style image of the target image and a class probability distribution corresponding to the local semantic feature.

[0021] Optionally, the global style feature is obtained by processing a global statistical feature of the sample target image based on a mask attention mechanism.

[0022] According to a second aspect of the embodiment of the present application, a multi-weather robust target detection device for power inspection is provided, which comprises:​

[0023] an image acquisition device configured to acquire a target image of a power system;

[0024] a target detection device configured to input the target image into a pre-trained target detection model to obtain a detection result output by the target detection model;

[0025] The target detection model is a model for target detection obtained by training a preset neural network model using pre-labeled sample target images, wherein the neural network model processes the sample target images to obtain global style features and local semantic features, and is trained by minimizing the orthogonal loss of the global style features and the local semantic features.

[0026] According to a third aspect of the embodiments of the present application, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program.

[0027] In the scheme provided by the embodiments of the present application, the global style features are extracted from the sample target images in the process of training the model, and the influence caused by the environment, such as weather changes, will be reflected in the global style features. By minimizing the orthogonal loss of the global style features and the local semantic features, the space of the style and the semantic is ensured to be independent of each other, and the global style features are prevented from interfering with the semantic judgment. In this way, the model can detect the local semantic features more accurately, and the model trained in this way can improve the accuracy of target detection.

[0028] Additional aspects and advantages of the present application will be made apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 is a flowchart of a power inspection multi-weather robust target detection method provided by the embodiments of the present application;

[0030] Figure 2 is a flowchart of a first model training method provided by the embodiments of the present application;

[0031] Figure 3 is a flowchart of a second model training method provided by the embodiments of the present application;

[0032] Figure 4 is a flowchart of a third model training method provided by the embodiments of the present application;

[0033] Figure 5is a flowchart of a fourth model training method provided by an embodiment of the present application;

[0034] Figure 6 is a structural diagram of a multi-weather robust target detection device for power inspection provided by an embodiment of the present application. DETAILED DESCRIPTION

[0035] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, in which the same or similar notations represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0036] A multi-weather robust target detection method for power inspection, a storage medium and an electronic device of an embodiment of the present application are described below with reference to the accompanying drawings.

[0037] In an embodiment of the present application, referring to Figure 1 , a multi-weather robust target detection method for power inspection is provided, the method comprising the following steps S101-S102.

[0038] S101: obtaining a target image of a power system;

[0039] S102: inputting the target image into a pre-trained target detection model to obtain a detection result output by the target detection model;

[0040] The target detection model is a model for target detection obtained by training a preset neural network model using pre-labeled sample target images, wherein the neural network model processes the sample target images to obtain global style features and local semantic features, and is trained by minimizing the orthogonal loss of the global style features and the local semantic features.

[0041] The target image and the sample target image can be images obtained by photographing a scene where the power system exists using a photographing device, which can be a mobile phone, a camera, a camera, etc. The present application does not limit this.

[0042] For the sample target image, the targets existing in the image can be pre-labeled, including power equipment bodies such as towers, insulators, cross arms, etc., and can also include environmental risk targets such as bird nests and flying wires (such as kites and steel wires wrapped on the line). When labeling, the position and category of the target can be labeled.

[0043] In an embodiment of the present application, the global style features are obtained by processing global statistical features of the sample target image based on a mask attention mechanism.

[0044] As Figure 3 shown, in the target detection model, a semantic branch is arranged, and the semantic branch extracts local semantic features by using a deep convolution or a Transformer structure. The input of the semantic branch can be an original sample target image or an enhanced image in the following embodiments .

[0045] The specific extraction process can be as follows: first, the feature map of the sample target image is processed into (H·W) tokens with a length of C, to form a sequence:

[0046] Wherein, R represents a sample target image set; H and W respectively represent the height and width of the sample target image; and C represents the output channel number.

[0047] Then, Swin (Shifted Window, shifted window) / ViT (Vision Transformer)-w and other local window self-attention are used to perform self-attention within the window and shift-attention across the window, to output a token sequence . Subsequently, the tokens within each s×s window are averaged and pooled, where s is a preset length, to obtain a pooled vector (j=1, 2, …, n) to form a new semantic feature map :

[0048]

[0049] Statistical features such as channel mean, variance and Gram matrix are obtained by performing statistical aggregation on the high-level feature map corresponding to the entire sample target image, and the statistical features are split into a group of statistical tokens. Then, a mask attention mechanism is used to introduce a learnable mask M in the 0-1 range into the scoring matrix of multi-head self-attention, so as to selectively shield or strengthen the correlation between different statistical tokens. The weighted result output by the attention weight modulated by the soft-max normalization is the global style feature.

[0050] As Figure 4 shown, the enhanced image is input into the style branch, global statistical feature modeling is performed, statistical features are obtained, global style features are obtained through learnable mask attention, and a predicted mask, that is, the above-mentioned learnable mask, is output. The global style feature and the local semantic feature calculate an orthogonal loss , and the predicted mask is used to calculate a mask loss ​The specific calculation method is described in subsequent embodiments, which is not described here.

[0051] In this way, not only the modeling ability of long-range dependencies of self-attention is retained, but also the filtering of non-style information is realized through the mask, so that a discriminative style representation orthogonal to semantic features and independent of spatial position is obtained. The abstract ability of convolutional features is retained by using high-level feature maps, and the position information is eliminated by statistics, so that the style representation is more compact and easier to be orthogonal to semantic representation.

[0052] By setting the above style branch and semantic branch, the original image features in the target image are decomposed into a semantic-related subspace and a domain / style-related subspace, and are represented by local semantic features and global style features respectively. The core purpose is to let the subsequent target detection network only rely on high-order semantic information directly related to class discrimination, and eliminate low-order style information drifting with scene, weather and imaging device changes.

[0053] Target detection according to global style features and local semantic features: as shown in Figure 3 The local semantic features are input into the target detection network in the model, and the class of the target in the target image is predicted and output through classification. The class output by the model is the class with the maximum probability in the class probability distribution, and the position of the target in the target image is output through the bounding box regression.

[0054] In this way, the class and position of the detected target can be included in the detection result.

[0055] The above decoupling process can bring three quantifiable changes:

[0056] Domain discriminability significantly decreases: the local semantic features are input into the domain classifier, and the accuracy decreases from 86.3% to 21.4% (≈random), indicating that the domain information has been successfully eliminated;

[0057] Intra-class compactness is improved: t-SNE (t-distributed Stochastic Neighbor Embedding) visualization shows that the uncoupled features are mixed in a “stretching” shape in different domains; after decoupling, samples of the same class are clustered into a tight cluster, and the Silhouette coefficient increases from 0.27 to 0.53;

[0058] Cross-domain detection robustness is improved: in target detection, the local semantic features replace the original features, and the cross-domain meanAveragePrecision (cross-domain mAP) is improved by more than 4.8%, especially the false detection rate of small targets and adverse weather scenes is reduced by 11.2%.

[0059] These results indicate that the decoupled semantic features are "purer," retaining discriminative semantics and removing context-dependent noise related to domain, scale, and illumination, thereby enhancing generalization ability.

[0060] In the solution provided by this invention, global style features are extracted from the sample target image during the model training process, so that the influence of the environment, such as weather changes, will be reflected in the global style features. By minimizing the orthogonal loss of global style features and local semantic features, the space of style and semantics is ensured to be independent of each other, and the global style features are prevented from interfering with semantic judgment. In this way, the model can detect local semantic features more accurately, and the trained model can improve the accuracy of target detection.

[0061] In one embodiment of the present invention, the sample target images input into the neural network for training can first undergo data augmentation, specifically including:

[0062] Image segmentation is performed on the sample target image to obtain the target region and background region of the sample target image;

[0063] The target region is locally enhanced, and the background region is globally transformed to obtain the enhanced sample target image.

[0064] One approach is to use an instance segmentation model (such as Mask R-CNN or Segment Anything) to generate binary masks for each target. ( ) represents a pixel Belongs to the (One target), thereby separating the target area and the background area.

[0065] The target region is represented as:

[0066]

[0067] (Element-wise product extraction of target pixels);

[0068] The background area is represented as:

[0069] (The remaining area after removing all targets);

[0070] in, This represents the target image of the sample.

[0071] Perform local enhancement transformation on the target region, including:

[0072] Light intensity adjustment (brightness, contrast, hue), geometric perturbation (±5° rotation, small-range translation and scaling), and texture perturbation (Gaussian noise, local sharpening).

[0073] For the background region, simulate bad weather by global transformation, including:

[0074] Haze: large-scale Gaussian blur , where is a Gaussian kernel with variance .

[0075] Rain / snow: superimpose particle noise and add motion blur.

[0076] Strong light: global brightness / contrast enhancement or local highlight superposition.

[0077] Superimpose the enhanced target region and the background region according to the original spatial position to obtain the enhanced sample target image .

[0078] where,

[0079]

[0080] A (·) represents a local enhancement transformation, and G (·) represents a global transformation. Joint mask (assuming that the targets do not overlap, ). The data augmentation process is uniformly represented in matrix form as:

[0081]

[0082] In the above formulas, the letters represent the same parameters have the same meaning.

[0083] The image is segmented into a target region and a background region by using an instance segmentation model, and targeted enhancement is applied respectively, and finally fused to generate an enhanced image, simulating target detection in different environments and different styles. Under the limited training data set lacking of extreme weather data, high-precision target detection can still be realized. This method has good computing efficiency, low resource occupation, and is suitable for edge computing devices and actual engineering deployment, meeting the needs of real-time performance and convenient deployment.

[0084] By explicitly decoupling the style and semantic features of the data enhanced image, the performance degradation problem of target detection under complex weather and variable imaging conditions can be reduced, and the generalization performance of the model can be improved.

[0085] As shown in Figure 2 , the original image I can be a sample target image, or a target image input into the model after training for detection. The original image I is input into an instance segmentation model to generate a target mask , and the target region and the background region are separated in the data enhancement module.

[0086] For the target region , target-oriented enhancement is performed, that is, a local enhancement transformation is used, and a photometric / geometry / texture transformation is performed to obtain an enhanced target region ; for the background region , global weather simulation is performed, that is, a global transformation is used, including simulating haze / rain / snow / strong light, to obtain an enhanced background region ; after fusing the enhanced images, an enhanced image is obtained .

[0087] In an embodiment of the present application, the neural network model is trained in the following manner:

[0088] The total loss is determined based on the orthogonal loss and a detection loss of a detection result output by the neural network model;

[0089] The model parameters of the neural network model are adjusted according to the total loss until a preset training completion condition is reached.

[0090] The manner of adjusting the model parameters can be back propagation, and the preset number of iterations of the model or convergence is reached, and then the preset training completion condition is reached.

[0091] In an embodiment of the present application, the orthogonal loss of the global style feature and the local semantic feature is generated in the following manner:

[0092]

[0093] wherein, denotes the batch size of the sample target image; and denote the global style feature and the local semantic feature of the i-th sample target image, respectively.

[0094] In an embodiment of the present application, the detection loss includes a classification loss and a bounding box regression loss; the classification loss and the bounding box regression loss are obtained based on a detection result for the local semantic feature.

[0095] As shown in FIG. 1, the local semantic feature is processed through an RPN (Region Proposal Network) network and an ROI (Region of Interest) head to obtain a detection result including classification prediction and bounding box regression. Figure 3

[0096] ​​​The ROI head is the module responsible for processing the candidate regions generated by the RPN network. It includes a classification subnetwork and a regression subnetwork. The classification subnetwork classifies the targets within the candidate regions, and the regression subnetwork fine-tunes the target positions to optimize the accuracy of the bounding box coordinates.

[0097] In this case, the classification loss is obtained by calculating the cross-entropy. And calculate the bounding box regression loss. As shown below:

[0098]

[0099] The bounding box regression loss is calculated using the Smooth L1 loss function:

[0100]

[0101] in, This represents the number of candidate regions in the target sample image; C is the number of target categories. For classification prediction, it means that the first... The candidate region belongs to the first The probability of the category; This represents the number of positive samples, where positive samples are candidate regions marked as foreground areas whose intersection-union ratio (IU) with any ground truth bounding box is greater than a preset IU threshold. Since only positive samples require bounding box regression, in... Chinese Normalize. The loss function is Smooth L1. Indicates the predicted bounding box. Represents the actual bounding box. Indicates the first One valid label for each candidate region in Class C (including background).

[0102] The Smooth L1 loss function works when the error is small (less than a preset threshold). When the error is large, L2 penalty is used to maintain smoothness, and it degenerates into L1 to suppress outliers. It is more robust to bounding box regression than pure L2.

[0103] The predicted bounding box is the bounding box representing the target location predicted by the model during the bounding box regression process; the true bounding box is the pre-labeled bounding box.

[0104] In one embodiment of the present invention, the total loss further includes cross-scale consistency loss, which is obtained based on the bounding box prediction results of the multi-scale feature maps of local semantic features.

[0105] Cross-scale consistency loss The consistency of the detection results of the same target on different scale feature maps is constrained. The definition is as follows:

[0106]

[0107] Wherein, Indicates the total number of target instances participating in the calculation of the cross-scale consistency loss; 、 Respectively, the predicted bounding box of the same target on the and scale feature map; Indicates the calculation of the intersection over union.

[0108] As Figure 3 The target detection network extracts multi-scale feature maps from local semantic features. The extraction method can be a feature pyramid network (FPN, Feature Pyramid Networks), which combines high-level features and low-level features in local semantic features to construct multi-scale feature maps. The cross-scale bounding box prediction obtains the predicted bounding box on each scale feature map, and the intersection over union based on the predicted bounding box is calculated to obtain the cross-scale consistency loss.

[0109] In the convolutional network, the low-level features are the texture, edge and other pixel-level details extracted by the first few layers, and the high-level features are the abstract semantic information such as object shape and category obtained by deep aggregation.

[0110] In the power inspection scene, "context" not only refers to background semantics, that is, the appearance distribution of the target is affected by the information such as spatial geometry, semantic co-occurrence, scene globality and scale hierarchy, but also implies the key condition of scale hierarchy: the appearance and geometric information of the same target corresponding to the FPN scale feature map under different shooting distances or resolutions is significantly different. If the network does not explicitly align these differences, when the scene migration causes the scale distribution to drift, positioning drift and missed detection often occur - this is the "context dependence" emphasized in this work.

[0111] In order to weaken this dependence, a cross-scale consistency loss is introduced in the training stage, which directly constrains the predicted boxes of the same target on different scale feature maps to be consistent, so as to maximize the intersection over union by minimizing the loss function, and explicitly align the geometric representation under different scale contexts, and internalize the "scale" context as an optimized geometric consistency constraint in the feature space.

[0112] In one embodiment of the present application, the total loss further includes a cross-style consistency loss, which is obtained based on the class probability distribution of different style images of the target image and the class probability distribution corresponding to the local semantic features.

[0113] The predicted class distribution of different style images of the same scene generated by data enhancement is respectively and As shown in Figure 5 , different style images are generated by the original image I,

[0114] Different style images refer to enhanced images generated by different enhancement transformation methods such as photometric adjustment and geometric perturbation. For example, for the same original image I, a photometrically adjusted image X for I is generated, and an image Y for I is generated by geometric perturbation. Then, X and Y are different style images of I.

[0115] In the foregoing embodiment, either the local enhancement transformation or the global transformation can be used to generate different style images, or a combination of any of the local enhancement transformation and the global transformation can be used for generation, and the embodiments of the present application do not limit this.

[0116] The predicted class distribution refers to the probability vector formed by the normalized prediction probability of each class in the C classes in the classification prediction, as shown in Figure 5 It can be obtained based on the detection network, and the detection network can be a model trained in the prior art, such as a YOLO model.

[0117] Figure 3 In the foregoing embodiment, the original prediction in the class probability distribution is the predicted class of the sample target image, and the probability of belonging to different classes can also be represented by the probability vector of the class distribution.

[0118] The cross-style consistency loss is defined as The KL divergence form is:

[0119]

[0120] Wherein, is the number of sample target images, is the number of target classes, and respectively represent the prediction probability of the target belonging to the class of the different style image of the i-th sample target image, and the predicted class of the sample target image.

[0121] The loss constrains the model to maintain prediction consistency under style changes, thereby improving the robustness of the detection model to image style changes.

[0122] In one embodiment, to further clarify the attention of the model to the target region, a learnable mask attention supervision is proposed. The target detection model outputs the predicted mask of the target sample image, and calculates the mask loss with the real target mask , a binary cross-entropy supervision is performed:

[0123]

[0124] wherein, denotes a set of pixel positions in the sample target image; denotes a real mask pixel value, which is a pixel value of a pre-labeled mask; denotes a predicted mask pixel value, which is a pixel value of a predicted mask output by the target detection model.

[0125] In this case, a binary cross-entropy loss can be added to the total loss for training.

[0126] In an embodiment, to reduce the interference of low-level style changes on high-level semantic features, a statistical difference loss of low-level convolutional features can be added to the total loss as an auxiliary loss.

[0127] Statistical difference loss is defined as follows:

[0128]

[0129] wherein, by global statistical feature modeling shown in Figure 4 , the statistical mean and variance of low-level convolutional features of two different style images of the same sample target image are obtained as and , that is, low-level style statistical features, so as to calculate the statistical difference loss .

[0130] In summary, the total loss can be set as:

[0131]

[0132] The loss weight coefficients , , , , , , are hyperparameters, and the meanings of other parameters are consistent with the parameters of the same letter in the foregoing embodiments, which are adjusted according to actual requirements to balance the contributions of various optimization objectives. As shown in Figure 5 , the total loss is the sum of the cross-style consistency loss, the mask loss, the orthogonal loss, the statistical difference loss, the classification loss, the regression loss, and the cross-scale consistency loss, and the parameters are updated by back propagation of the total loss to complete the model training process.

[0133] Corresponding to the method embodiments described above, in an embodiment, referring to Figure 6The application further provides a multi-weather robust target detection device for power inspection, which comprises:

[0134] An image acquisition device 601 is configured to acquire a target image of a power system.

[0135] A target detection device 602 is configured to input the target image into a pre-trained target detection model to obtain a detection result output by the target detection model.

[0136] The target detection model is a model for target detection, which is obtained by training a preset neural network model using pre-labeled sample target images, wherein the neural network model processes the sample target images to obtain global style features and local semantic features, and is trained by minimizing the orthogonal loss of the global style features and the local semantic features.

[0137] According to a third aspect of the embodiments of the application, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the multi-weather robust target detection method for power inspection according to any of the above embodiments.

[0138] It should be noted that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a list of executable instructions for implementing logic functions, which can be embodied in any computer readable medium for use by or in connection with an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor or other system that can fetch the instructions from an instruction execution system, apparatus or device and execute the instructions, or in conjunction with these instruction execution systems, apparatus or devices. For the purpose of this specification, the "computer readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by or in connection with an instruction execution system, apparatus or device, or in conjunction with these instruction execution systems, apparatus or devices. More specific examples (non-exhaustive list) of computer readable medium include the following: electrical connections having one or more wires (electronic devices), portable computer diskettes (magnetic devices), random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memories), fiber optic devices, and portable compact disk read-only memories (CD ROMs). In addition, the computer readable medium can even be paper or other suitable medium on which the program can be printed, because the program can be electronically obtained, for example, by optical scanning of the paper or other medium, followed by editing, interpreting or otherwise processing the program as necessary, and then storing it in a computer memory.

[0139] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, through software or firmware in storage media which are executable by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following technologies, known in the art, or their combinations, can be employed: discrete logic circuitry having logic gates for implementing logic functions upon an application of data signals, application-specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field-programmable gate arrays (FPGA), and so on.

[0140] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are contained in at least one embodiment or example of the present application. In the present description, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0141] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0142] In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0143] In the present application, unless specifically defined otherwise, the terms "mounting", "connected", "connecting", "fixed", and the like should be construed broadly, for example, can be fixed connection, can also be detachable connection, or integral; can be mechanical connection, can also be electrical connection; can be directly connected, can also be indirectly connected through an intermediate medium, can be the internal communication of two elements or the interaction relationship of two elements, unless specifically defined otherwise. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0144] In the present application, unless specifically defined otherwise, the first feature is "on" or "under" the second feature can be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, the first feature "above", "over" and "on" the second feature can be that the first feature is directly above or obliquely above the second feature, or only indicates that the horizontal height of the first feature is higher than that of the second feature. The first feature "below", "under" and "under" the second feature can be that the first feature is directly below or obliquely below the second feature, or only indicates that the horizontal height of the first feature is less than that of the second feature.

[0145] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A multi-weather robust target detection method for power inspection, characterized in that, The method comprises: acquiring a target image of a power system; inputting the target image into a pre-trained target detection model to obtain a detection result output by the target detection model; The target detection model is a model for target detection obtained by training a preset neural network model using pre-labeled sample target images, wherein the neural network model processes the sample target images to obtain global style features and local semantic features, and is trained by minimizing the orthogonal loss of the global style features and the local semantic features. The global style feature and the local semantic feature are generated in the following manner : in, This indicates the batch size of the sample target images; and They represent the first Global style features and local semantic features of a sample target image.

2. The method of claim 1, wherein, Before the neural network model processes the sample target images, the method further comprises: performing image segmentation on the sample target images to obtain target regions and background regions of the sample target images; performing local enhancement transformation on the target regions and global transformation on the background regions to obtain enhanced sample target images.

3. The method of claim 1, wherein, The neural network model is trained in the following manner: determine a total loss based on the orthogonal loss and a detection loss of the detection result output by the neural network model; adjust model parameters of the neural network model according to the total loss until a preset training completion condition is reached.

4. The method of claim 3, wherein, The detection loss includes a classification loss and a bounding box regression loss; the classification loss and the bounding box regression loss are obtained based on the detection result of the local semantic features.

5. The method of claim 3, wherein, The total loss further includes a cross-scale consistency loss obtained according to the bounding box prediction results of multi-scale feature maps of the local semantic features.

6. The method of claim 3, wherein, The total loss further includes a cross-style consistency loss obtained based on the class probability distribution of different style images of the target image and the class probability distribution corresponding to the local semantic features.

7. The method of claim 1, wherein, The global style features are obtained by processing global statistical features of the sample target images based on a mask attention mechanism.

8. A multi-weather robust target detection apparatus for power patrol, characterized in that, The device comprises: an image acquisition device for acquiring a target image of a power system; a target detection device for inputting the target image into a pre-trained target detection model to obtain a detection result output by the target detection model; The target detection model is a model for target detection obtained by training a preset neural network model using pre-labeled sample target images, wherein the neural network model processes the sample target images to obtain global style features and local semantic features, and is trained by minimizing the orthogonal loss of the global style features and the local semantic features; The global style feature and the local semantic feature are generated in the following manner : in, This indicates the batch size of the sample target images; and They represent the first Global style features and local semantic features of a sample target image.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the power inspection multi-weather robust target detection method according to any one of claims 1-7.

Images