CN112906654A - Anti-vibration hammer detection method based on deep learning algorithm - Google Patents
Anti-vibration hammer detection method based on deep learning algorithm Download PDFInfo
- Publication number
- CN112906654A CN112906654A CN202110331746.0A CN202110331746A CN112906654A CN 112906654 A CN112906654 A CN 112906654A CN 202110331746 A CN202110331746 A CN 202110331746A CN 112906654 A CN112906654 A CN 112906654A
- Authority
- CN
- China
- Prior art keywords
- network
- data set
- data
- inspection robot
- deep learning
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
-
- 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
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Multimedia (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a detection method of an electric appliance. The technical scheme is as follows: the method for detecting the vibration damper based on the deep learning algorithm comprises the following steps: step 1, collecting video data containing a vibration damper, and obtaining images frame by frame to generate a data set; meanwhile, related images are searched through a network and added into a data set; step 2, preprocessing the collected data, and expanding the data to generate similar images; step 3, marking the stockbridge dampers in the collected data set to obtain the coordinates of the candidate frame containing the target object; step 4, marking the preprocessed data set, inputting the marked data set into a MobileNet V3 network, and extracting feature graphs of three dimensions after network processing; step 5, inputting the feature diagram into a Yolo V3 module for training; inputting the trained optimal neural network model parameters into the inspection robot; and 6, detecting the vibration damper on the power transmission line by the line inspection robot. The method has good safety and low cost.
Description
Technical Field
The invention relates to a detection method of an electric power apparatus, in particular to a detection method of a shockproof hammer of a power transmission line.
Background
Transmission lines are important components of power systems and are responsible for the long-distance transmission of electrical energy. The lead vibrates violently under the influence of strong wind and bends frequently for a long time, so that the lead is fatigue-damaged, and serious potential safety hazards exist. At present, the main method adopted to reduce the frequency of the electric wire vibration caused by wind is to use a hardware fitting. A vibration damper is one of the important vibration-proof components of a high-voltage cable. By using the inertia of the anti-vibration hammer head, it consumes the vibration energy transmitted to the wire by wind and reduces the vibration damage of the overhead transmission line. The transmission line is in different regions and climates, so the natural environment is relatively severe. The vibration damper is susceptible to corrosion and rust due to prolonged exposure to harsh outdoor environments. If the clamping jig becomes loose, the vibratory hammer can be lost. Slippage, shifting, overturning, etc. occurs, which can lead to poor performance of the damper. If these problems cannot be solved in time, a serious power transmission accident is easily caused. Such transmission accidents have serious consequences, such as large-scale blackouts or transmission line damages, adversely affect the stable operation of the power system, and cause serious economic losses to the country and society. Therefore, correctly positioning the vibration damper, finding out the fault in time and taking remedial measures are very important for promoting the effective use of the power transmission line and prolonging the service life of the power transmission line. However, since the terrain varies greatly throughout a country, many areas of the transmission line must cross mountains and rivers as well as certain communication holes. These factors bring great inconvenience to inspection and maintenance of the transmission line.
At present, domestic and foreign inspection mainly depends on the movement of inspectors along the line, and various inspection equipment is manually used for inspection, but the inspection efficiency is low, the cost is high, and the phenomena of error inspection and detection omission are many. Power grid companies in advanced areas at home and abroad use helicopters or unmanned aerial vehicles to inspect high-voltage lines, thereby improving efficiency. However, the cost of using a helicopter is too high. And unmanned aerial vehicle is at some communication blind areas wayward, is influenced by transmission line and other common barriers easily, and its operation is also not complete safety. These are all issues that need to be addressed urgently. Therefore, the invention provides the inspection robot for inspection, which replaces the inspector to inspect the line, thereby reducing the labor intensity of inspection operation, reducing the inspection cost, improving the inspection efficiency, improving the safety of inspection operation and further improving the management and maintenance level of the line.
The precision of the traditional vibration damper falling defect detection based on machine learning is difficult to reach 80%. Deep learning is a technique developed based on artificial neural networks. In recent years, deep learning algorithms are rapidly developing. The difference between it and traditional machine learning techniques is that it can achieve 95% accuracy by learning from an independent network without relying on manual feature extraction. According to the existing research results, the deep learning algorithm has better recognition and classification performance than the conventional image algorithm. However, the hierarchy of network models with generally good performance is more complex. This means that a lot of computing power must be spent to provide powerful performance. It is desirable to integrate the convenience of convolutional neural networks in computer vision into some portable embedded devices to achieve a more convenient life. To better meet the demand of many low computing power devices, lightweight networks have recently become a research hotspot. MobileNet V3 is a lightweight network that enables deep learning algorithms to be applied to certain low-power edge devices. yoloV3 is a single stage algorithm, which is relatively fast. According to the invention, the MobileNet V3+ yoloV3 is used, and a lightweight network is embedded into the inspection robot, so that the detection work of the vibration damper can be better completed.
Disclosure of Invention
The invention aims to overcome the defects of the background technology, and provides a vibration damper detection method based on a deep learning algorithm.
In order to achieve the purpose, the invention provides the following technical scheme:
the method for detecting the vibration damper based on the deep learning algorithm is sequentially carried out according to the following steps:
step 4, labeling the preprocessed data set, inputting the labeled data set into a MobileNet V3 network, and extracting characteristic graphs of three dimensions after network processing, wherein the characteristic graphs are respectively 13 × 13,26 × 26 and 52 × 52;
and 6, detecting the vibration damper on the power transmission line by the line inspection robot.
And the preprocessing operation in the step 2 comprises noise increasing, atomizing, translation, rotation and twisting.
The detection step in the step 6 is as follows: the inspection robot collects the pictures in the process, and then carries out defogging operation on the pictures; the process of this operation is to remove some impurity factors present in the picture, so that the features of the picture are clearer. Then judging the processed picture by using the current optimal model; thereby obtaining the detection result.
The MobileNet V3 network processing in step 4 includes: after the size of an input image is adjusted to 416 multiplied by 416, sequentially carrying out a plurality of structural processing of a convolution layer, a pooling layer and a deep convolution in a network, and finally outputting feature maps of three dimensions from three branches of a sixth bottleneck layer, a twelfth bottleneck layer and a last convolution layer of the network respectively; the feature maps of three dimensions are respectively 13 × 13,26 × 26 and 52 × 52.
In step 6, the inspection robot also collects data in the process and reserves the data for the next training model to improve the inspection accuracy.
The invention has the beneficial effects that: the inspection robot is adopted, so that inspection can be automatically carried out on the power transmission line, manual control is not needed, and labor cost is greatly saved. The inspection robot can also be charged in time through the power transmission line, and the condition that the electric power is insufficient in the working process is not worried about. Compare helicopter, unmanned aerial vehicle, all have great advantage in aspects such as security, cost. Many target detection algorithms require a great deal of effort to operate, and are difficult to implement for some common devices. Many existing technologies adopt a network data transmission mode to transmit data to a cloud end, and perform online detection through a remote server. However, the distribution of the transmission lines is wide, and the positions of some transmission lines are communication blind areas, so that real-time detection cannot be performed. Therefore, the invention adopts MobileNet V3+ YooloV 3 (the prior art) to embed the lightweight network into the inspection robot, so that the deep learning algorithm with high accuracy has wider application, and the problem that the communication blind area can not be monitored in real time is well solved.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention.
Fig. 2 is a schematic view of the working state of the inspection robot.
Fig. 3 is a network structure diagram of the darknet 53.
Fig. 4 is a comparison graph before and after a residual network modification.
FIG. 5 is a schematic diagram of the structure of the Yolo V3 algorithm module.
FIG. 6 is a state diagram of the MobileNet V3 lightweight identification network in combination with the Yolo V3 single-phase algorithm.
Fig. 7 is an example of three images in a data set collected in embodiment 1.
Fig. 8 is an example of three images after processing one of the pictures of fig. 8.
FIG. 9 is an example of labeling the processed image of FIG. 8.
Fig. 10 is a diagram showing the effect of detection after an image including a vibration damper is tested.
Detailed Description
The following describes the method for detecting a vibration damper based on a deep learning algorithm according to the embodiment shown in the drawings.
And 2, because the collected data set is limited by various conditions, various scenes in the real environment, such as heavy fog, rain, snow and the like, can be simulated better. And carrying out operations such as noise increasing, atomization, translation, rotation, distortion and the like on the collected data, and expanding the data to generate a similar image so as to achieve the purpose of data enhancement.
In deep learning, the more complex the convolutional neural network model is, the stronger the ability to express objects, which will result in good training data and poor test data results. Therefore, massive data are needed to avoid the over-fitting condition, and the trained model is ensured to have good detection effect on new data. In order to avoid the influence of too high fitting degree on the detection effect, a large amount of sample data support is needed. The richer the sample data is, the higher the detection accuracy is. The purpose of step 1 and step 2 is to enrich the data set of the training sample.
And 3, marking the shockproof hammer in the acquired data set image by using an image marking tool labelIm to obtain candidate frame coordinates and a label containing the target object.
And 4, inputting the calibrated simulation data set into a MobileNet V3 for feature extraction.
The backbone network of the original YOLO V3 algorithm was Darknet53, which was a superposition of residual units. Joseph Redman's experiments show that the Darknet-53 model outperforms ResNet-101, ResNet-152 and Darknet-19 in terms of a balance between accuracy and classification efficiency.
The network structure of Darknet53 is shown in FIG. 3 (the default input in the figure is 256 × 256 pictures, but in practice YOLO v3 suggests an input picture size of 416 × 416, so 256 × 256, 128 × 128, 64 × 64, 32 × 32, 16 × 16, 8 × 8 in FIG. 3 should be modified to 416 × 416, 208 × 208, 104 × 104, 52 × 52, 26 × 26, 13 × 13, respectively).
To make the classification accuracy high, many models become deeper and deeper in depth, as well as more and more complex, as is the depth residual network (ResNet), which reaches layer 152. However, such a large and complex model is difficult to be applied in some practical application scenarios. Such as low performance embedded control development boards. Firstly, models are too large and face the problem of insufficient memory, and secondly, some scenarios require low latency or fast response. Therefore, the invention abandons Darknet53 and adopts lightweight network MobileNet V3. The method integrates the ideas of the following three models: the depth separable convolution of MobileNetV1, the inverse residual structure with linear bottlenecks of MobileNetV2, and the lightweight attention model based on squeeze and excitation structure of mnsety. The following four points are the main improvements of MobileNet V3 (see fig. 4):
1) introducing SE structures
The SE structure was added to the bottlenet structure and placed after the depthwise filter. In the structure including SE, the channel of the expansion layer was changed to 1/4 as it was. In so doing, not only is accuracy improved, but also no more time is consumed.
2) Modifying tail structure
Before the avg pooling of mobilenetv2, a 1x1 convolutional layer was present, which was now placed behind the avg pooling. The size of the profile was then changed by avg pooling. The characteristic diagram is reduced from 7 × 7 to 1 × 1, and after the characteristic diagram is reduced to 1 × 1, the dimension is improved by using 1 × 1. In this way, the amount of calculation can be greatly reduced. To further reduce the amount of computation, the first 3 × 3 convolution and the first 1 × 1 convolution are directly deleted, and the structure shown in the second row of fig. 4 becomes. (as can be seen from fig. 4: comparison before and after modification, the modified version can discard three expensive layers at the network end without loss of accuracy).
3) Modifying the number of channels
The header convolution kernel in mobilene v2 uses 32 x 3, while the header convolution kernels in large and small versions of mobilene 3 use 16 x 3, which both guarantees accuracy and reduces the speed by 3 ms.
4) Changes in non-linear transformations
H-swish was used instead of swish. The sigmoid calculation is relatively time consuming, which is more prominent when moving the terminal, so ReLU6(x +3)/6 will be used to approximate the replacement sigmoid. The advantages of ReLU are that 1, the limitation of the platform is reduced, the calculation operation can be carried out on any software and hardware platform, and 2, the potential precision loss is eliminated by using h-swish to replace switch in the quantification process. In the quantization mode, the efficiency is improved by about 15%, and in the deep network, the effect of h-swish is more obvious.
And 5, inputting the features extracted by the MobileNet V3 into a Yolo V3 module for training. And (5) performing loop iteration and optimizing the model parameters. And inputting the trained optimal neural network model parameters into the inspection robot, and putting the inspection robot into practice.
The single-phase detection framework eliminates the region proposal and realizes complete end-to-end, wherein the Yolo algorithm is the most classical one. Compared with a two-stage network, the accuracy is slightly lower, but the detection speed is greatly improved, and the effect of real-time target detection can be achieved. For the sake of fast R-CNN, Yolo V2 introduced the Anchor mechanism. And the design of the anchor is improved from traditional manual to K-Means clustering generation. And meanwhile, the deep and shallow layer characteristics are connected, so that the detection precision of the small object is obviously improved. To further improve the detection effect, Yolo V3 uses a better basic classification network and classifier. By using the residual error network structure for reference, the Yolo V3 forms a deeper network layer, and multi-scale detection is adopted, so that the mAP and small object detection effect is improved.
The major improvements of Yolo V3 (see fig. 5) are: adjusting a network structure; detecting an object by using the multi-scale features; object classification was performed using Logistic instead of softmax. In the aspect of basic image feature extraction, the Yolo V3 adopts a mode of fusing a plurality of scales for prediction. The original Yolo V2 has a passhigh layer, which is intended to improve the accuracy of detection of small objects. In Yolo V3, an fps (feature pyramid) like and fusion approach was used. The more detailed grid cells are, the more detailed objects can be detected by detecting on the feature maps of a plurality of scales. The improvement is quite obvious in terms of the detection effect of small targets. Finally, in Yolo V3, the object classification softmax is changed to logistic. Instead of predicting object classes using softmax, prediction is done using logistic output, which can support multi-labeled objects (e.g., one Person has two labels, Woman and Person).
According to the invention, a MobileNet V3 lightweight identification network and a Yolo V3 single-stage algorithm are combined (as shown in FIG. 6), and the shockproof hammer detection is carried out by means of the inspection robot, so that the requirements on hardware are reduced, high accuracy can be realized, and the real-time requirement is met. Because the accuracy of the Yolo V3 and the fastRCNN two-stage algorithm are not different, the accuracy in practical application is faster, and the accuracy in practical application needs higher hardware cost. The MobileNet V3 is lightweight network, allowing for perfect operation on low power embedded devices due to the scaling of the network architecture.
And 6, detecting the vibration damper on the power transmission line by the line inspection robot. The detection steps are as follows: the inspection robot collects the pictures in the process, and then carries out defogging operation on the pictures (the purpose of the defogging operation is to remove some impurity factors in the pictures so as to make the features of the pictures clearer); then judging the processed picture by using the current optimal model; thereby obtaining the detection result.
The inspection robot also collects data in the process and reserves the data for the next training model; in the long run, the accuracy will be greatly improved.
The invention has the characteristics that:
1. the MobileNet V3 lightweight recognition network and the Yolo V3 single-stage algorithm are combined, and the shockproof hammer detection is performed by means of the inspection robot, so that the requirements on hardware are reduced, the high accuracy can be realized, and the real-time requirement is met. Because the accuracy of the Yolo V3 and the fastRCNN two-stage algorithm are not different, the accuracy in practical application is faster, and the accuracy in practical application needs higher hardware cost. The MobileNet V3 is lightweight network, allowing for perfect operation on low power embedded devices due to the scaling of the network architecture.
2. Compared with the existing unmanned aerial vehicle and helicopter which are widely applied, the line patrol robot has the advantages that the cost is saved, and compared with the unmanned aerial vehicle which is not easy to control in some special environments, the line patrol robot is safer. The inspection robot can automatically charge on a power transmission line, and the phenomenon that the inspection robot cannot work due to insufficient electric quantity is avoided. And the inspection robot can move along the power transmission line by itself without manual management, so that the labor cost is greatly saved.
3. Because the transmission lines are widely distributed, some transmission lines are distributed in communication blind areas, the cost of more communication equipment can be increased when online detection is realized, and the actual detection effect can be influenced due to the poor quality of network transmission images. And the MobileNet V3+ Yolo V3 can locally realize a deep learning algorithm, and the requirement on network signals is not high.
Example 1
1) Collecting the picture and video data of the shockproof hammer (as shown in figure 7) from multiple angles, and adding the picture and video data into a data set;
2) the collected images are pre-processed to generate similar images (see fig. 8; in fig. 8: the image a is an original image, the image B is an image added with Gaussian noise, the image c is an image added with salt-pepper noise), and the data set is expanded;
3) the object of target detection is annotated using an image annotation tool such as labelImg (as shown in fig. 9):
4) inputting the calibrated simulation data set into a MobileNet V3 network for feature extraction
Specifically, by modifying the MobileNet network, for an input image with a size of 416 × 416, the dimensions of a finally output feature map are respectively 13 × 13,26 × 26, and 52 × 52, which respectively correspond to 13 × 13,26 × 26, and 52 × 52 grids of the input image, each grid predicts the sizes of multiple boxes (boxes), each box (box) contains 4 coordinate values, 1 confidence and a fixed number condition class probability, and the number of classes is 1, so the dimensions of the output feature map for the MobileNet network are 13 × 13x 3x (5+1), 26x 26x 3x (5+1), and 52x 52x 3x (5+ 1). And finally, inputting the three characteristics into a Yolo module, and finally outputting the confidence of the category.
5) Training and optimizing model
In the step, a Pythroch scientific calculation frame is used as an example, and the MobileNet + YOLO V3 algorithm is completed for detecting the damper. The main operating environments include: a Linux server, Pytorch, 4 × GeForce Tesla V10032G graphics card, CUDA 10.0, CUDNN7.6.1. The number of each sample was iteratively trained to 64, 16 batches total, and 1000 iterations were performed; the momentum factor is set to 0.9; attenuation coefficient was set to 0.0005; the learning rate adjustment strategy usage step and the initial learning rate are set to 0.001. In the training process, a multi-scale training strategy is adopted, and meanwhile, more images produced in the step 2) are added to increase training samples, so that the robustness of the network and the accuracy of images with different qualities are improved.
6) Detection verification
The jar test model tested 500 images containing the jar. These pictures have different backgrounds, lighting conditions and types of impact hammers. The effect of damper detection is shown in fig. 10, where the percentages shown in the figure represent the confidence of the detected damper (i.e., the probability that the detected object is the damper; and when the confidence is greater than 50%, the detection is accurate, the object in the box is determined to be the damper). Actual verification tests show that the improved detection algorithm can effectively detect the vibration damper and can realize real-time detection on the low-power-consumption embedded development board.
Claims (5)
1. The method for detecting the vibration damper based on the deep learning algorithm is sequentially carried out according to the following steps:
step 1, manually controlling a line inspection robot, collecting video data containing a vibration damper, and drawing frame by frame to generate a data set; meanwhile, related images are searched through a network and added into a data set; to ensure the integrity of the data set;
step 2, preprocessing the collected data, and expanding the data to generate similar images so as to achieve the purpose of data enhancement;
step 3, labeling the shockproof hammer in the collected data set by using image labeling software labelImg to obtain coordinates and a label of a candidate frame containing the target object;
step 4, labeling the preprocessed data set, inputting the labeled data set into a MobileNet V3 network, and extracting characteristic graphs of three dimensions after network processing, wherein the characteristic graphs are respectively 13 × 13,26 × 26 and 52 × 52;
step 5, inputting the three-dimensional feature maps extracted by the MobileNet V3 into a Yolo V3 module for training; performing loop iteration to optimize model parameters; inputting the trained optimal neural network model parameters into the inspection robot;
and 6, detecting the vibration damper on the power transmission line by the line inspection robot.
2. The deep learning algorithm-based stockbridge damper detection method according to claim 1, characterized in that: and the preprocessing operation in the step 2 comprises noise increasing, atomizing, translation, rotation and twisting.
3. The deep learning algorithm-based stockbridge damper detection method according to claim 2, characterized in that: the detection step in the step 6 is as follows: the inspection robot collects the pictures in the process, and then carries out defogging operation on the pictures; then judging the processed picture by using the current optimal model; thereby obtaining the detection result.
4. The deep learning algorithm-based stockbridge damper detection method according to claim 3, characterized in that: the MobileNet V3 network processing in step 4 includes: after the size of an input image is adjusted to 416 multiplied by 416, sequentially carrying out a plurality of structural processing of a convolution layer, a pooling layer and a deep convolution in a network, and finally outputting feature maps of three dimensions from three branches of a sixth bottleneck layer, a twelfth bottleneck layer and a last convolution layer of the network respectively; the feature maps of three dimensions are respectively 13 × 13,26 × 26 and 52 × 52.
5. The deep learning algorithm-based stockbridge damper detection method according to claim 4, wherein: in step 6, the inspection robot also collects data in the process and reserves the data for the next training model to improve the inspection accuracy.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110331746.0A CN112906654A (en) | 2021-03-29 | 2021-03-29 | Anti-vibration hammer detection method based on deep learning algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110331746.0A CN112906654A (en) | 2021-03-29 | 2021-03-29 | Anti-vibration hammer detection method based on deep learning algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112906654A true CN112906654A (en) | 2021-06-04 |
Family
ID=76109160
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110331746.0A Pending CN112906654A (en) | 2021-03-29 | 2021-03-29 | Anti-vibration hammer detection method based on deep learning algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112906654A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113468992A (en) * | 2021-06-21 | 2021-10-01 | 四川轻化工大学 | Construction site safety helmet wearing detection method based on lightweight convolutional neural network |
| CN113792578A (en) * | 2021-07-30 | 2021-12-14 | 北京智芯微电子科技有限公司 | Method, device and system for detecting abnormity of transformer substation |
| CN113869122A (en) * | 2021-08-27 | 2021-12-31 | 国网浙江省电力有限公司 | Distribution network engineering reinforced control method |
| CN116229278A (en) * | 2023-05-10 | 2023-06-06 | 广东电网有限责任公司珠海供电局 | Method and system for detecting rust defect of vibration damper of power transmission line |
| CN116298674A (en) * | 2023-02-14 | 2023-06-23 | 四川轻化工大学 | Same-pole double-circuit line fault phase selection based on MobileNet V3 |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107133943A (en) * | 2017-04-26 | 2017-09-05 | 贵州电网有限责任公司输电运行检修分公司 | A kind of visible detection method of stockbridge damper defects detection |
| CN111104906A (en) * | 2019-12-19 | 2020-05-05 | 南京工程学院 | Transmission tower bird nest fault detection method based on YOLO |
| CN112070730A (en) * | 2020-08-27 | 2020-12-11 | 宁波市电力设计院有限公司 | Anti-vibration hammer falling detection method based on power transmission line inspection image |
-
2021
- 2021-03-29 CN CN202110331746.0A patent/CN112906654A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107133943A (en) * | 2017-04-26 | 2017-09-05 | 贵州电网有限责任公司输电运行检修分公司 | A kind of visible detection method of stockbridge damper defects detection |
| CN111104906A (en) * | 2019-12-19 | 2020-05-05 | 南京工程学院 | Transmission tower bird nest fault detection method based on YOLO |
| CN112070730A (en) * | 2020-08-27 | 2020-12-11 | 宁波市电力设计院有限公司 | Anti-vibration hammer falling detection method based on power transmission line inspection image |
Non-Patent Citations (4)
| Title |
|---|
| 张明锐等: "《中国高铁丛书 高铁牵引供电系统》", 31 January 2019, 上海科学技术文献出版社 * |
| 杨罡等: "基于无人机前端和SSD算法的输电线路部件检测模型对比研究", 《太原理工大学学报》 * |
| 王夏夏: ""可用于自动巡航的电力设施目标的搜索与识别研究"", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》 * |
| 董召杰: "基于YOLOv3的电力线关键部件实时检测", 《电子测量技术》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113468992A (en) * | 2021-06-21 | 2021-10-01 | 四川轻化工大学 | Construction site safety helmet wearing detection method based on lightweight convolutional neural network |
| CN113468992B (en) * | 2021-06-21 | 2022-11-04 | 四川轻化工大学 | Construction site safety helmet wearing detection method based on lightweight convolutional neural network |
| CN113792578A (en) * | 2021-07-30 | 2021-12-14 | 北京智芯微电子科技有限公司 | Method, device and system for detecting abnormity of transformer substation |
| CN113869122A (en) * | 2021-08-27 | 2021-12-31 | 国网浙江省电力有限公司 | Distribution network engineering reinforced control method |
| CN116298674A (en) * | 2023-02-14 | 2023-06-23 | 四川轻化工大学 | Same-pole double-circuit line fault phase selection based on MobileNet V3 |
| CN116298674B (en) * | 2023-02-14 | 2023-12-26 | 四川轻化工大学 | Same-pole double-circuit line fault phase selection based on MobileNet V3 |
| CN116229278A (en) * | 2023-05-10 | 2023-06-06 | 广东电网有限责任公司珠海供电局 | Method and system for detecting rust defect of vibration damper of power transmission line |
| CN116229278B (en) * | 2023-05-10 | 2023-08-04 | 广东电网有限责任公司珠海供电局 | Method and system for detecting rust defect of vibration damper of power transmission line |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112906654A (en) | Anti-vibration hammer detection method based on deep learning algorithm | |
| CN109785289B (en) | Transmission line defect detection method and system and electronic equipment | |
| CN110033453A (en) | Based on the power transmission and transformation line insulator Aerial Images fault detection method for improving YOLOv3 | |
| CN110569841B (en) | Contact gateway key component target detection method based on convolutional neural network | |
| CN112200178B (en) | Transformer substation insulator infrared image detection method based on artificial intelligence | |
| CN113515655B (en) | Fault identification method and device based on image classification | |
| CN114973032B (en) | Deep convolutional neural network-based photovoltaic panel hot spot detection method and device | |
| CN113252701B (en) | Cloud edge cooperation-based power transmission line insulator self-explosion defect detection system and method | |
| CN113076992A (en) | Household garbage detection method and device | |
| CN112614130A (en) | Unmanned aerial vehicle power transmission line insulator fault detection method based on 5G transmission and YOLOv3 | |
| CN111507398A (en) | Transformer substation metal instrument corrosion identification method based on target detection | |
| CN113485432A (en) | Photovoltaic power station electroluminescence intelligent diagnosis system and method based on unmanned aerial vehicle | |
| CN114863118A (en) | Self-learning identification system and method based on external hidden danger of power transmission line | |
| CN114612803B (en) | Improved CENTERNET transmission line insulator defect detection method | |
| CN113902793B (en) | Method, system and electronic equipment for predicting end-to-end building height based on single-vision remote sensing image | |
| CN114359167B (en) | Insulator defect detection method based on lightweight YOLOv < 4 > under complex scene | |
| CN117496223A (en) | Light insulator defect detection method and device based on deep learning | |
| CN108470141B (en) | Statistical feature and machine learning-based insulator identification method in distribution line | |
| Di et al. | Research on Real-Time Power Line Damage Detection Method Based on YOLO Algorithm | |
| Zhang et al. | A Real-Time Wind Turbine Blade Damage Detection Method Based on an Improved YOLOv5 Algorithm | |
| Huang et al. | Insulator detection based on deep learning method in aerial images for power line patrol | |
| CN112070730A (en) | Anti-vibration hammer falling detection method based on power transmission line inspection image | |
| Tong et al. | Research on surface defect detection technology of long-distance and long-span FAST cable | |
| CN116912563A (en) | Defect identification method and device for power grid equipment | |
| CN117853780A (en) | Power transmission line hardware detection method based on improved YOLOv5 model |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210604 |