Power equipment detection method based on 3D point cloud saliency learning

Abstract

The invention discloses a power equipment detection method based on 3D point cloud saliency learning, and belongs to the technical field of deep learning and three-dimensional target detection. Aiming at the problems of background point interference caused by complicated detection targets and insufficient key feature extraction of a detection model during detection of power transformation equipment by an existing method, a significance learning module is constructed in a feature extraction stage on the basis of a Second algorithm, and a channel attention enhancement module and a residual space attention module are designed, so that the detection accuracy of the power transformation equipment is improved. Focusing the spatial position information of the point cloud to obtain the saliency characteristics of the point cloud in the power scene, and reducing the noise interference in the detection scene, thereby achieving the purpose of improving the model detection precision; salient features are sent to a sparse convolution module for feature extraction, and then a loss function WIoU loss of a 3D bounding box as a whole for regression is constructed in a region recommendation network module, so that the accuracy of model prediction is further enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of deep learning and three-dimensional target detection, and specifically relates to a power equipment detection method based on 3D point cloud saliency learning. Background Art

[0002] Inspection of power equipment in substations is a necessary means to ensure the safe and stable operation of substations. At present, the main methods for inspecting substation equipment are manual inspection, robot inspection, and image recognition technology based on video intelligent monitoring. However, manual inspection is inefficient and the accuracy cannot be guaranteed. The latter two methods are limited by the difficulty of measuring distance with image data and cannot accurately inspect substation equipment.

[0003] With the rapid development of computer vision, 3D target detection technology is becoming more and more mature. In order to cope with the increasingly complex maintenance tasks of power scenes, the recognition of key components through 3D target detection algorithms has become the current mainstream trend. However, the cloud scene of substations is relatively complex and the data scale is also very large. When using existing methods to detect substation equipment, there will be problems such as interference from background points due to the complexity of detection targets, which leads to insufficient detection and extraction of key features by the detection model. Therefore, at this stage, it is urgent to propose a substation equipment detection method that can reduce the influence of background point interference in complex substation scenes and obtain key features in the scene to meet the needs of safe operation and maintenance of substations. Summary of the invention

[0004] In view of the problems that when existing methods are used to detect substation equipment, background point interference may occur due to complex detection targets and the detection model cannot extract key features sufficiently. The present invention provides a power equipment detection method based on 3D point cloud saliency learning.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for detecting electric power equipment based on 3D point cloud saliency learning, the method comprising the following steps:

[0007] Step 1: Create a KITTI format dataset for power scenarios;

[0008] The specific operation of step 1 to establish the KITTI format dataset of the power scene is: using a laser radar to collect point cloud data, using a camera to collect RGB images, using a point cloud annotation tool to annotate the targets in the collected point cloud, and arranging them according to the KITTI dataset label format to obtain a self-built dataset for training and testing; the dataset is divided into a training set, a validation set, and a test set according to the proportion;

[0009] Step 2: Construct a SalSecond target detection model including a saliency learning module, a sparse convolution module, and a region recommendation network module; where:

[0010] The saliency learning module enhances channel attention while ensuring the flow of information between groups, and learns different channel features in groups to obtain more refined global features of the point cloud; then, through point-by-point spatial attention learning, it further obtains salient features in the scene and reduces the impact of background points on detection accuracy;

[0011] The sparse convolution module extracts the obtained significant features;

[0012] The region recommendation network module performs prediction regression. By constructing the WIoU loss function and paying attention to the positional relationship between the prediction box and the true value box, the model regression accuracy is improved. The specific operations are as follows:

[0013] Taking the target detection model Second as the benchmark model, a saliency learning module is constructed in the feature extraction stage, and an enhanced channel attention module and a residual space attention module are designed to focus on the spatial position information of the point cloud, reduce the noise interference in the detection scene, and obtain the saliency features of the point cloud in the power scene. The saliency features are sent to the sparse convolution module for feature extraction, and then enter the regional recommendation network module to construct the 3D bounding box as the overall regression loss function WIoU loss, which solves the problem that the original model loss function ignores the coupling between different parameters of the bounding box, resulting in poor regression accuracy;

[0014] Furthermore, the step 2 uses the traditional target detection model Second as the benchmark model, constructs a saliency learning module in the feature extraction stage, designs an enhanced channel attention module (ECA) and a residual space attention module (RSA), focuses on the spatial position information of the point cloud, reduces noise interference in the detection scene, and obtains the saliency features of the point cloud in the power scene; the specific operations are:

[0015] The voxelized point cloud data is input into the network, processed by the enhanced channel attention module to obtain the global feature G, and then processed by the residual space attention module to obtain the salient feature S of the point cloud; the process is as follows:

[0016] Step 2.1: The enhanced channel attention module divides the input channel into three non-interfering sub-networks, and inputs the voxelized point cloud data aggregated by the maximum pooling and average pooling operations separately, focusing on the fine-grained features of the point cloud from different dimensions; then, the Concat operation is performed on the fine-grained features of different dimensions for splicing; in order to solve the problem of insufficient global feature learning caused by the lack of feature interaction between different groups and the reduction of information expression ability due to the separate calculation of each sub-network, the Shuffle operation is introduced to evenly mix the fine-grained features of different dimensions and then learn them again through the three sub-networks and splice the final results; finally, nonlinear mapping is performed through the activation function to obtain higher-dimensional global features; the process is described by the following formula:

[0017]

[0018] Among them, i represents sub-networks of different groups, ω i is the parameter learned by the fully connected layer in each sub-network, W max +W avg represents the features of the voxelized point cloud after maximum pooling and average pooling, σ is the Sigmoid activation function; f i (x) represents the global feature obtained by concatenating the results of the first grouping;

[0019] The results obtained from each sub-network are then concatenated and a Shuffle operation is performed. The Shuffle operation is an operation that mixes different features. Specifically, the features calculated from different channels in the first half of the network are first reshaped, then reshaped again after transposition, and then evenly mixed. Finally, the learning is performed again through three different channels, and the learned features of different dimensions are concatenated to obtain global features. This can avoid information isolation caused by different groups of calculations and enhance the learning ability of data representation. The formula is as follows:

[0020]

[0021] W l represents the parameters of different fully connected layers during the second grouping operation; σ represents the Sigmoid activation function;

[0022] Step 2.2: The residual spatial attention module adopts a self-attention structure to focus on the spatial distance of the point cloud and the correlation between shape features; introduces a residual connection mechanism to strengthen the relationship between input and output and improve the model's ability to capture significant features; specifically:

[0023] First, the global features obtained by the enhanced channel attention module are mapped into three parameters Q, K, and V through three convolutional layers. Then, the relationship between different points is calculated by transposing Q and K. Finally, softmax is used for normalization to obtain the attention weight A and multiply it with V to obtain the significant feature. The process is described by the following formula:

[0024]

[0025] S=V×A+G (4)

[0026] Among them, S is the significant feature obtained, K T Represents matrix transpose, dk is the dimension of key value K;

[0027] Furthermore, in step 2, the region recommendation network module is entered to construct the 3D bounding box as a loss function WIoU loss for overall regression. The specific operation to solve the problem that the original model loss function ignores the coupling between different parameters of the bounding box and causes the regression accuracy to deteriorate is:

[0028] Step 2.3: Construct the WIoU loss, a loss function for regressing the 3D bounding box as a whole, to solve the problem that the original model loss function ignores the coupling between different parameters of the bounding box, resulting in poor regression accuracy; this process is expressed by the following formula:

[0029] WIoU=L DIoU3D +αv (5)

[0030] Among them, L DIoU3D It is the DIoU in a three-dimensional scene, and its calculation formula is as follows:

[0031] L DIoU3D =1+A DIoU3D (B p ,B g ) (6)

[0032]

[0033] Among them, B g is the true target box, B p To predict the bounding box, V p is the area of ​​the prediction box, V g is the area of ​​the truth box, h i For B p With B g The intersection height, R i For B p With B g The volume of the intersection area, b is B p With B gThe distance between the center points, d is the length of the diagonal of the smallest enclosed rectangle formed by the intersection of the prediction box and the true value box; A DIoU3D (B p ,B g ) represents the predicted box B p With the truth box B g DIoU; IoU 3D (B p ,B g ) represents the IoU (intersection over union) between the predicted box and the candidate box in three-dimensional space;

[0034] In order to make the prediction results more accurate and solve the problem that the DIoU of three-dimensional scenes ignores the consistency of the aspect ratio between the predicted box and the true value box, a penalty term αv is added on the basis of the DIoU of three-dimensional scenes. α represents the balance parameter and v represents the correction factor. The calculation process is as follows:

[0035]

[0036] Among them, L1 and L represent the length of the true value box and the predicted box, W1 and W represent the width of the true value box and the predicted box; IoU represents the result of dividing the overlap between the predicted box and the true value box by the combined part of the two regions.

[0037] Step 3: Training strategy formulation and parameter setting;

[0038] The boundary ranges of the point cloud data on the x, y, and z axes in three-dimensional space are respectively from 0m to 70.4m in width W, from -40m to 40m in height H, and from -1m to 5m in depth D;

[0039] When voxelizing, the voxel size selected is the width v w =0.05m, height v h =0.05m, depth v d =0.1m, and limit each voxel to contain at most 5 points, and the upper limit of the number of voxels is 16000;

[0040] During model training, the parameters were set by the Adam optimizer as follows: the initial learning rate was set to 0.003, and gradually decayed with the increase of training rounds. Each batch processed 4 samples, and a total of 1400 rounds of iterations were performed;

[0041] The momentum parameters β1 and β2 of the Adam optimizer are set to 0.9 and 0.999 respectively, the decay weight of the learning rate decay strategy is 0.01, and the decay factor is 0.8.

[0042] Step 4: Analysis of power equipment component test results.

[0043] Compared with the prior art, the present invention has the following advantages:

[0044] (1) Good accuracy performance. The performance of this model is evaluated by measuring the average AP value of the points that the machine needs to pay attention to in the suspension insulator, transformer, bushing and conductor. The experimental results show that the saliency learning module and WIoU loss help improve the detection accuracy of the model for substation equipment in the substation scenario.

[0045] (2) Strong robustness. The model of the present invention can overcome the challenges brought by small and complex equipment and background interference in the substation scene while having real-time detection, effectively suppress the noise in the detection scene, and effectively improve the problem of insufficient feature extraction in the substation scene.

[0046] (3) Rich detection types. The types that can be detected by the present invention include but are not limited to insulators, transformers, bushings and conductors. Therefore, the present invention is a defect detection method with good generalization. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A step diagram of a power equipment detection method based on 3D point cloud saliency learning;

[0048] Figure 2 This is the overall framework structure diagram;

[0049] Figure 3 To enhance the channel attention module diagram;

[0050] Figure 4 is the residual space attention module diagram;

[0051] Figure 5 Schematic diagram of three-dimensional DIoU. DETAILED DESCRIPTION

[0052] In order to gain a deeper understanding of the present invention, we will provide a comprehensive and detailed description of the present invention. However, the present invention has multiple implementations and is not limited to the specific examples listed herein. The presentation of these examples is intended to deepen the comprehensive understanding of the disclosure of the present invention.

[0053] A method for detecting electric equipment based on 3D point cloud saliency learning, the method comprising the following steps: Figure 1 As shown:

[0054] Step 1: Establish a KITTI format dataset for power scenarios. The original data was collected in a substation scene. The Ouster OS-1-64 line laser radar was used to collect point cloud data. The laser radar had a horizontal resolution of 2048 and a vertical field of view of 45° (±22.5°). A wide-angle 120-degree camera was used to collect RGB images with a camera resolution of 1920*1080. Then, the Labelcloud point cloud annotation tool was used to annotate the objects in the collected point cloud and organize them according to the KITTI dataset label format to obtain a self-built dataset that can be used for training and testing. The self-built dataset was divided into training set, validation set, and test set in a ratio of 7:2:1. The self-built dataset includes the points that the machine needs to pay attention to in the four categories of "insulator", "transformer", "bushing", and "conductor". The distribution of different categories in the dataset is shown in Table 1.

[0055] Table 1 Dataset distribution table

[0056]

[0057] Step 2: Construct the SalSecond target detection model including the saliency learning module, sparse convolution module, and region recommendation network module; its overall model structure is as follows Figure 2 As shown; among which:

[0058] The saliency learning module enhances channel attention while ensuring the flow of information between groups, and learns different channel features in groups to obtain more refined global features of the point cloud; then, through point-by-point spatial attention learning, it further obtains salient features in the scene and reduces the impact of background points on detection accuracy;

[0059] The sparse convolution module extracts the obtained significant features;

[0060] The regional recommendation network performs prediction regression and constructs the WIoU loss function. At the same time, it pays attention to the positional relationship between the prediction box and the true value box to improve the model regression accuracy.

[0061] Taking the traditional target detection model Second as the benchmark model, a saliency learning module is constructed in the feature extraction stage, and an enhanced channel attention module and a residual space attention module are designed to focus on the spatial position information of the point cloud, reduce the noise interference in the detection scene, and obtain the saliency features of the point cloud in the power scene; the saliency features are sent to the sparse convolution module for feature extraction, and then enter the regional recommendation network module to construct the 3D bounding box as the overall regression loss function WIoU loss, which solves the problem that the original model loss function ignores the coupling between different parameters of the bounding box, resulting in poor regression accuracy;

[0062] Step 2.1: The enhanced channel attention module divides the input channel into three non-interfering sub-networks, and inputs the voxelized point cloud data aggregated by the maximum pooling and average pooling operations respectively, focusing on the fine-grained features of the point cloud from different dimensions; then, the Concat operation is performed on the fine-grained features of different dimensions for splicing; the Shuffle operation is introduced to evenly mix the fine-grained features of different dimensions, and then the three sub-networks are learned again and the final results are spliced; finally, nonlinear mapping is performed through the activation function to obtain higher-dimensional global features; the module structure is as follows Figure 3 As shown; the process is described by the following formula:

[0063]

[0064] Among them, i represents sub-networks of different groups, ω i is the parameter learned by the fully connected layer in each sub-network, W max +W avg represents the features of the voxelized point cloud after maximum pooling and average pooling, σ is the Sigmoid activation function; f i (x) represents the global feature obtained by concatenating the results of the first grouping;

[0065] The Shuffle operation is an operation that mixes different features. Specifically, the features calculated by different channels in the first half of the network are first reshaped, reshaped again after transposition, and then evenly mixed. Finally, they are learned again through three different channels, and the learned features of different dimensions are spliced ​​to obtain global features. The formula is as follows:

[0066]

[0067] W l represents the parameters of different fully connected layers during the second grouping operation; σ represents the Sigmoid activation function;

[0068] Step 2.2: The residual spatial attention module adopts a self-attention structure to focus on the spatial distance of the point cloud and the correlation between shape features; introduces a residual connection mechanism to strengthen the relationship between input and output and improve the model's ability to capture significant features; the module structure is as follows Figure 4 Shown: Specifically:

[0069] First, the global features obtained by the enhanced channel attention module are mapped into three parameters Q, K, and V through three convolutional layers. Then, the relationship between different points is calculated by transposing Q and K. Finally, softmax is used for normalization to obtain the attention weight A and multiply it with V to obtain the significant feature. The process is described by the following formula:

[0070]

[0071] S=V×A+G (4)

[0072] Among them, S is the significant feature obtained, K T Represents matrix transpose, dk is the dimension of key value K.

[0073] Step 2.3: Construct the WIoU loss, a loss function for regressing the 3D bounding box as a whole, to solve the problem that the original model loss function ignores the coupling between different parameters of the bounding box, resulting in poor regression accuracy; this process is expressed by the following formula:

[0074] WIoU=L DIoU3D +αv (5)

[0075] Among them, L DIoU3D It is the DIoU in a three-dimensional scene, and its calculation formula is as follows:

[0076] L DIoU3D =1+A DIoU3D (B p ,B g ) (6)

[0077]

[0078] Among them, B g is the true target box, B p is the predicted bounding box, V p is the area of ​​the prediction box, V g is the area of ​​the truth box, h i For B p With B g The intersection height, R i For B p With B g The volume of the intersection area, b is B p With B g The distance between the center points, d is the length of the diagonal of the smallest enclosed rectangle formed by the intersection of the prediction box and the true value box; A DIoU3D (B p ,B g ) represents the predicted box B p With the truth box B g DIoU; IoU 3D (B p ,B g ) represents the IoU (intersection over union) between the predicted box and the candidate box in three-dimensional space;

[0079] The schematic diagram of DIoU in a three-dimensional scene is as follows Figure 5 As shown;

[0080] A penalty term αv is added on the basis of DIoU in three-dimensional scenes, where α represents the balance parameter and v represents the correction factor. This solves the problem that DIoU in three-dimensional scenes ignores the consistency of aspect ratio between the predicted box and the true value box. The calculation process is as follows:

[0081]

[0082]

[0083] Among them, L1 and L represent the length of the true value box and the predicted box, W1 and W represent the width of the true value box and the predicted box; IoU represents the result of dividing the overlapping part of the predicted box and the true value box by the combined part of the two areas.

[0084] Step 3: Formulate training strategies and set parameters; train and test the deep learning model architecture constructed based on this method, and detect different devices in the substation scenario; the specific operations are:

[0085] The boundary ranges of the point cloud data on the x, y, and z axes in three-dimensional space are respectively from 0m to 70.4m in width W, from -40m to 40m in height H, and from -1m to 5m in depth D;

[0086] When voxelizing, the voxel size selected is the width v w =0.05m, height v h =0.05m, depth v d =0.1m, and limit each voxel to contain at most 5 points, and the upper limit of the number of voxels is 16000;

[0087] During model training, the parameters were set by the Adam optimizer as follows: the initial learning rate was set to 0.003, and gradually decayed with the increase of training rounds. Each batch processed 4 samples, and a total of 1400 rounds of iterations were performed;

[0088] The momentum parameters β1 and β2 of the Adam optimizer are set to 0.9 and 0.999 respectively, the decay weight of the learning rate decay strategy is 0.01, and the decay factor is 0.8.

[0089] Step 4: Analysis of test results of power equipment components

[0090] The performance of this model is evaluated on the self-built dataset by measuring the average AP value of the points that the machine needs to pay attention to in the four categories of insulators, transformers, bushings and conductors, and compared with other mainstream target detection models and performed a visual analysis. The experimental results are shown in Table 2. The experimental results show that the average mean accuracy of the model on the self-built dataset is improved by 2.93% compared with the baseline model, and the detection accuracy of the points that the machine needs to pay attention to in the four categories of insulators, transformers, bushings and conductors is improved by 3.43%, 2.03%, 2.89% and 3.36% respectively.

[0091] Table 2 Performance comparison of different detection methods on self-built datasets

[0092]

[0093]

[0094] In order to verify the effectiveness of the saliency learning module and WIoU loss, ablation experiments were performed on the self-built datasets. The results are shown in Table 3. The results show that after constructing the saliency learning module on the basis of the baseline network, the detection accuracy of each category is improved by 2.19%, 1.28%, 2.00% and 2.21% respectively. It can be seen that the introduction of the saliency learning module improves the feature extraction ability of the model. In order to further improve the accuracy of position regression, the original position regression loss function is replaced with WIOU loss. The results show that the detection accuracy of each device is improved by 0.69%, 0.62%, 1.22% and 1.15% respectively, which proves the effectiveness of this method.

[0095] Table 3 Comparison of ablation test results on self-built datasets

[0096]

[0097] The contents not described in detail in the specification of the present invention belong to the prior art known to the professional and technical personnel in the field. Although the illustrative specific embodiments of the present invention are described above to facilitate the understanding of the present invention by the technical personnel in the field, it should be clear that the present invention is not limited to the scope of the specific embodiments. For the ordinary technical personnel in the field, as long as various changes are within the spirit and scope of the present invention defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A power equipment detection method based on 3D point cloud saliency learning, characterized in that: The method comprises the following steps: Step 1: Create a KITTI format dataset for power scenarios; Step 2: SalSecond object detection model including saliency learning module, sparse convolution module, and region recommendation network module; Step 3: Training strategy formulation and parameter setting; Step 4: Analysis of power equipment component test results.

2. The method for detecting electric power equipment based on 3D point cloud saliency learning according to claim 1, characterized in that: The specific operations of step 1 to establish a KITTI format dataset for power scenarios are: using a lidar to collect point cloud data, using a camera to collect RGB images, using a point cloud annotation tool to annotate the targets in the collected point cloud, and organizing them according to the KITTI dataset label format to obtain a self-built dataset for training and testing; the self-built dataset is divided into a training set, a validation set, and a test set according to the proportion.

3. The method for detecting electric power equipment based on 3D point cloud saliency learning according to claim 2, characterized in that: The step 2 constructs a SalSecond target detection model including a saliency learning module, a sparse convolution module, and a region recommendation network module; wherein: The saliency learning module enhances channel attention while ensuring the flow of information between groups, and learns different channel features in groups to obtain more refined global features of the point cloud; then, through point-by-point spatial attention learning, it further obtains salient features in the scene and reduces the impact of background points on detection accuracy; The sparse convolution module extracts the obtained significant features; The region recommendation network module performs prediction regression and improves the model regression accuracy by constructing the WIoU loss function and paying attention to the positional relationship between the prediction box and the true value box.

4. The method for detecting electric power equipment based on 3D point cloud saliency learning according to claim 3 is characterized in that: The step 2 constructs a SalSecond target detection model including a saliency learning module, a sparse convolution module, and a region recommendation network module; the specific operations are: Taking the target detection model Second as the benchmark model, a saliency learning module is constructed in the feature extraction stage, and an enhanced channel attention module and a residual space attention module are designed to focus on the spatial position information of the point cloud, reduce the noise interference in the detection scene, and obtain the saliency features of the point cloud in the power scene; The significant features are sent to the sparse convolution module for feature extraction, and then enter the region recommendation network module to construct the 3D bounding box as the overall regression loss function WIoU loss, which solves the problem that the original model loss function ignores the coupling between different parameters of the bounding box, resulting in poor regression accuracy.

5. The method for detecting electric power equipment based on 3D point cloud saliency learning according to claim 4, characterized in that: The target detection model Second is used as the benchmark model, a saliency learning module is constructed in the feature extraction stage, an enhanced channel attention module and a residual space attention module are designed, the spatial position information of the point cloud is focused, the noise interference in the detection scene is reduced, and the saliency features of the point cloud in the power scene are obtained; the specific operations are: The voxelized point cloud data is input into the network and processed by the enhanced channel attention module to obtain the global feature G, and then processed by the residual space attention module to obtain the salient feature S of the point cloud; The process is as follows: Step 2.1: The enhanced channel attention module divides the input channel into three non-interfering sub-networks, and inputs the voxelized point cloud data aggregated by the maximum pooling and average pooling operations respectively, focusing on the fine-grained features of the point cloud from different dimensions; then, the Concat operation is performed on the fine-grained features of different dimensions for splicing; the Shuffle operation is introduced to evenly mix the fine-grained features of different dimensions, and then the three sub-networks are learned again and the final results are spliced; finally, nonlinear mapping is performed through the activation function to obtain higher-dimensional global features; The process is described by the following formula: Among them, i represents sub-networks of different groups, ω i is the parameter learned by the fully connected layer in each sub-network, W max +W avg represents the features of the voxelized point cloud after maximum pooling and average pooling, σ is the Sigmoid activation function; f i (x) represents the global feature obtained by concatenating the results of the first grouping; Then the results obtained from each sub-network are concatenated and the Shuffle operation is performed; specifically, the features calculated from different channels in the first half of the network are first reshaped, then reshaped again after transposition, and then evenly mixed, and finally learned again through 3 different channels, and the learned features of different dimensions are concatenated to obtain the global feature G; the formula is as follows: W l represents the parameters of different fully connected layers during the second grouping operation; σ represents the Sigmoid activation function; Step 2.2: The residual spatial attention module adopts a self-attention structure to focus on the spatial distance of the point cloud and the correlation between shape features; introduces a residual connection mechanism to strengthen the relationship between input and output and improve the model's ability to capture significant features; specifically: First, the global features obtained by the enhanced channel attention module are mapped into three parameters Q, K, and V through three convolutional layers, and then the relationship between different points is calculated using the transpose of Q and K. Finally, softmax is used for normalization to obtain the attention weight A and multiply it by V to obtain the significant feature S. The process is described by the following formula: S=V×A+G (4) Among them, S is the significant feature obtained, K T Represents matrix transpose, dk is the dimension of key value K.

6. The method for detecting electric power equipment based on 3D point cloud saliency learning according to claim 5, characterized in that: The significant features are sent to the sparse convolution module for feature extraction, and then enter the region recommendation network module to construct the 3D bounding box as the overall regression loss function WIoU loss. The specific operation to solve the problem that the original model loss function ignores the coupling between different parameters of the bounding box and causes the regression accuracy to deteriorate is: Step 2.3: Construct the WIoU loss, a loss function for regressing the 3D bounding box as a whole, to solve the problem that the original model loss function ignores the coupling between different parameters of the bounding box, resulting in poor regression accuracy; this process is expressed by the following formula: W IoU =L DIoU3D +αv (5) Among them, L DIoU3D It is the DIoU in a three-dimensional scene, and its calculation formula is as follows: L DIoU3D =1+A DIoU3D (B p ,B h ) (6) Among them, B g is the true target box, B p is the predicted bounding box, V p is the area of ​​the prediction box, V g is the area of ​​the truth box, h i For B p With B g The intersection height, R i For B p With B g The volume of the intersection area, b is B p With B g The distance between the center points, d is the length of the diagonal of the smallest enclosed rectangle formed by the intersection of the prediction box and the true value box; A DIoU3D (B p ,B g ) represents the predicted box B p With the truth box B g DIoU; IoU 3D (B p ,B g ) represents the IoU between the predicted box and the candidate box in three-dimensional space; A penalty term αv is added on the basis of DIoU in three-dimensional scenes, where α represents the balance parameter and v represents the correction factor. This solves the problem that DIoU in three-dimensional scenes ignores the consistency of aspect ratio between the predicted box and the true value box. The calculation process is as follows: Among them, L1 and L represent the length of the true value box and the predicted box, W1 and W represent the width of the true value box and the predicted box; IoU represents the result of dividing the overlapping part of the predicted box and the true value box by the combined part of the two areas.

7. The method for detecting electric power equipment based on 3D point cloud saliency learning according to claim 6, characterized in that: The specific operations of step 3 training strategy formulation and parameter setting are as follows: The boundary ranges of the point cloud data on the x, y, and z axes in three-dimensional space are respectively from 0m to 70.4m in width W, from -40m to 40m in height H, and from -1m to 5m in depth D; When voxelizing, the voxel size selected is the width v w =0.05m, height v h =0.05m, depth v d =0.1m, and limit each voxel to contain at most 5 points, and the upper limit of the number of voxels is 16000; During model training, the parameters were set by the Adam optimizer as follows: the initial learning rate was set to 0.003, and gradually decayed with the increase of training rounds. Each batch processed 4 samples, and a total of 1400 rounds of iterations were performed; The momentum parameters β1 and β2 of the Adam optimizer are set to 0.9 and 0.999 respectively, the decay weight of the learning rate decay strategy is 0.01, and the decay factor is 0.8.

Images