CN116486166A - Power transmission line foreign matter identification detection method based on edge calculation - Google Patents
Power transmission line foreign matter identification detection method based on edge calculation Download PDFInfo
- Publication number
- CN116486166A CN116486166A CN202310470792.8A CN202310470792A CN116486166A CN 116486166 A CN116486166 A CN 116486166A CN 202310470792 A CN202310470792 A CN 202310470792A CN 116486166 A CN116486166 A CN 116486166A
- Authority
- CN
- China
- Prior art keywords
- feature
- network
- loss function
- transmission line
- edge calculation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 53
- 238000004364 calculation method Methods 0.000 title claims abstract description 32
- 230000005540 biological transmission Effects 0.000 title claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 16
- 238000013528 artificial neural network Methods 0.000 claims abstract description 15
- 230000006870 function Effects 0.000 claims description 31
- 238000000034 method Methods 0.000 claims description 26
- 230000001629 suppression Effects 0.000 claims description 9
- 238000013135 deep learning Methods 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 4
- 238000009432 framing Methods 0.000 claims description 3
- 239000007788 liquid Substances 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 9
- 238000013136 deep learning model Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009194 climbing Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/096—Transfer learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/766—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using regression, e.g. by projecting features on hyperplanes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/94—Hardware or software architectures specially adapted for image or video understanding
- G06V10/95—Hardware or software architectures specially adapted for image or video understanding structured as a network, e.g. client-server architectures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of a transmission line foreign matter identification detection method based on edge calculation, and discloses a transmission line foreign matter identification detection method based on edge calculation, which comprises the following steps: s1, constructing an SSD network model, wherein the SSD network model comprises a convolution neural network and a multi-scale feature detection network, the convolution neural network is used for extracting image features, edge calculation based on NPU neural network processor hardware is adopted, namely data processing is carried out on the edge of equipment, the equipment is used for processing, partial calculation results can be sent to a cloud for deep learning model training and updating, and therefore data transmission efficiency and data safety are improved.
Description
Technical Field
The invention relates to the technical field of a transmission line foreign matter identification and detection method based on edge calculation, in particular to a transmission line foreign matter identification and detection method based on edge calculation.
Background
At present, the method for detecting the foreign matters on the power transmission equipment mainly uses an unmanned aerial vehicle to shoot a picture of the power transmission equipment, then the picture is transmitted back to a server for image recognition and detection, and the method is an efficient and reliable method, so that workers can be prevented from climbing up the high-position inspection equipment, the foreign matters on the power transmission equipment can be quickly found, the finding rate is improved, in addition, for a large power transmission line, the time and the cost can be saved, and the working efficiency is improved by shooting the picture by using the unmanned aerial vehicle, therefore, the unmanned aerial vehicle technology is adopted to detect the power transmission equipment in a mode of selecting more enterprises and institutions, the method can detect the foreign matters on the power transmission equipment, but has the defects that a great deal of manpower and time are required for manual investigation, and real-time detection cannot be realized, so that when a problem occurs in a power transmission system, the problem of the foreign matters is difficult to be found and handled in time. While the detection of the image recognition method for re-using the unmanned aerial vehicle to take the photo can save a lot of manpower, a lot of storage and calculation capacities of a server are required, and the safety of data transmission cannot be ensured, so that the methods cannot well meet the actual requirements to solve the problems;
edge computing is an emerging computing model, the development of which stems from the cloud computing model that is popular today. In the cloud computing mode, data needs to be transmitted to the cloud for processing, and the method has many advantages such as convenience, cost saving and the like, but also has many problems, wherein the most important is that the problems of bandwidth and delay are more and more serious due to the fact that the data volume is too large, and the safety of processing massive data is difficult to guarantee. Unlike cloud computing, edge computing stores data locally, fundamentally solving security and privacy issues. In addition, the edge calculation is closer to the data source, only the processed data is transmitted to the cloud center, the time delay caused by the data transmission speed and the bandwidth limitation is greatly reduced, and the processing data pressure of the cloud center is obviously relieved.
Disclosure of Invention
The invention aims to provide a transmission line foreign matter identification detection method based on edge calculation, which solves the problems in the background technology.
In order to achieve the above purpose, the present invention provides the following technical solutions: a transmission line foreign matter identification detection method based on edge calculation comprises the following steps:
s1, constructing an SSD network model, wherein the SSD network model comprises a convolution neural network and a multi-scale feature detection network, the convolution neural network is used for extracting image features, and the multi-scale feature detection network reduces the size by carrying out pooling operation on feature images extracted by the convolution neural network;
s2, using a plurality of feature maps of different convolution layers to classify objects and returning the offset of a target frame;
s3, obtaining an output detection result by using a Non Maximum Suppression method;
s4, calculating a loss function through an SSD algorithm, and adding the confidence level of classification and the loss value of target frame regression during calculation;
s5, performing multi-task deep learning.
Preferably, in the step S1, the convolutional neural network performs multiple convolution and pooling operations on the input image to gradually reduce the size of the feature map and extract feature information of different layers, and the feature map is sent to a subsequent multi-scale feature detection network for target detection and framing.
Preferably, in step S1, the multiscale feature detection network sets up an additional convolution layer on the existing SSD network to obtain a feature map with a larger degree of distinction, so that it is more beneficial to detect objects with different scales, the receptive field of the low-level feature map is smaller, and the receptive field of the high-level feature map is larger.
Preferably, in step S1, the multi-scale feature detection network extracts feature graphs obtained by different convolution layers as input, sets the scale of a priori frame for each feature graph, and performs convolution operation and classification and regression operation on the feature graphs to obtain classification information and position information, so that the multi-scale feature detection network is suitable for running on embedded equipment, and can effectively improve detection speed and accuracy through the arrangement of the multi-scale feature detection network and the priori frame, and meanwhile avoid the problems of repeated detection and false detection.
Preferably, in the step S2, the default frames with different shapes such as wide, flat, thin and high frames are used to better adapt to the targets with different shapes such as rectangle, circle and ellipse, when the default frames are selected, the number and the size of the default frames are determined according to the size of each feature map, when the feature map is smaller, the smaller default frames are used, when the feature map is larger, the larger default frames are used, in addition, the aspect ratio of the object is considered when the default frames are selected, so as to ensure that the matching degree of the default frames and the targets is as high as possible, and the default frames with different shapes are used, thereby better adapting to the targets with different shapes.
Preferably, in the step S3, the Non Maximum Suppression operation flow is as follows:
A1. dividing all frames according to categories, and eliminating background categories;
A2. for the bounding boxes (b_box) in each object class, arranging in descending order according to the classification confidence;
A3. in a certain class, selecting a boundary BOX B_BOX1 with highest confidence, removing B_BOX1 from an input list, and adding the B_BOX1 into an output list;
A4. calculating the intersection ratio IoU of the B_BOX1 and the rest B_BOX2 one by one, and removing the B_BOX2 at the input if IoU (B_BOX1, B_BOX2) > threshold TH;
A5. repeating the steps 3-4 until the input list is empty, and completing the traversal of one object class;
A6. repeating the steps of 2 to 5 until Non Maximum Suppression treatment of all the object classes is completed;
A7. and outputting the list, and ending the algorithm.
Preferably, in the step S4, the SSD algorithm calculates a total loss function of the loss functions as:
wherein: n is the number of matched default frames; alpha is a balance factor; x is the confidence of whether the matched box belongs to a certain category; c is a category confidence predictive value; l is a prediction frame; g is a real frame; lconf is confidence loss, and a cross entropy loss function is adopted; lloc is the Smooth L1 loss function of the prediction frame and the real frame, wherein the cross entropy loss function is:
in which the number of the components in the liquid,prediction frame and real frameThe Smooth L1 loss function of (2) is:
the smoth L1 loss function is:
the VGG network in SSD is replaced by a lightweight convolutional neural network MobileNet, and network compression and acceleration are achieved through the use of depth separable convolution.
Preferably, in the step S5, the input data is sent to the model to be predicted and compared with the real label during the deep learning, so as to calculate the loss function, and in order to ensure that the multiple tasks are balanced during the training during the deep learning, a dynamic weight average method is adopted to dynamically adjust the loss weight of each task in each training round, so that the importance degree of different tasks is equal, and a dynamic weight average method is adopted to adaptively adjust the task weight, so that the loss convergence speed of each subtask is the same.
The invention provides a transmission line foreign matter identification detection method based on edge calculation. The transmission line foreign matter identification detection method based on edge calculation has the following beneficial effects:
(1) According to the transmission line foreign matter identification detection method based on the edge calculation, the edge calculation based on NPU neural network processor hardware is adopted, namely data processing is carried out on the edge of equipment, the equipment is used for processing, partial calculation results can be sent to a cloud for deep learning model training and updating, so that the transmission efficiency and the data safety of data are improved, meanwhile, the method also adopts a multi-task deep learning technology, namely data of different tasks are combined into the same deep learning model for training and identification, and the complexity and the training time of data processing are reduced;
(2) According to the transmission line foreign matter identification detection method based on edge calculation, the MobileNet and the optimized SSD target detection method are adopted to directly perform processing calculation on edge equipment, then detected foreign matter images are uploaded to a cloud center server or a local area, compared with the SSD method based on a VGG network, the optimized MobileNet and the SSD target detection method are about 5 times Faster in running speed on a CPU, and compared with the Faster-RCNN, the optimized MobileNet and the SSD target detection method are about 58 times Faster;
(3) According to the transmission line foreign matter identification detection method based on edge calculation, VGG in an SSD original is replaced by MobileNet, mobileNet, the calculation complexity of standard convolution operation is reduced from O (k 2 c) to O (k 2 cd), k is the size of a convolution kernel, c is the number of channels, d is the size of an output feature diagram, the design enables MobileNet to have fewer parameters and faster reasoning speed than a traditional convolution neural network, meanwhile, higher accuracy can still be maintained, VGG network of extracted features in an SSD model is replaced by MobileNet so that the VGG network is conveniently deployed on embedded equipment, the detection speed is improved, and MobileNet adopts a depth separable convolution design, so that model parameters and calculation complexity are greatly reduced, and meanwhile, higher accuracy is maintained.
Drawings
FIG. 1 is a schematic diagram of an SSD network according to the present invention;
fig. 2 is a schematic diagram of the first half of the MobileNet network structure according to the present invention;
fig. 3 is a schematic diagram of the second half of the MobileNet network structure according to the present invention;
fig. 4 is a deep learning flow chart of the present invention.
Detailed Description
The technical scheme of the present invention is further described in non-limiting detail below with reference to the preferred embodiments and the accompanying drawings. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, based on the embodiments of the invention, which are apparent to those of ordinary skill in the art without undue burden are within the scope of the invention
Referring to fig. 1-4, the present invention provides the following technical solutions: a transmission line foreign matter identification detection method based on edge calculation comprises the following steps:
s1, constructing an SSD network model, wherein the SSD network model comprises a convolution neural network and a multi-scale feature detection network, the convolution neural network is used for extracting image features, the convolution neural network carries out repeated convolution and pooling operations on an input image so as to gradually reduce the size of feature images, and extracts feature information of different layers, the feature images are sent into a subsequent multi-scale feature detection network for target detection and framing, the SSD network model has the characteristic of parameter sharing, the calculation amount can be obviously reduced, high-quality feature information can be extracted, the multi-scale feature detection network reduces the size by carrying out pooling operations on the feature images extracted by the convolution neural network, the multi-scale feature detection network comprises a plurality of additional convolution layers arranged on the prior SSD network, the feature images with larger discrimination degree are obtained, so that objects under different scales can be detected more easily, the receptive field of the low-level feature images is smaller, the receptive field of the high-level feature images is larger, the multi-scale feature detection network takes the feature images obtained by extracting different convolution layers as input, the scale of the prior frame is set for each feature image, the scale of the prior frame is in direct proportion to the scaling degree of the feature images, the convolution operation is carried out on the feature images, and classification and regression operation are carried out respectively, so that classification information and position information are obtained, the method is suitable for being operated on embedded equipment, the detection speed and the accuracy can be effectively improved through the arrangement of the multi-scale feature detection network and the prior frame, and meanwhile, the problems of repeated detection and false detection are avoided;
s2, classifying objects by using a plurality of feature maps of different convolution layers, returning the offset of target frames, and better adapting to targets of different shapes such as rectangle, circle, ellipse and the like by using default frames of different shapes such as wide flat and thin high frames, when selecting the default frames, firstly determining the number and the size of the default frames according to the size of each feature map, when the feature map is smaller, using smaller default frames, and when the feature map is larger, using larger default frames, and in addition, considering aspect ratio of the objects when selecting the default frames, so as to ensure that the matching degree of the default frames and the targets is as high as possible, and using default frames of different shapes, thereby better adapting to targets of different shapes, improving the accuracy and the robustness of an algorithm, and simultaneously avoiding the problem of detection errors caused by using single-shape default frames;
s3, obtaining an output detection result by using a Non Maximum Suppression method, wherein the Non Maximum Suppression operation flow is as follows:
A1. dividing all frames according to categories, and eliminating background categories;
A2. for the bounding boxes (b_box) in each object class, arranging in descending order according to the classification confidence;
A3. in a certain class, selecting a boundary BOX B_BOX1 with highest confidence, removing B_BOX1 from an input list, and adding the B_BOX1 into an output list;
A4. calculating the intersection ratio IoU of the B_BOX1 and the rest B_BOX2 one by one, and removing the B_BOX2 at the input if IoU (B_BOX1, B_BOX2) > threshold TH;
A5. repeating the steps 3-4 until the input list is empty, and completing the traversal of one object class;
A6. repeating the steps of 2 to 5 until Non Maximum Suppression treatment of all the object classes is completed;
A7. outputting a list, and ending the algorithm;
s4, calculating a loss function through an SSD algorithm, adding the classified confidence coefficient and the target frame regression loss value during calculation, wherein the SSD algorithm calculates the total loss function of the loss function as follows:
wherein: n is the number of matched default frames; alpha is a balance factor; x is the confidence of whether the matched box belongs to a certain category; c is a category confidence predictive value; l is a prediction frame; g is a real frame; lconf is confidence loss, and a cross entropy loss function is adopted; lloc is the Smooth L1 loss function of the prediction frame and the real frame, wherein the cross entropy loss function is:
in which the number of the components in the liquid,the Smooth L1 loss function of the prediction frame and the real frame is as follows:
the smoth L1 loss function is:
replacing VGG network in SSD source with a lightweight convolutional neural network MobileNet, and realizing network compression and acceleration by using depth separable convolution;
s5, performing multi-task deep learning, wherein input data are sent into a model to be predicted when the deep learning is performed, and compared with a real label, a loss function can be calculated, and the loss function is divided into three parts, namely three parts: the method comprises the steps of (1) for obj loss, lbox and lcls, for obj loss, learning according to the overlapping rate (Intersection over Union, ioU) between the target and the anchor frame corresponding to the region if the target exists, wherein the larger IoU indicates that the target is more consistent with the region, for lbox, the smaller the corresponding loss is, the more accurate the coordinates of the target frame are indicated, finally, the smaller the value is, the more accurate the judgment of the target class is, and the attention is paid to the fact that lcls and lbox only accumulate loss values in the region where the target exists, the loss values of the non-target region are zero, in order to ensure that a plurality of tasks are balanced in training, a dynamic weight average method is adopted for dynamically adjusting the loss weight of each task in each round of training, so that the importance degree of different tasks is equal, a dynamic weight average method can be adopted for adaptively adjusting the weight of the target frame, and finally, the value of the target class loss is equal to the loss value of each task in each round of training is equal, and the specific loss of the task is equal to the first round of 62 is set as the corresponding training function before the training is reduced to the value of the first round of the training function (62):
if (less, indicating that the loss is reduced by the t-1 th training round, the task gets better learning, so the attention to the task can be properly reduced, and the weight of the kth task is calculated as:
wherein T controls the flatness of the weight distribution, and a larger T causes the weights of all tasks to approach 1
The weights of the two rounds before training are set to 1.
Claims (8)
1. The transmission line foreign matter identification detection method based on edge calculation is characterized by comprising the following steps of:
s1, constructing an SSD network model, wherein the SSD network model comprises a convolution neural network and a multi-scale feature detection network, the convolution neural network is used for extracting image features, and the multi-scale feature detection network reduces the size by carrying out pooling operation on feature images extracted by the convolution neural network;
s2, using a plurality of feature maps of different convolution layers to classify objects and returning the offset of a target frame;
s3, obtaining an output detection result by using a Non Maximum Suppression method;
s4, calculating a loss function through an SSD algorithm, and adding the confidence level of classification and the loss value of target frame regression during calculation;
s5, performing multi-task deep learning.
2. The method according to claim 1, wherein in step S1, the convolutional neural network performs a plurality of convolution and pooling operations on the input image to gradually reduce the size of feature maps, and extract feature information of different levels, and the feature maps are sent to a subsequent multi-scale feature detection network for target detection and framing.
3. The method for identifying and detecting foreign matters in power transmission lines based on edge calculation according to claim 1, wherein in the step S1, the multiscale feature detection network obtains a feature map with a larger degree of distinction by erecting an additional convolution layer on an existing SSD network, the receptive field of a lower-layer feature map is smaller, and the receptive field of a higher-layer feature map is larger.
4. The method for identifying and detecting foreign bodies in a power transmission line based on edge calculation according to claim 1, wherein in the step S1, a multi-scale feature detection network takes feature graphs obtained by separately extracting different convolution layers as input, sets the scale of a prior frame for each feature graph, the scale of the prior frame is proportional to the scaling degree of the feature graphs, and performs convolution operation and classification and regression operation on the feature graphs to obtain classification information and position information.
5. The method according to claim 1, wherein in the step S2, by using default frames with different shapes such as wide flat and thin high frames, objects with different shapes such as rectangle, circle, ellipse are better adapted, when selecting default frames, the number and size of default frames are first determined according to the size of each feature map, when feature maps are smaller, smaller default frames are used, and when feature maps are larger, larger default frames are used, and in addition, aspect ratio examples of objects are also considered when selecting default frames, so as to ensure that the matching degree between default frames and objects is as high as possible.
6. The method for identifying and detecting the foreign matter on the power transmission line based on the edge calculation of claim 1, wherein in the step S3, the Non Maximum Suppression operation flow is as follows:
A1. dividing all frames according to categories, and eliminating background categories;
A2. for the bounding boxes (b_box) in each object class, arranging in descending order according to the classification confidence;
A3. in a certain class, selecting a boundary BOX B_BOX1 with highest confidence, removing B_BOX1 from an input list, and adding the B_BOX1 into an output list;
A4. calculating the intersection ratio IoU of the B_BOX1 and the rest B_BOX2 one by one, and removing the B_BOX2 at the input if IoU (B_BOX1, B_BOX2) > threshold TH;
A5. repeating the steps 3-4 until the input list is empty, and completing the traversal of one object class;
A6. repeating the steps of 2 to 5 until Non Maximum Suppression treatment of all the object classes is completed;
A7. and outputting the list, and ending the algorithm.
7. The method for identifying and detecting foreign objects on a power transmission line based on edge calculation according to claim 1, wherein in the step S4, the SSD algorithm calculates a total loss function of the loss functions as follows:
wherein: n is the number of matched default frames; alpha is a balance factor; x is the confidence of whether the matched box belongs to a certain category; c is a category confidence predictive value; l is a prediction frame; g is a real frame; lconf is confidence loss, and a cross entropy loss function is adopted; lloc is the Smooth L1 loss function of the prediction frame and the real frame, wherein the cross entropy loss function is:
in which the number of the components in the liquid,
the Smooth L1 loss function of the prediction frame and the real frame is as follows:
the smoth L1 loss function is:
8. the method according to claim 1, wherein in step S5, the input data is sent to the model for prediction and compared with the real labels to calculate a loss function, and for a single task, we need to minimize the loss function to update the model parameters for better performance, and in order to ensure that multiple tasks are balanced in training, a dynamic weight average method is used to dynamically adjust the loss weight of each task in each training round to make the importance degree of different tasks equal.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310470792.8A CN116486166A (en) | 2023-04-27 | 2023-04-27 | Power transmission line foreign matter identification detection method based on edge calculation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310470792.8A CN116486166A (en) | 2023-04-27 | 2023-04-27 | Power transmission line foreign matter identification detection method based on edge calculation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116486166A true CN116486166A (en) | 2023-07-25 |
Family
ID=87219180
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310470792.8A Pending CN116486166A (en) | 2023-04-27 | 2023-04-27 | Power transmission line foreign matter identification detection method based on edge calculation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116486166A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117520950A (en) * | 2024-01-04 | 2024-02-06 | 贵州大学 | Multi-target UAV fault diagnosis method based on attention knowledge sharing network |
-
2023
- 2023-04-27 CN CN202310470792.8A patent/CN116486166A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117520950A (en) * | 2024-01-04 | 2024-02-06 | 贵州大学 | Multi-target UAV fault diagnosis method based on attention knowledge sharing network |
CN117520950B (en) * | 2024-01-04 | 2024-03-19 | 贵州大学 | Multi-target UAV fault diagnosis method based on attention knowledge sharing network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111126472B (en) | SSD (solid State disk) -based improved target detection method | |
US11798132B2 (en) | Image inpainting method and apparatus, computer device, and storage medium | |
WO2021238019A1 (en) | Real-time traffic flow detection system and method based on ghost convolutional feature fusion neural network | |
CN108416307B (en) | Method, device and equipment for detecting pavement cracks of aerial images | |
CN112150821B (en) | Lightweight vehicle detection model construction method, system and device | |
CN107169954B (en) | Image significance detection method based on parallel convolutional neural network | |
WO2023193401A1 (en) | Point cloud detection model training method and apparatus, electronic device, and storage medium | |
CN111160249A (en) | Multi-class target detection method of optical remote sensing image based on cross-scale feature fusion | |
WO2022227770A1 (en) | Method for training target object detection model, target object detection method, and device | |
TW202207077A (en) | Text area positioning method and device | |
WO2022262757A1 (en) | Model training method, image detection method, and detection device | |
CN109948457B (en) | Real-time target recognition method based on convolutional neural network and CUDA acceleration | |
CN110991444A (en) | Complex scene-oriented license plate recognition method and device | |
CN110415260B (en) | Smoke image segmentation and identification method based on dictionary and BP neural network | |
CN109522831B (en) | Real-time vehicle detection method based on micro-convolution neural network | |
CN110443279B (en) | Unmanned aerial vehicle image vehicle detection method based on lightweight neural network | |
CN113221956B (en) | Target identification method and device based on improved multi-scale depth model | |
CN114973057B (en) | Video image detection method and related equipment based on artificial intelligence | |
CN116486166A (en) | Power transmission line foreign matter identification detection method based on edge calculation | |
CN114863189B (en) | Intelligent image identification method based on big data | |
CN111524113A (en) | Lifting chain abnormity identification method, system, equipment and medium | |
CN116129135A (en) | Tower crane safety early warning method based on small target visual identification and virtual entity mapping | |
CN116071315A (en) | Product visual defect detection method and system based on machine vision | |
CN112446292B (en) | 2D image salient object detection method and system | |
CN116740572A (en) | Marine vessel target detection method and system based on improved YOLOX |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |