Power transmission line detection method and device based on image recognition, equipment and medium

Abstract

Embodiments of the present application disclose a power transmission line detection method, device and equipment based on image recognition and a medium. The method comprises: acquiring image data of a power transmission line to be detected, performing image quality enhancement processing on the image data to obtain target image data, and performing feature extraction to obtain feature data corresponding to the image data; inputting the feature data into a preset region proposal network to obtain a feature map of the image data and target proposal region data corresponding to the feature map; inputting the feature map and the target proposal region data into a preset region of interest pooling layer to obtain pooling information, and completing detection according to the pooling information. Through image quality enhancement processing on the image data, the feature data in the image data is enhanced, so that the feature data in the image data is easier to be extracted, thereby reducing the omission of the feature data during image recognition, ensuring the effect of image recognition, and making the detection process of the power transmission line to be detected more reliable and accurate.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method, apparatus, equipment and medium for detecting power transmission lines based on image recognition. Background Technology

[0002] Power transmission lines are a crucial component of power transmission and distribution systems. However, due to long-term use and the influence of the external environment, power transmission lines may develop various defects, such as poor contact, breakage, and wear. These defects can lead to serious consequences such as power system failures, line outages, and even fires. Therefore, accurate and efficient defect detection of power transmission lines is essential.

[0003] Over the past few decades, many traditional methods for detecting defects in power transmission lines have been proposed and applied. However, these methods typically rely on manual feature engineering and simple machine learning algorithms, which limit their detection performance and are easily affected by complex scenes and changes in lighting. Therefore, accurate detection of defects in power transmission lines remains a challenging problem. Summary of the Invention

[0004] In view of this, the present invention provides a method, apparatus, equipment and medium for detecting transmission lines based on image recognition, which solves the problem of low accuracy in the prior art when detecting defects in transmission lines.

[0005] To achieve one, some, or all of the above objectives, or other objectives, the present invention proposes a transmission line detection method based on image recognition, comprising:

[0006] Image data of the transmission line to be inspected is acquired, and image quality enhancement processing is performed on the image data to obtain target image data;

[0007] Feature extraction is performed on the target image data to obtain feature data corresponding to the image data;

[0008] The feature data is input into a preset region proposal network to obtain a feature map of the image data and target proposal region data corresponding to the feature map.

[0009] The feature map and the target proposed region data are input into a preset region of interest pooling layer to obtain pooling information, and the transmission line to be detected is detected based on the pooling information.

[0010] Optionally, the step of performing image quality enhancement processing on the image data to obtain target image data includes:

[0011] The image data is input into a densely connected convolutional block in a preset image quality enhancement network, and the first feature data of the image data is extracted through a two-dimensional convolutional layer in the densely connected convolutional block.

[0012] Based on the preset feature fusion mechanism of the densely connected convolutional blocks, the difference data between different densely connected convolutional blocks in the preset image quality enhancement network is obtained, and the second feature data of the image data is obtained based on the difference data.

[0013] The first feature data and the second feature data are input into the global feature fusion layer in the preset image quality enhancement network to obtain the enhanced data of the image data.

[0014] The target image data is constructed based on the enhanced data.

[0015] Optionally, the step of extracting features from the target image data to obtain feature data corresponding to the image data includes:

[0016] The target image data is input into a preset target detection network to obtain the initial feature data of the target image data;

[0017] The initial feature data is input into a preset cascaded bidirectional feature pyramid network to obtain bidirectional feature data of the target image data;

[0018] The bidirectional feature data are fused using a weighted fusion method to obtain fused feature data, which is then used as the feature data corresponding to the image data.

[0019] Optionally, the step of inputting the target image data into a preset target detection network to obtain the initial feature data of the target image data includes:

[0020] The resolution data of the target image data is scaled according to a preset joint coefficient using a joint scaling method to obtain the input data;

[0021] The network parameters of the preset target detection network are scaled based on the preset joint coefficients to obtain the target network.

[0022] The input data is input into the target network to obtain the initial feature data of the target image data.

[0023] Optionally, the step of inputting the initial feature data into a preset cascaded bidirectional feature pyramid network to obtain bidirectional feature data of the target image data includes:

[0024] Delete the node in the preset cascaded bidirectional feature pyramid network that has only one input data to obtain the target cascaded bidirectional feature pyramid network.

[0025] The bidirectional paths in the target cascaded bidirectional feature pyramid network are used as feature network layers, and the initial feature data is input into the feature network layers to obtain the bidirectional feature data of the target image data.

[0026] Optionally, the step of inputting the feature data into a preset region proposal network to obtain a feature map of the image data and target proposal region data corresponding to the feature map includes:

[0027] The feature data is input into the first branch network of the proposed network of the preset region, and the feature map of the image data is determined through the preset normalization layer in the first branch network.

[0028] The feature data is input into the second branch network of the preset region proposal network, and the initial proposed region data corresponding to the feature map is corrected through the preset convolutional layer in the second branch network to obtain the target proposed region data.

[0029] Optionally, the step of inputting the feature map and the target proposed region data into a preset region of interest pooling layer to obtain pooling information, and detecting the transmission line to be detected based on the pooling information, includes:

[0030] The proposed region corresponding to the target proposed region data is mapped and reduced through the first processing layer of the preset interest region pooling layer to obtain the reduced proposed region data.

[0031] The proposed narrowing region data is fitted using the preset regression algorithm of the preset interest region pooling layer to obtain fitted data, and the fitted data is used as the first information in the pooling information.

[0032] The feature map is divided into standard feature maps of a preset specification by the second processing layer of the preset region of interest pooling layer;

[0033] The standard feature map is classified by the preset classification layer of the preset interest region pooling layer to obtain a defect feature map, and the defect feature map is used as the second information in the pooling information.

[0034] The transmission line to be tested is detected based on the defect feature map and the fitted data.

[0035] On the other hand, embodiments of this application provide a transmission line detection device based on image recognition, the device comprising:

[0036] The data acquisition module is used to acquire image data of the transmission line to be inspected and to perform image quality enhancement processing on the image data to obtain target image data;

[0037] The feature extraction module is used to extract features from the target image data to obtain feature data corresponding to the image data;

[0038] A labeling module is used to input the feature data into a preset region proposal network to obtain a feature map of the image data and target proposal region data corresponding to the feature map;

[0039] The detection module is used to input the feature map and the target proposed region data into a preset region of interest pooling layer to obtain pooling information, and to detect the transmission line to be detected based on the pooling information.

[0040] On the other hand, embodiments of this application provide a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program is executed by a processor to perform the steps of the image recognition-based transmission line detection method described above.

[0041] Implementing the embodiments of the present invention will have the following beneficial effects:

[0042] By acquiring image data of the transmission line to be inspected and performing image quality enhancement processing on the image data, target image data is obtained. Feature extraction is performed on the target image data to obtain feature data corresponding to the image data. The feature data is input into a preset region proposal network to obtain feature maps in the image data and target proposed region data corresponding to the feature maps. The feature maps and target proposed region data are input into a preset region of interest pooling layer to obtain pooling information, and the transmission line to be inspected is detected based on the pooling information. By performing image quality enhancement processing on the image data, the feature data in the image data is enhanced, making the feature data easier to extract, thereby reducing the omission of feature data during image recognition, ensuring the effectiveness of image recognition, and making the detection process of the transmission line to be inspected more reliable and accurate. By fully leveraging the advantages of image augmentation and enhancement technology and image recognition, features in transmission line defect images are effectively extracted, achieving fast and accurate transmission line defect image detection. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] in:

[0045] Figure 1 This is a flowchart of a transmission line detection method based on image recognition provided in an embodiment of this application;

[0046] Figure 2 This is a framework diagram of a transmission line detection device based on image recognition provided in an embodiment of this application;

[0047] Figure 3 This is a schematic diagram of the structure of a transmission line detection device based on image recognition provided in an embodiment of this application;

[0048] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0049] Figure 5 This is a schematic diagram of the structure of a computer-readable storage medium provided in an embodiment of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] The image recognition-based transmission line detection method provided in this invention can be applied in application environments including user terminals and / or server terminals, wherein the user terminal communicates with the server terminal via a network. The server terminal can acquire image data of the transmission line to be detected and perform image quality enhancement processing on the image data to obtain target image data; extract features from the target image data to obtain feature data corresponding to the image data; input the feature data into a preset region proposal network to obtain a feature map in the image data and target proposed region data corresponding to the feature map; input the feature map and the target proposed region data into a preset region of interest pooling layer to obtain pooling information, and detect the transmission line to be detected based on the pooling information. In this invention, by performing image quality enhancement processing on the image data, the feature data in the image data is enhanced, making the feature data in the image data easier to extract, thereby reducing the omission of feature data during image recognition, ensuring the effectiveness of image recognition, and making the detection process of the transmission line to be detected more reliable and accurate. By fully leveraging the advantages of image augmentation and enhancement technologies and image recognition, this invention effectively extracts features from images of power transmission line defects, achieving rapid and accurate defect detection. The user end can be, but is not limited to, various personal computers, laptops, smartphones, tablets, smart camera devices, and portable wearable devices. The server end can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0052] For example, in practical applications, image data of the transmission line to be detected can be obtained through a client. The client sends the image data of the transmission line to be detected to a server. The server performs image quality enhancement processing on the image data to obtain target image data; extracts features from the target image data to obtain feature data corresponding to the image data; inputs the feature data into a preset region proposal network to obtain a feature map in the image data and target proposed region data corresponding to the feature map; inputs the feature map and the target proposed region data into a preset region of interest pooling layer to obtain pooling information, and detects the transmission line to be detected based on the pooling information.

[0053] For example, when inspecting the transmission line to be inspected based on the pooling information, an image recognition model can be used to identify the defect type corresponding to the defect feature map in the pooling information.

[0054] like Figure 1 As shown in the figure, this application provides a transmission line detection method based on image recognition, including:

[0055] S101. Acquire image data of the transmission line to be detected, and perform image quality enhancement processing on the image data to obtain target image data;

[0056] For example, image data of the transmission line to be inspected can be acquired directly through an image acquisition device, or it can be acquired through data transmission.

[0057] For example, image quality enhancement is a key technology in digital image processing. In image applications, image quality directly affects the effectiveness of image applications. For instance, in the process of image recognition, image quality directly affects the result of image recognition.

[0058] For example, in image recognition scenarios, image augmentation techniques are a method to expand the dataset by enhancing the diversity of training data. By performing random transformations, rotations, flips, and other operations on images, the diversity of training data can be increased, improving the model's generalization ability and reducing overfitting.

[0059] S102. Perform feature extraction on the target image data to obtain feature data corresponding to the image data;

[0060] For example, image feature extraction exists due to machine vision. The process by which a computer extracts relevant pixels that constitute an image for recognition, and then analyzes and assigns these pixels to their respective features, is image feature extraction. From a transformation or mapping perspective, it involves transforming a group of measured values ​​of a certain pattern.

[0061] For example, the effectiveness of feature extraction is often determined by the image quality of the image to be extracted and the performance of the algorithm used for feature extraction. Therefore, improving the image quality and / or improving the performance of the algorithm used for feature extraction can improve the effectiveness of feature extraction.

[0062] S103. Input the feature data into a preset region proposal network to obtain the feature map in the image data and the target proposal region data corresponding to the feature map.

[0063] For example, a proposed regional network is used to calibrate a specific region, that is, to generate a calibration framework for that specific region.

[0064] For example, the feature data is input into a preset region proposal network to generate a proposed region corresponding to the feature map, and it is determined whether the generated proposed region contains wire defects. Then, the proposed region containing wire defects and the feature map of wire defects are sent to the region of interest pooling layer to extract information for subsequent determination of defect category and correction of proposed region boundary.

[0065] S104. Input the feature map and the target proposed region data into a preset region of interest pooling layer to obtain pooling information, and detect the transmission line to be detected based on the pooling information.

[0066] For example, when inspecting the transmission line to be inspected based on the pooling information, an image recognition model can be used to identify the defect type corresponding to the defect feature map in the pooling information. The defect type is, for example, broken strands, foreign objects, etc.

[0067] By acquiring image data of the transmission line to be inspected and performing image quality enhancement processing on the image data, target image data is obtained. Feature extraction is performed on the target image data to obtain feature data corresponding to the image data. The feature data is input into a preset region proposal network to obtain feature maps in the image data and target proposed region data corresponding to the feature maps. The feature maps and target proposed region data are input into a preset region of interest pooling layer to obtain pooling information, and the transmission line to be inspected is detected based on the pooling information. By performing image quality enhancement processing on the image data, the feature data in the image data is enhanced, making the feature data easier to extract, thereby reducing the omission of feature data during image recognition, ensuring the effectiveness of image recognition, and making the detection process of the transmission line to be inspected more reliable and accurate. By fully leveraging the advantages of image augmentation and enhancement technology and image recognition, features in transmission line defect images are effectively extracted, achieving fast and accurate transmission line defect image detection.

[0068] In one possible implementation, the step of performing image quality enhancement processing on the image data to obtain target image data includes:

[0069] The image data is input into a densely connected convolutional block in a preset image quality enhancement network, and the first feature data of the image data is extracted through a two-dimensional convolutional layer in the densely connected convolutional block.

[0070] Based on the preset feature fusion mechanism of the densely connected convolutional blocks, the difference data between different densely connected convolutional blocks in the preset image quality enhancement network is obtained, and the second feature data of the image data is obtained based on the difference data.

[0071] The first feature data and the second feature data are input into the global feature fusion layer in the preset image quality enhancement network to obtain the enhanced data of the image data.

[0072] The target image data is constructed based on the enhanced data.

[0073] For example, the steps of constructing a preset image quality enhancement network include: building a feature extraction module consisting of two concatenated two-dimensional convolutional layers to extract features from the input low-resolution image; building a residual dense module consisting of densely connected convolutional blocks, a local feature fusion mechanism, a local residual learning mechanism, a continuous memory mechanism, and a global feature fusion layer; specifically, building a densely connected convolutional block consisting of six concatenated two-dimensional convolutional layers and six activation function layers; setting a local feature fusion mechanism for the densely connected convolutional block, that is, by connecting the output state of the previous residual dense block and the output state of all preceding layers in the current residual dense block. The network extracts local dense features by adaptively preserving and fusing information (difference data); the local residual learning mechanism allows the network to learn residual representations instead of directly learning the original feature representations, thereby improving network performance without increasing computational burden; the continuous memory mechanism allows the output of one residual dense block to directly access every layer of the next residual dense block. This mechanism realizes the continuous state transfer, ensuring that each convolutional layer can access all subsequent layers within the residual dense block and pass on the information that needs to be preserved; a global feature fusion layer is constructed, consisting of one connection layer, one convolutional layer, one weighted summation layer, and one batch normalization layer.

[0074] For example, the densely connected convolutional block structure is fixed as follows: two-dimensional convolutional layer → activation function layer → two-dimensional convolutional layer → activation function layer → two-dimensional convolutional layer → activation function layer → two-dimensional convolutional layer → activation function layer → two-dimensional convolutional layer → activation function layer → two-dimensional convolutional layer → activation function layer → two-dimensional convolutional layer → activation function layer.

[0075] For example, the contiguous memory mechanism is implemented by passing the state of the previous residual dense block to each level of the current residual dense block. Let F d-1 and F d Let G and D be the input and output of the (d-th)th residual dense block, respectively, and both the input and the output have G0 feature maps. The output of the c-th convolutional layer of the d-th residual dense block can be expressed as:

[0076] F d,c =σ(W d,c [F d-1 ,F d,1 ,…,F d,c-1 ])

[0077] Where σ represents a nonlinear function (ReLU activation function), W d,c Let F represent the weights of the c-th convolutional layer. d,c-1 This is the input for the dc-th residual dense block, with all biases ignored.

[0078] The global feature fusion layer comprises one connection layer, one convolutional layer, one weighted summation layer, and one batch normalization layer. The structure of the global feature fusion layer is fixed as follows: connection layer → 2D convolutional layer → weighted summation layer → batch normalization layer. The kernel size of the 2D convolutional layer is set to 3×3, and the size of the zero-input padding on each side remains fixed. The activation function of each activation function layer is set to the ReLU activation function. In a densely connected convolutional block, except for the last activation function layer, the output of the previous activation function layer is used as the input of any subsequent 2D convolutional layer in a skip connection manner.

[0079] In one possible implementation, the step of extracting features from the target image data to obtain feature data corresponding to the image data includes:

[0080] The target image data is input into a preset target detection network to obtain the initial feature data of the target image data;

[0081] The initial feature data is input into a preset cascaded bidirectional feature pyramid network to obtain bidirectional feature data of the target image data;

[0082] The bidirectional feature data are fused using a weighted fusion method to obtain fused feature data, which is then used as the feature data corresponding to the image data.

[0083] For example, the preset object detection network adopts the object detection model (EfficientNet). EfficientNet is an efficient convolutional neural network structure that uses composite coefficients to balance the network's depth, width, and resolution, thereby improving model performance while reducing model parameters. EfficientNet has achieved significant results in image classification tasks and can be quickly applied to other tasks with similar characteristics through transfer learning methods, such as the detection of defects in power transmission lines (transmission lines) in this task.

[0084] For example, the steps of constructing a preset object detection network include: concatenating seven convolutional networks (P1, P2, P3, P4, P5, P6, P7) to construct the EfficientNet backbone network.

[0085] For example, a pre-defined cascaded bidirectional feature pyramid network (BiFPN network) performs feature fusion on five convolutional layers: P3, P4, P5, P6, and P7.

[0086] In one possible implementation, the step of inputting the target image data into a preset target detection network to obtain initial feature data of the target image data includes:

[0087] The resolution data of the target image data is scaled according to a preset joint coefficient using a joint scaling method to obtain the input data;

[0088] The network parameters of the preset target detection network are scaled based on the preset joint coefficients to obtain the target network.

[0089] The input data is input into the target network to obtain the initial feature data of the target image data.

[0090] For example, the process of scaling the network parameters of the preset target detection network to obtain the target network is as follows: fix the preset joint coefficients. Assuming more than twice the resources are available, a neural network search is then performed on α, β, and γ. α, β, and γ are fixed as constants, and the baseline network is scaled to obtain the target network (B1, B2, B3, B4, B5, B6, B7).

[0091] For example, joint coefficients are used to determine the network width, depth, and input image (the target image data) resolution. Perform uniform scaling. The scaling formula is as follows:

[0092] depth:

[0093] width:

[0094] Resolution:

[0095] Here, α, β, and γ are constants that can be determined through a small grid search. α is a user-specified coefficient that controls how many resources are available for model scaling, while α, β, and γ specify how these additional resources are allocated to network width, depth, and resolution, respectively.

[0096] For example, in the joint scaling method, when the constraint is α·β 2 ·γ 2When α ≈ 2, EfficientNet achieves its optimal values, i.e., α = 1.2, β = 1.1, γ = 1.15. The target network (B1, B2, B3, B4, B5, B6, B7) has the following output channels: B1 has an input resolution of 224*224 and 32 output channels; B2 has an input resolution of 112*112 and 16 output channels; B3 has an input resolution of 112*112 and 25 output channels; B4 has an input resolution of 56*56 and 40 output channels; B5 has an input resolution of 28*28 and 80 output channels; B6 has an input resolution of 14*14 and 112 output channels; and B7 has an input resolution of 14*14 and 192 output channels.

[0097] In one possible implementation, the step of inputting the initial feature data into a preset cascaded bidirectional feature pyramid network to obtain bidirectional feature data of the target image data includes:

[0098] Delete the node in the preset cascaded bidirectional feature pyramid network that has only one input data to obtain the target cascaded bidirectional feature pyramid network.

[0099] The bidirectional paths in the target cascaded bidirectional feature pyramid network are used as feature network layers, and the initial feature data is input into the feature network layers to obtain the bidirectional feature data of the target image data.

[0100] For example, in layers P3, P4, P5, P6, and P7, nodes with only one input are removed; these nodes receive only one input and have no feature fusion. If the original input and output nodes are at the same level, an additional edge is added between them. Each bidirectional (top-down and bottom-up) path is treated as a feature network layer, and the same layer is repeated multiple times to achieve higher-level feature fusion. A weighted fusion method similar to an attention mechanism is added to layers P3, P4, P5, P6, and P7. This method can be expressed as:

[0101]

[0102] Where ω i Let I represent the learnable weights in the i-th convolutional layer, j represent the number of repetitions of the convolutional layer in the EfficientNet backbone, and ∈ represent the expected value. i For the feature data of the i-th layer convolutional network, ω j represents the weights of the i-th layer convolutional network during the j-th convolution.

[0103] The defect fusion characteristics of the P6 layer power transmission line can be represented as follows:

[0104]

[0105]

[0106] in This is an intermediate feature of the 6th layer of the top-down path. This represents the output features of the 6th layer of the bottom-up path, where Conv() is the vector convolution operation function, ∈ is the expected value, Resize() is the allocation function, ω′1 is the actual weight in the 1st convolutional network, ω′2 is the actual weight in the 2nd convolutional network, and ω′3 is the actual weight in the 3rd convolutional network. (P6) in For the input data of the 6th layer of the top-down path, P7 in This is the input data for the 7th layer of the top-down path. The output features are from the 5th layer of the bottom-up path; all other features are constructed in a similar manner. Feature fusion is performed using depthwise separable convolutions, with batch normalization and activation processing added after each convolution.

[0107] In one possible implementation, the step of inputting the feature data into a preset region proposal network to obtain a feature map of the image data and target proposed region data corresponding to the feature map includes:

[0108] The feature data is input into the first branch network of the proposed network of the preset region, and the feature map of the image data is determined through the preset normalization layer in the first branch network.

[0109] The feature data is input into the second branch network of the preset region proposal network, and the initial proposed region data corresponding to the feature map is corrected through the preset convolutional layer in the second branch network to obtain the target proposed region data.

[0110] For example, a proposed network for a predefined region is constructed by convolutional layers, classification branches (first branch network), and bounding box regression branches (second branch network).

[0111] For example, a classification branch consisting of one 2D convolutional layer and one normalization layer is constructed. The branch structure is: convolutional layer → normalization layer, used to determine whether the generated proposed region contains wire defects; the kernel size of the 2D convolutional layer is set to 1×1, and the normalization layer uses the softmax function. A bounding box regression branch consisting of one 2D convolutional layer is constructed to correct the boundary of the proposed region; the kernel size of the 2D convolutional layer in the bounding box regression branch is set to 1×1.

[0112] In one possible implementation, the step of inputting the feature map and the target proposed region data into a preset region of interest pooling layer to obtain pooling information, and then detecting the transmission line to be detected based on the pooling information, includes:

[0113] The proposed region corresponding to the target proposed region data is mapped and reduced through the first processing layer of the preset interest region pooling layer to obtain the reduced proposed region data.

[0114] The proposed narrowing region data is fitted using the preset regression algorithm of the preset interest region pooling layer to obtain fitted data, and the fitted data is used as the first information in the pooling information.

[0115] The feature map is divided into standard feature maps of a preset specification by the second processing layer of the preset region of interest pooling layer;

[0116] The standard feature map is classified by the preset classification layer of the preset interest region pooling layer to obtain a defect feature map, and the defect feature map is used as the second information in the pooling information.

[0117] The transmission line to be tested is detected based on the defect feature map and the fitted data.

[0118] For example, the target proposed region data is mapped down by a factor of 16 using the spatial-scale parameter to map back to the feature map size, i.e., the spatial-scale is set to 0.0625; the feature map region corresponding to each proposed region is horizontally divided into a grid of size w×h, where w and h are set to 7; max pooling is performed on each part of the grid, and after processing, the output structure of proposed regions of different sizes is fixed to size w×h, thus achieving a fixed-length output.

[0119] For example, a defect classification module consisting of one fully connected layer and one normalization layer is constructed. The structure of the classification module is fixed as: fully connected layer → normalization layer; the normalization layer uses the softmax function. A regression layer is constructed to perform bounding box regression on the proposed region; specifically, a four-dimensional vector (x, y, w, h) is used to represent the center point coordinates and width and height of the bounding box, respectively. The purpose of the bounding box regression is to make the original bounding box A = [A x A y A w A h After input, a bounding box GT = [G] is obtained through mapping. x G y G w G hA closer regression bounding box: Since the original bounding box A is relatively close to the true bounding box GT in this example, a linear regression model is selected for fitting, that is, translation is performed first, and then scaling is performed.

[0120] For example, the translation amount (t) between the bounding box of the bounding box regression module and the true bounding box x , t y ) and scaling factor (t) w , t h This can be represented as:

[0121]

[0122]

[0123] Among them, (x a y a w a h a ) represent the center point coordinates and width and height of the actual bounding box, respectively, and log() is the logarithmic function.

[0124] In one possible implementation, such as Figure 2 As shown, an image recognition-based transmission line detection method includes: data acquisition: acquiring images of transmission lines in multiple experimental pilot scenarios; data preprocessing: organizing and labeling the acquired transmission line images to form a dataset of 1500 transmission line images containing three types: normal, broken strands, and scattered strands, and dividing it into training, validation, and test sets; network construction: constructing a transmission line defect image feature fusion network based on EfficientDet; network training: inputting the transmission line defect training sample set into the constructed network in batches for network training, and verifying the current detection performance of the network using a transmission line defect validation sample set after each batch of training; defect detection: inputting the transmission line defect test sample set into the trained transmission line defect image feature fusion network based on EfficientDet to obtain the detection results.

[0125] For example, during network training, the stochastic gradient descent algorithm is used to update the model parameters. The loss function used by the entire network can be divided into two parts: classification loss: the normalized function loss calculated by the classification branch, used for network training to classify whether the proposed region contains wire defects; and bounding box loss: the loss used for network training for bounding box regression. In this method, a smoothing loss is used, which can be represented as follows:

[0126]

[0127]

[0128] Where ti Indicates the predicted bounding box. This represents the true bounding box. Because in practice, N... cls and N reg The gap is too large. To predict the loss value between the bounding box and the ground truth bounding box, soomth... L1 () represents the smoothing loss function. Parameters:

[0129]

[0130] Balance N cls and N reg , where N cls and N reg These represent the total number of anchor points and the feature map size, respectively; the loss function of the entire network can be expressed as:

[0131]

[0132] Where i represents the index of the anchor bounding box, p i This represents the predicted probability of anchor point i. The actual label of anchor point i; t i Indicates the predicted bounding box. Represents the actual bounding box.

[0133] In one possible implementation, embodiments of this application provide a transmission line detection device based on image recognition, such as... Figure 3 As shown, the device includes:

[0134] The data acquisition module 201 is used to acquire image data of the transmission line to be detected and to perform image quality enhancement processing on the image data to obtain target image data;

[0135] Feature extraction module 202 is used to extract features from the target image data to obtain feature data corresponding to the image data;

[0136] The labeling module 203 is used to input the feature data into a preset region proposal network to obtain a feature map of the image data and target proposal region data corresponding to the feature map;

[0137] The detection module 204 is used to input the feature map and the target proposed region data into a preset region of interest pooling layer to obtain pooling information, and to detect the transmission line to be detected based on the pooling information.

[0138] By acquiring image data of the transmission line to be inspected and performing image quality enhancement processing on the image data, target image data is obtained. Feature extraction is performed on the target image data to obtain feature data corresponding to the image data. The feature data is input into a preset region proposal network to obtain feature maps in the image data and target proposed region data corresponding to the feature maps. The feature maps and target proposed region data are input into a preset region of interest pooling layer to obtain pooling information, and the transmission line to be inspected is detected based on the pooling information. By performing image quality enhancement processing on the image data, the feature data in the image data is enhanced, making the feature data easier to extract, thereby reducing the omission of feature data during image recognition, ensuring the effectiveness of image recognition, and making the detection process of the transmission line to be inspected more reliable and accurate. By fully leveraging the advantages of image augmentation and enhancement technology and image recognition, features in transmission line defect images are effectively extracted, achieving fast and accurate transmission line defect image detection.

[0139] In one possible implementation, such as Figure 4 As shown, this application embodiment provides an electronic device 300, including: a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it performs the following: acquiring image data of a transmission line to be detected, and performing image quality enhancement processing on the image data to obtain target image data; extracting features from the target image data to obtain feature data corresponding to the image data; inputting the feature data into a preset region proposal network to obtain a feature map in the image data and target proposal region data corresponding to the feature map; inputting the feature map and the target proposal region data into a preset region of interest pooling layer to obtain pooling information, and detecting the transmission line to be detected based on the pooling information.

[0140] By acquiring image data of the transmission line to be inspected and performing image quality enhancement processing on the image data, target image data is obtained. Feature extraction is performed on the target image data to obtain feature data corresponding to the image data. The feature data is input into a preset region proposal network to obtain feature maps in the image data and target proposed region data corresponding to the feature maps. The feature maps and target proposed region data are input into a preset region of interest pooling layer to obtain pooling information, and the transmission line to be inspected is detected based on the pooling information. By performing image quality enhancement processing on the image data, the feature data in the image data is enhanced, making the feature data easier to extract, thereby reducing the omission of feature data during image recognition, ensuring the effectiveness of image recognition, and making the detection process of the transmission line to be inspected more reliable and accurate. By fully leveraging the advantages of image augmentation and enhancement technology and image recognition, features in transmission line defect images are effectively extracted, achieving fast and accurate transmission line defect image detection.

[0141] In one possible implementation, such as Figure 5 As shown, this application embodiment provides a computer-readable storage medium 400 storing a computer program 411. When executed by a processor, the computer program 411 performs the following: acquiring image data of a transmission line to be detected, and performing image quality enhancement processing on the image data to obtain target image data; extracting features from the target image data to obtain feature data corresponding to the image data; inputting the feature data into a preset region proposal network to obtain a feature map in the image data and target proposed region data corresponding to the feature map; inputting the feature map and the target proposed region data into a preset region of interest pooling layer to obtain pooling information, and detecting the transmission line to be detected based on the pooling information.

[0142] By acquiring image data of the transmission line to be inspected and performing image quality enhancement processing on the image data, target image data is obtained. Feature extraction is performed on the target image data to obtain feature data corresponding to the image data. The feature data is input into a preset region proposal network to obtain feature maps in the image data and target proposed region data corresponding to the feature maps. The feature maps and target proposed region data are input into a preset region of interest pooling layer to obtain pooling information, and the transmission line to be inspected is detected based on the pooling information. By performing image quality enhancement processing on the image data, the feature data in the image data is enhanced, making the feature data easier to extract, thereby reducing the omission of feature data during image recognition, ensuring the effectiveness of image recognition, and making the detection process of the transmission line to be inspected more reliable and accurate. By fully leveraging the advantages of image augmentation and enhancement technology and image recognition, features in transmission line defect images are effectively extracted, achieving fast and accurate transmission line defect image detection.

[0143] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0144] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0145] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, etc., or any suitable combination thereof.

[0146] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0147] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0148] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

[0149] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A transmission line detection method based on image recognition, characterized in that, include: Image data of the transmission line to be inspected is acquired, and image quality enhancement processing is performed on the image data to obtain target image data; Feature extraction is performed on the target image data to obtain feature data corresponding to the image data; The feature data is input into a preset region proposal network to obtain a feature map of the image data and target proposal region data corresponding to the feature map. The feature map and the target proposed region data are input into a preset region of interest pooling layer to obtain pooling information, and the transmission line to be detected is detected based on the pooling information. The proposed network for the preset region consists of a convolutional layer, a first branch network, and a second branch network. The step of inputting the feature data into a preset region proposal network to obtain a feature map of the image data and target proposal region data corresponding to the feature map includes: The feature data is input into the first branch network of the proposed network of the preset region, and the feature map of the image data is determined through the preset normalization layer in the first branch network. The feature data is input into the second branch network of the preset region proposal network, and the initial proposed region data corresponding to the feature map is corrected through the preset convolutional layer in the second branch network to obtain the target proposed region data. The step of inputting the feature map and the target proposed region data into a preset region of interest pooling layer to obtain pooling information, and then detecting the transmission line to be detected based on the pooling information, includes: The proposed region corresponding to the target proposed region data is mapped and reduced through the first processing layer of the preset interest region pooling layer to obtain the reduced proposed region data. The proposed narrowing region data is fitted using the preset regression algorithm of the preset interest region pooling layer to obtain fitted data, and the fitted data is used as the first information in the pooling information. The feature map is divided into standard feature maps of a preset specification by the second processing layer of the preset region of interest pooling layer; The standard feature map is classified by the preset classification layer of the preset interest region pooling layer to obtain a defect feature map, and the defect feature map is used as the second information in the pooling information. The transmission line to be tested is detected based on the defect feature map and the fitted data.

2. The transmission line detection method based on image recognition as described in claim 1, characterized in that, The step of performing image quality enhancement processing on the image data to obtain target image data includes: The image data is input into a densely connected convolutional block in a preset image quality enhancement network, and the first feature data of the image data is extracted through a two-dimensional convolutional layer in the densely connected convolutional block. Based on the preset feature fusion mechanism of the densely connected convolutional blocks, the difference data between different densely connected convolutional blocks in the preset image quality enhancement network is obtained, and the second feature data of the image data is obtained based on the difference data. The first feature data and the second feature data are input into the global feature fusion layer in the preset image quality enhancement network to obtain the enhanced data of the image data. The target image data is constructed based on the enhanced data.

3. The transmission line detection method based on image recognition as described in claim 1, characterized in that, The step of extracting features from the target image data to obtain feature data corresponding to the image data includes: The target image data is input into a preset target detection network to obtain the initial feature data of the target image data; The initial feature data is input into a preset cascaded bidirectional feature pyramid network to obtain bidirectional feature data of the target image data; The bidirectional feature data are fused using a weighted fusion method to obtain fused feature data, which is then used as the feature data corresponding to the image data.

4. The transmission line detection method based on image recognition as described in claim 3, characterized in that, The step of inputting the target image data into a preset target detection network to obtain the initial feature data of the target image data includes: The resolution data of the target image data is scaled according to a preset joint coefficient using a joint scaling method to obtain the input data; The network parameters of the preset target detection network are scaled based on the preset joint coefficients to obtain the target network. The input data is input into the target network to obtain the initial feature data of the target image data.

5. The transmission line detection method based on image recognition as described in claim 3, characterized in that, The step of inputting the initial feature data into a preset cascaded bidirectional feature pyramid network to obtain bidirectional feature data of the target image data includes: Delete the node in the preset cascaded bidirectional feature pyramid network that has only one input data to obtain the target cascaded bidirectional feature pyramid network. The bidirectional paths in the target cascaded bidirectional feature pyramid network are used as feature network layers, and the initial feature data is input into the feature network layers to obtain the bidirectional feature data of the target image data.

6. A transmission line detection device based on image recognition, characterized in that, The device includes: The data acquisition module is used to acquire image data of the transmission line to be inspected and to perform image quality enhancement processing on the image data to obtain target image data; The feature extraction module is used to extract features from the target image data to obtain feature data corresponding to the image data; A labeling module is used to input the feature data into a preset region proposal network to obtain a feature map of the image data and target proposal region data corresponding to the feature map; The detection module is used to input the feature map and the target proposed region data into a preset region of interest pooling layer to obtain pooling information, and to detect the transmission line to be detected based on the pooling information; The proposed network for the preset region consists of a convolutional layer, a first branch network, and a second branch network. The labeling module is further configured to input the feature data into the first branch network of the preset region proposal network, and determine the feature map of the image data through the preset normalization layer in the first branch network; input the feature data into the second branch network of the preset region proposal network, and correct the initial proposed region data corresponding to the feature map through the preset convolutional layer in the second branch network to obtain the target proposed region data; The detection module is further configured to: map and reduce the proposed region corresponding to the target proposed region data through the first processing layer of the preset region of interest pooling layer to obtain reduced proposed region data; fit the reduced proposed region data through the preset regression algorithm of the preset region of interest pooling layer to obtain fitted data, and use the fitted data as the first information in the pooling information; divide the feature map into standard feature maps of preset specifications through the second processing layer of the preset region of interest pooling layer; classify the standard feature maps through the preset classification layer of the preset region of interest pooling layer to obtain defect feature maps, and use the defect feature maps as the second information in the pooling information; and detect the transmission line to be detected based on the defect feature maps and the fitted data.

7. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the image recognition-based transmission line detection method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the image recognition-based transmission line detection method as described in any one of claims 1 to 5.

Images