CN110796107A - Power inspection image defect identification method and system and power inspection unmanned aerial vehicle - Google Patents
Power inspection image defect identification method and system and power inspection unmanned aerial vehicle Download PDFInfo
- Publication number
- CN110796107A CN110796107A CN201911065444.2A CN201911065444A CN110796107A CN 110796107 A CN110796107 A CN 110796107A CN 201911065444 A CN201911065444 A CN 201911065444A CN 110796107 A CN110796107 A CN 110796107A
- Authority
- CN
- China
- Prior art keywords
- network
- target
- quantization
- inspection image
- classification
- 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
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Quality & Reliability (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for identifying defects of an electric power inspection image, which comprises the following steps: creating and cascading a target detection network and a plurality of classification networks; acquiring a plurality of frames of inspection image samples, labeling targets in the inspection image samples, and generating a training sample set; training the cascade network by adopting a training sample set, wherein the quantization parameter of each network layer is related to the quantization series and the quantization range of the network layer in which the network layer is positioned; and identifying the defects in the newly acquired inspection image by adopting the trained cascade network. The invention can provide an effective FPGA airborne identification system aiming at the operation and maintenance of the power grid tower and the overhead line, and the corresponding quantization function can ensure that different channels of different network layers can be properly quantized, thereby furthest keeping the accuracy of the network; by cascading the target detection network and the classification network, the defect detection accuracy is greatly improved, and the unmanned aerial vehicle inspection of the power grid really realizes automatic identification.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a power inspection image defect identification method and system and a power inspection unmanned aerial vehicle.
Background
The transmission line of national grid construction reaches several million kilometers, and the operation and maintenance of such long-distance distribution network overhead line need comprehensively rely on the comprehensive inspection of unmanned aerial vehicle. The common unmanned aerial vehicle platform is provided with a high-speed image module, an infrared imaging sensing module, an ultraviolet imaging sensing module, an ultrahigh frequency partial discharge sensing module, an ultrasonic partial discharge sensing module and other devices so as to complete comprehensive inspection operation. The high-speed image module mainly achieves a video image acquisition function of visible light under a flight condition, carries a high-resolution visible light camera through the unmanned aerial vehicle to inspect and shoot visible light images, and can detect defect information such as strand scattering, strand breakage, insulator falling, pin falling, nut missing, tower footing tree shielding, tower wire net bird nest and the like of wires of a distribution network line. The infrared imaging sensing module (thermal infrared imager) is mainly used for detecting heating faults of components such as wire joints, wire clamps and the like. The solution is based on manual graph judgment to identify defects, and the overhauling efficiency is low.
In order to solve this problem, the defect detection of some parts is then identified by the respective identification system. For example, the invention with the patent number CN110069975A provides a method for identifying a wire stranding defect in a "wire stranding identification method and system based on a neural network", the invention with the patent number CN110309865A provides a method for identifying a pin defect in a "pin defect classification image identification method for an unmanned aerial vehicle inspection power transmission line", and the invention with the patent number CN110378222A provides a method for identifying a damper defect in a "damper target detection and defect identification method and device for a power transmission line".
However, the prior art can only detect one or a limited number of defects, and in fact, one inspection task usually needs to detect most or all parts of the whole transmission line, the types of transmission line parts and the types of defects are various, and if the targeted identification system is replaced each time, the inspection task becomes quite heavy, and if an identification system capable of detecting multiple targets is created, the following two problems exist: firstly, the identification system is complex in structure and difficult to create; and for an identification system for simultaneously detecting various defects, the parameters adopted by the identification modes of different targets are different, even parameters such as abnormal weight exist, the calculated amount of the whole identification system is large, and the actual detection precision and accuracy are low.
Disclosure of Invention
The invention aims to provide a method and a system for identifying the defects of an electric power inspection image and an electric power inspection unmanned aerial vehicle, and provides an effective FPGA (field programmable gate array) airborne identification system aiming at the operation and maintenance of a power grid tower and an overhead line, wherein corresponding quantization functions can ensure that different channels of different network layers can be properly quantized, and the accuracy of a network is kept to the maximum extent; by cascading the target detection network and the classification network, the defect detection accuracy is greatly improved, and the unmanned aerial vehicle inspection of the power grid really realizes automatic identification.
In order to achieve the above object, with reference to fig. 1, the present invention provides a method for identifying defects in an electric power inspection image, where the method includes:
s1: the method comprises the steps that a target detection network and a plurality of classification networks are created and cascaded, the target detection network is used for positioning target image information in an inspection image and intercepting and transmitting the positioned target image information to the next-stage classification network, and the classification networks are used for judging whether defects exist in the corresponding target image information or not;
s2: acquiring a plurality of frames of inspection image samples, marking targets in the inspection image samples, wherein marking items comprise target types, target positions, defect types and defect positions, and generating a training sample set;
s3: training the cascade network by adopting a training sample set, wherein the quantization parameter of each network layer is related to the quantization series and the quantization range of the network layer in which the network layer is positioned;
s4: and identifying the defects in the newly acquired inspection image by adopting the trained cascade network.
In a further embodiment, in step S2, the target types include wires, insulators, vibration dampers, pin nuts, towers, and bird nests.
In a further embodiment, the quantization policy of the cascade network is: the network training adopts floating point operation, and the network prediction adopts integer operation.
In a further embodiment, the quantization strategy comprises the steps of:
establishing correspondence between real numbers and bit expressions of the values thereof;
the network is trained by floating point operation, quantization effect is inserted in the forward propagation process of training, and floating point operation is maintained in the backward propagation process.
In a further embodiment, said establishing correspondence between a real number and a bit representation of its value is:
let the affine mapping of the real number r be:
r＝S(q-Z)
wherein r is a real number, S is a scale constant, and Z is a zero constant, the operation of multiplying the two nxn matrices is completed by integer operation:
let r(i,j)＝S(q(i,j)-Z) is a matrix element r(i,j)The quantization parameter is (S, Z); the definition by matrix multiplication can be:
in the above formula, the first and second carbon atoms are,and (5) calculating off line.
In a further embodiment, the inserting quantization effect in the training forward propagation process refers to: in forward propagation, the weights are quantized before convolution with the input, the excitation function is quantized point by point when used for inference, and the quantization parameter of each network layer is related to the quantization series and quantization range of the network layer where the quantization parameter is located.
In a further embodiment, in the quantization process, the quantization function q used is:
clamp(r；,a,b)：＝Min(Max(x,a),b)
wherein r is the real value to be quantized, [ a, b ] is the quantization range, n is the number of quantization steps, [ DEG ] is the nearest neighbor rounding;
the quantization range of weight quantization is represented by a: min weights, b Max weights; the quantization range of the excitation is obtained by an exponential moving average method in the training process.
In a further embodiment, the loss function of the target detection network is composed of a coordinate error, an IOU error, and a classification error, and the loss function is respectively:
error of target coordinate
Target IOU error
Target classification error
In the formula, an input image is divided into S multiplied by S grids, and if the center position of a target group channel falls into a certain grid, the grid is responsible for detecting the object;
b Bounding boxes and confidence coefficients thereof and C class probabilities are predicted for each grid; the Bounding Box information (x, y, w, h) is the offset of the center position of the object relative to the grid position, and the width and the height of the object, and are normalized; the confidence degree reflects whether the target is contained or not and the accuracy of the position under the condition of containing the target, and is defined asWherein Pr (object) is e {0, 1 };
wherein λ iscoordFor the weight of the positioning error, λnoobjAnd λnodefectWeight of the target classification error;
judging whether the jth Bounding Box in the ith grid is responsible for the target or not, and the Bounding Box with the largest IOU of the group of the target is responsible for the coordinate prediction of the target;
and judging whether the center point of the target falls into a grid i, wherein the grid contains the center of the target, namely the grid is responsible for predicting the class probability of the target.
Based on the identification method, the invention also provides an electric power inspection image defect identification system, which comprises a cascaded target detection network and a plurality of classification networks, wherein the target detection network is used for positioning target image information in an inspection image and intercepting the positioned target image information to transmit to the next-stage classification network, and the classification networks are used for judging whether defects exist in the corresponding target image information;
the classification network adopts SVM or Alexnet network.
The invention also provides an electric power inspection unmanned aerial vehicle which comprises an unmanned aerial vehicle body, an image acquisition device and an FPGA (field programmable gate array) onboard identification system carrying the electric power inspection image defect identification system according to claim 9.
The electric power inspection image defect identification system provided by the invention is an FPGA airborne identification system, and the real-time processing capacity reaches 4K/60 fps. The electric power inspection image defect identification method covers target part positioning and defect detection and comprises a target detection network and a plurality of subsequent classification networks. The target detection network is similar to a YOLO V3 convolutional neural network, and is used for positioning target information such as distribution network line wires, insulators, vibration dampers, small hardware fittings (pins, nuts and the like), tower foundations, bird nests and the like, intercepting the target image information and transmitting the target image information to a next-level classification network (Alexnet, SVM); the classification networks are respectively a wire classification network, an insulator classification network, a vibration damper classification network, a small hardware fitting classification network and a tower foundation classification network and are used for judging whether the corresponding target is defective or not.
Compared with the prior art, the technical scheme of the invention has the following remarkable beneficial effects:
(1) an effective FPGA airborne identification system is provided for operation and maintenance of power grid towers and overhead lines, and corresponding quantization functions can ensure that different channels of different network layers can be properly quantized, so that the accuracy of the network is kept to the maximum extent.
(2) By cascading the target detection network and the classification network, the defect detection accuracy is greatly improved, and the unmanned aerial vehicle inspection of the power grid really realizes automatic identification.
It should be understood that all combinations of the foregoing concepts and additional concepts described in greater detail below can be considered as part of the inventive subject matter of this disclosure unless such concepts are mutually inconsistent. In addition, all combinations of claimed subject matter are considered a part of the presently disclosed subject matter.
The foregoing and other aspects, embodiments and features of the present teachings can be more fully understood from the following description taken in conjunction with the accompanying drawings. Additional aspects of the present invention, such as features and/or advantages of exemplary embodiments, will be apparent from the description which follows, or may be learned by practice of specific embodiments in accordance with the teachings of the present invention.
Drawings
The drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component that is illustrated in various figures may be represented by a like numeral. For purposes of clarity, not every component may be labeled in every drawing. Embodiments of various aspects of the present invention will now be described, by way of example, with reference to the accompanying drawings, in which:
fig. 1 is a flow chart of the power inspection image defect identification method of the invention.
Fig. 2 is a schematic diagram of the structure of the cascaded network of the present invention.
FIG. 3 is a schematic diagram of convolutional layer forward inference for pure integer arithmetic in accordance with the present invention.
FIG. 4 is a schematic diagram of the principles of simulated quantization training of convolutional layers of the present invention.
Detailed Description
In order to better understand the technical content of the present invention, specific embodiments are described below with reference to the accompanying drawings.
With reference to fig. 1, the invention aims to provide an electric power inspection image defect identification method, which can greatly improve the detection accuracy and enable unmanned aerial vehicle inspection of a power grid to really realize automatic identification.
The method for identifying the defects of the power inspection image comprises two parts of building a training data set and creating a cascade network model.
And regarding the construction of the training data set, the initial training data set is formed by pictures acquired by the approach observation of the unmanned aerial vehicle, and then the final training data set is completed by manually marking the positions of the targets and the defects.
Working principle about the cascaded network model: the target detection network firstly positions the ground wire, the insulator, the vibration damper, the pin nut, the tower footing, the bird nest and the like, and then the corresponding classification network judges whether the target has defects one by one.
With reference to fig. 2, the cascade network includes an object detection network and a plurality of subsequent classification networks. The loss function of the target detection network model consists of a coordinate error, an IOU (interaction over Union) error and a classification error.
Specifically, the formula is shown as follows:
error of target coordinate
Target IOU error
Target classification error
In the formula, the input image is divided into S × S grids, and if the center position of a target group channel falls into a certain grid, the grid is responsible for detecting the object. Each grid predicts B Bounding boxes and their confidence, and C class probabilities. The Bounding Box information (x, y, w, h) is the offset of the center position of the object from the grid position and the width and height, all normalized. The confidence degree reflects whether the target is contained or not and the accuracy of the position under the condition of containing the target, and is defined as
Where Pr (object) is ∈ {0, 1 }. Lambda [ alpha ]coordFor the weight of the positioning error, λnoobjAnd λnodefectWeight of the target classification error;
judging whether the jth Bounding Box in the ith grid is responsible for the target or not, and the Bounding Box with the largest IOU of the group of the target is responsible for the coordinate prediction of the target;
judging whether the center point of the target falls into a grid i, wherein the grid contains the center of the target, and predicting the targetThe class probability of (2). And training the SGD algorithm to obtain appropriate network parameters, and predicting to obtain target positioning and classification.
The classification network is a common SVM or Alexnet network.
The defect identification system is an FPGA airborne identification system, and the corresponding quantization strategy is as follows: the network training adopts floating point operation, and the network prediction adopts integer operation. Fig. 3 and 4 represent the convolutional layer prediction process for pure integer arithmetic and the convolutional layer training process for insertion of analog quantization, respectively. In fig. 3, both input and output are expressed as 8-bit integers, and convolution includes 8-bit integer arithmetic and 32-bit integer accumulation, according to the quantization strategy. The RELU6 nonlinear layer includes only 8-bit integer operations. In fig. 4, all variables and calculations are performed by 32-bit floating point operations, and weight quantization nodes and excitation quantization nodes are inserted to simulate the variable quantization effect.
The Logistic function in the network is directly realized by a fixed-point method. The specific quantization strategy is as follows:
first, a correspondence between a real number and a bit representation of its value is established, the affine mapping of the real number r being: r ═ S (q-Z) where r is a real number, S is a scale constant, and Z is a zero point constant. The multiplication of the two nxn matrices can be performed by integer operations: let r(i,j)＝S(q(i,j)-Z) is a matrix element r(i,j)The quantization parameter is (S, Z). Definition by matrix multiplication:
in the above formula, the first and second carbon atoms are,
the calculation can be done off-line.
Secondly, we train the network with floating point operations, but insert quantization effects in the forward propagation process of the training, and maintain the floating point operations in the backward propagation process. In forward propagation, weights are quantized before convolution with the input, and the excitation function is quantized point-by-point when used for inference. The quantization parameter of each network layer depends on the quantization series and the quantization range, and the specific quantization function q is as follows:
clamp(r；,a,b)：＝Min(Max(x,a),b)
wherein r is the real number to be quantized, [ a, b ]]For quantization range, n is the number of quantization levels, and "·" is the nearest neighbor rounding. For 8-bit quantization, we can fix the quantization level n to 28. The quantization range of weight quantization we can be represented by a: min weights, b Max weights. The quantization range of the excitation is obtained by an exponential moving average method (exponential moving average) during the training process. Different from the existing quantization mode for quantizing all channels of the same network layer with the same precision, the method still has excellent performance even if the weight ranges of different output channels are very different or some abnormal weights exist.
In this disclosure, aspects of the present invention are described with reference to the accompanying drawings, in which a number of illustrative embodiments are shown. Embodiments of the present disclosure are not necessarily defined to include all aspects of the invention. It should be appreciated that the various concepts and embodiments described above, as well as those described in greater detail below, may be implemented in any of numerous ways, as the disclosed concepts and embodiments are not limited to any one implementation. In addition, some aspects of the present disclosure may be used alone, or in any suitable combination with other aspects of the present disclosure.
Although the present invention has been described with reference to the preferred embodiments, it is not intended to be limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, the protection scope of the present invention should be determined by the appended claims.
Claims (10)
1. The electric power inspection image defect identification method is characterized by comprising the following steps:
s1: the method comprises the steps that a target detection network and a plurality of classification networks are created and cascaded, the target detection network is used for positioning target image information in an inspection image and intercepting and transmitting the positioned target image information to the next-stage classification network, and the classification networks are used for judging whether defects exist in the corresponding target image information or not;
s2: acquiring a plurality of frames of inspection image samples, marking targets in the inspection image samples, wherein marking items comprise target types, target positions, defect types and defect positions, and generating a training sample set;
s3: training the cascade network by adopting a training sample set, wherein the quantization parameter of each network layer is related to the quantization series and the quantization range of the network layer in which the network layer is positioned;
s4: and identifying the defects in the newly acquired inspection image by adopting the trained cascade network.
2. The power inspection image defect identification method according to claim 1, wherein in step S2, the target types include wires, insulators, vibration dampers, pin nuts, towers, bird nests.
3. The power inspection image defect identification method according to claim 1, wherein the quantization strategy of the cascade network is: the network training adopts floating point operation, and the network prediction adopts integer operation.
4. The power inspection image defect identification method according to claim 3, wherein the quantification strategy includes the steps of:
establishing correspondence between real numbers and bit expressions of the values thereof;
the network is trained by floating point operation, quantization effect is inserted in the forward propagation process of training, and floating point operation is maintained in the backward propagation process.
5. The power inspection image defect identification method according to claim 4, wherein establishing correspondence between real numbers and bit representations of the values thereof is:
let the affine mapping of the real number r be:
r＝S(q-Z)
wherein r is a real number, S is a scale constant, and Z is a zero constant, the operation of multiplying the two nxn matrices is completed by integer operation:
let r(i，j)＝S(q(i，j)-Z) is a matrix element r(i，j)The quantization parameter is (S, Z); the definition by matrix multiplication can be:
in the above formula, the first and second carbon atoms are,
and (5) calculating off line.
6. The power inspection image defect identification method according to claim 4, wherein the inserting quantification effects during the forward propagation of training refers to: in forward propagation, the weights are quantized before convolution with the input, the excitation function is quantized point by point when used for inference, and the quantization parameter of each network layer is related to the quantization series and quantization range of the network layer where the quantization parameter is located.
7. The power inspection image defect identification method according to claim 6, wherein in the quantization process, a quantization function q is adopted as follows:
clamp(r；，a，b)：＝Min(Max(x，a)，b)
wherein r is the real number to be quantized, [ a, b ]]For the quantization range, n is the number of quantized levels,
rounding the nearest neighbor;
the quantization range of weight quantization is represented by a: min weights, b: max weights; the quantization range of the excitation is obtained by an exponential moving average method in the training process.
8. The power inspection image defect identification method according to claim 1, wherein the loss function of the target detection network is composed of coordinate errors, IOU errors and classification errors, and respectively:
error of target coordinate
Target IOU error
Target classification error
In the formula, an input image is divided into S multiplied by S grids, and if the center position of a target group channel falls into a certain grid, the grid is responsible for detecting the object;
each one of which isB Bounding boxes and confidence coefficients thereof and C class probabilities are predicted in a grid mode; the Bounding Box information (x, y, w, h) is the offset of the center position of the object relative to the grid position, and the width and the height of the object, and are normalized; the confidence degree reflects whether the target is contained or not and the accuracy of the position under the condition of containing the target, and is defined as
Wherein Pr (object) is e {0, 1 };
wherein λ iscoordFor the weight of the positioning error, λnoobjAnd λnodefectWeight of the target classification error;
judging whether the jth Bounding Box in the ith grid is responsible for the target or not, and the Bounding Box with the largest IOU of the group of the target is responsible for the coordinate prediction of the target;
and judging whether the center point of the target falls into a grid i, wherein the grid contains the center of the target, namely the grid is responsible for predicting the class probability of the target.
9. An electric power inspection image defect identification system based on the identification method of claim 1, characterized in that the identification system comprises a target detection network and a plurality of classification networks which are cascaded, wherein the target detection network is used for positioning target image information in an inspection image and intercepting the positioned target image information to transmit to the next-stage classification network, and the classification networks are used for judging whether defects exist in the corresponding target image information;
the classification network adopts SVM or Alexnet network.
10. The utility model provides an unmanned aerial vehicle is patrolled and examined to electric power, its characterized in that, unmanned aerial vehicle is patrolled and examined to electric power includes unmanned aerial vehicle body, image acquisition device, carries with according to claim 9 image defect identification system's FPGA machine carries on identification system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911065444.2A CN110796107A (en) | 2019-11-04 | 2019-11-04 | Power inspection image defect identification method and system and power inspection unmanned aerial vehicle |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911065444.2A CN110796107A (en) | 2019-11-04 | 2019-11-04 | Power inspection image defect identification method and system and power inspection unmanned aerial vehicle |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110796107A true CN110796107A (en) | 2020-02-14 |
Family
ID=69442510
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911065444.2A Pending CN110796107A (en) | 2019-11-04 | 2019-11-04 | Power inspection image defect identification method and system and power inspection unmanned aerial vehicle |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110796107A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111311967A (en) * | 2020-03-31 | 2020-06-19 | 普宙飞行器科技（深圳）有限公司 | Unmanned aerial vehicle-based power line inspection system and method |
| CN111383223A (en) * | 2020-03-19 | 2020-07-07 | 天津理工大学 | Charging pile part line sequence color error-proofing detection method |
| CN111415345A (en) * | 2020-03-20 | 2020-07-14 | 山东文多网络科技有限公司 | Transformer substation ultraviolet image intelligent inspection algorithm and device based on deep learning |
| CN111460995A (en) * | 2020-03-31 | 2020-07-28 | 普宙飞行器科技（深圳）有限公司 | Unmanned aerial vehicle-based power line inspection method and inspection system |
| CN111614358A (en) * | 2020-04-30 | 2020-09-01 | 北京的卢深视科技有限公司 | Method, system, device and storage medium for feature extraction based on sub-channel quantization |
| CN111881937A (en) * | 2020-06-22 | 2020-11-03 | 深圳金三立视频科技股份有限公司 | Transmission line hardware target detection and defect identification method and terminal |
| CN111896850A (en) * | 2020-07-03 | 2020-11-06 | 国网湖南省电力有限公司 | System and method for detecting and positioning ultrahigh frequency partial discharge of unmanned aerial vehicle |
| CN112001902A (en) * | 2020-08-19 | 2020-11-27 | 上海商汤智能科技有限公司 | Defect detection method and related device, equipment and storage medium |
| CN112051270A (en) * | 2020-09-10 | 2020-12-08 | 广州云从人工智能技术有限公司 | Power transmission line defect detection method, system, equipment and medium |
| CN112102443A (en) * | 2020-09-15 | 2020-12-18 | 国网电力科学研究院武汉南瑞有限责任公司 | Marking system and marking method suitable for substation equipment inspection image |
| CN112229845A (en) * | 2020-10-12 | 2021-01-15 | 国网河南省电力公司濮阳供电公司 | Unmanned aerial vehicle high-precision winding tower intelligent inspection method based on visual navigation technology |
| CN112926401A (en) * | 2021-01-29 | 2021-06-08 | 广州中科智巡科技有限公司 | Transmission line hardware corrosion detection method and system |
| CN113160210A (en) * | 2021-05-10 | 2021-07-23 | 深圳市水务工程检测有限公司 | Drainage pipeline defect detection method and device based on depth camera |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2013145898A1 (en) * | 2012-03-27 | 2013-10-03 | 株式会社日立製作所 | Light detection device and method thereof |
| CN104736962A (en) * | 2012-09-05 | 2015-06-24 | 科磊股份有限公司 | Method for estimating and correcting misregistration target inaccuracy |
| CN106157339A (en) * | 2016-07-05 | 2016-11-23 | 华南理工大学 | The animated Mesh sequence compaction algorithm extracted based on low-rank vertex trajectories subspace |
| US20170091925A1 (en) * | 2013-02-01 | 2017-03-30 | Kla-Tencor Corporation | Detecting Defects on a Wafer Using Defect-Specific and Multi-Channel Information |
| CN107688849A (en) * | 2017-07-28 | 2018-02-13 | 北京深鉴科技有限公司 | A kind of dynamic strategy fixed point training method and device |
| CN109325454A (en) * | 2018-09-28 | 2019-02-12 | 合肥工业大学 | A kind of static gesture real-time identification method based on YOLOv3 |
| CN109829462A (en) * | 2019-01-16 | 2019-05-31 | 中南大学 | A kind of a variety of flowers recognition methods based on semantic segmentation |
| CN110009090A (en) * | 2019-04-02 | 2019-07-12 | 北京市商汤科技开发有限公司 | Neural metwork training and image processing method and device |
| CN110147807A (en) * | 2019-01-04 | 2019-08-20 | 上海海事大学 | A kind of ship intelligent recognition tracking |
| CN110163836A (en) * | 2018-11-14 | 2019-08-23 | 宁波大学 | Based on deep learning for the excavator detection method under the inspection of high-altitude |
| CN110378420A (en) * | 2019-07-19 | 2019-10-25 | Oppo广东移动通信有限公司 | A kind of image detecting method, device and computer readable storage medium |
-
2019
- 2019-11-04 CN CN201911065444.2A patent/CN110796107A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2013145898A1 (en) * | 2012-03-27 | 2013-10-03 | 株式会社日立製作所 | Light detection device and method thereof |
| CN104736962A (en) * | 2012-09-05 | 2015-06-24 | 科磊股份有限公司 | Method for estimating and correcting misregistration target inaccuracy |
| US20170091925A1 (en) * | 2013-02-01 | 2017-03-30 | Kla-Tencor Corporation | Detecting Defects on a Wafer Using Defect-Specific and Multi-Channel Information |
| CN106157339A (en) * | 2016-07-05 | 2016-11-23 | 华南理工大学 | The animated Mesh sequence compaction algorithm extracted based on low-rank vertex trajectories subspace |
| CN107688849A (en) * | 2017-07-28 | 2018-02-13 | 北京深鉴科技有限公司 | A kind of dynamic strategy fixed point training method and device |
| CN109325454A (en) * | 2018-09-28 | 2019-02-12 | 合肥工业大学 | A kind of static gesture real-time identification method based on YOLOv3 |
| CN110163836A (en) * | 2018-11-14 | 2019-08-23 | 宁波大学 | Based on deep learning for the excavator detection method under the inspection of high-altitude |
| CN110147807A (en) * | 2019-01-04 | 2019-08-20 | 上海海事大学 | A kind of ship intelligent recognition tracking |
| CN109829462A (en) * | 2019-01-16 | 2019-05-31 | 中南大学 | A kind of a variety of flowers recognition methods based on semantic segmentation |
| CN110009090A (en) * | 2019-04-02 | 2019-07-12 | 北京市商汤科技开发有限公司 | Neural metwork training and image processing method and device |
| CN110378420A (en) * | 2019-07-19 | 2019-10-25 | Oppo广东移动通信有限公司 | A kind of image detecting method, device and computer readable storage medium |
Non-Patent Citations (2)
| Title |
|---|
| BENOIT JACOB 等: "Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference", 《HTTPS://ARXIV.ORG/PDF/1712.05877.PDF》 * |
| 李宁 等: "基于无人机图像识别技术的输电线路缺陷检测", 《电子设计工程》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111383223A (en) * | 2020-03-19 | 2020-07-07 | 天津理工大学 | Charging pile part line sequence color error-proofing detection method |
| CN111415345A (en) * | 2020-03-20 | 2020-07-14 | 山东文多网络科技有限公司 | Transformer substation ultraviolet image intelligent inspection algorithm and device based on deep learning |
| CN111415345B (en) * | 2020-03-20 | 2023-09-12 | 山东文多网络科技有限公司 | Intelligent substation ultraviolet image inspection method and device based on deep learning |
| CN111460995A (en) * | 2020-03-31 | 2020-07-28 | 普宙飞行器科技（深圳）有限公司 | Unmanned aerial vehicle-based power line inspection method and inspection system |
| CN111311967A (en) * | 2020-03-31 | 2020-06-19 | 普宙飞行器科技（深圳）有限公司 | Unmanned aerial vehicle-based power line inspection system and method |
| CN111614358B (en) * | 2020-04-30 | 2023-08-04 | 合肥的卢深视科技有限公司 | Feature extraction method, system, equipment and storage medium based on multichannel quantization |
| CN111614358A (en) * | 2020-04-30 | 2020-09-01 | 北京的卢深视科技有限公司 | Method, system, device and storage medium for feature extraction based on sub-channel quantization |
| CN111881937A (en) * | 2020-06-22 | 2020-11-03 | 深圳金三立视频科技股份有限公司 | Transmission line hardware target detection and defect identification method and terminal |
| CN111896850A (en) * | 2020-07-03 | 2020-11-06 | 国网湖南省电力有限公司 | System and method for detecting and positioning ultrahigh frequency partial discharge of unmanned aerial vehicle |
| CN112001902A (en) * | 2020-08-19 | 2020-11-27 | 上海商汤智能科技有限公司 | Defect detection method and related device, equipment and storage medium |
| CN112051270A (en) * | 2020-09-10 | 2020-12-08 | 广州云从人工智能技术有限公司 | Power transmission line defect detection method, system, equipment and medium |
| CN112051270B (en) * | 2020-09-10 | 2021-08-13 | 广州云从人工智能技术有限公司 | Power transmission line defect detection method, system, equipment and medium |
| CN112102443A (en) * | 2020-09-15 | 2020-12-18 | 国网电力科学研究院武汉南瑞有限责任公司 | Marking system and marking method suitable for substation equipment inspection image |
| CN112229845A (en) * | 2020-10-12 | 2021-01-15 | 国网河南省电力公司濮阳供电公司 | Unmanned aerial vehicle high-precision winding tower intelligent inspection method based on visual navigation technology |
| CN112926401A (en) * | 2021-01-29 | 2021-06-08 | 广州中科智巡科技有限公司 | Transmission line hardware corrosion detection method and system |
| CN113160210A (en) * | 2021-05-10 | 2021-07-23 | 深圳市水务工程检测有限公司 | Drainage pipeline defect detection method and device based on depth camera |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110796107A (en) | Power inspection image defect identification method and system and power inspection unmanned aerial vehicle | |
| CN112183788B (en) | Domain adaptive equipment operation detection system and method | |
| CN110827251B (en) | Power transmission line locking pin defect detection method based on aerial image | |
| CN112001260A (en) | Cable trench fault detection method based on infrared and visible light image fusion | |
| CN112508950B (en) | Anomaly detection method and device | |
| Hao et al. | Detection of bird nests on power line patrol using single shot detector | |
| CN113962973A (en) | Power transmission line unmanned aerial vehicle intelligent inspection system and method based on satellite technology | |
| CN117496223A (en) | Light insulator defect detection method and device based on deep learning | |
| CN116912589A (en) | Medium-voltage distribution cable state evaluation method based on multi-mode fusion | |
| CN116338392A (en) | Method, device and equipment for identifying lightning discharge defects of glass insulator | |
| CN111667461A (en) | Method for detecting abnormal target of power transmission line | |
| CN117809083B (en) | Cable joint fault detection method and system based on infrared or ultraviolet images | |
| CN114241292A (en) | Power transmission line icing monitoring method and device based on intelligent gateway | |
| CN114119454A (en) | Device and method for smoke detection of power transmission line | |
| CN111476062A (en) | Lane line detection method and device, electronic equipment and driving system | |
| CN110782431B (en) | High-voltage wire icing area detection method based on deep learning | |
| CN112434600A (en) | Unmanned aerial vehicle inspection method and system | |
| CN113486873B (en) | Transformer substation equipment inspection method and system based on big data and artificial intelligence | |
| CN116895036A (en) | Deep learning-based farmland protection early warning method and device | |
| CN113705001B (en) | Power transmission line fault detection method, system and embedded equipment | |
| CN116738355A (en) | Electric power multi-mode feature level fusion method considering data difference | |
| CN114677777A (en) | Equipment inspection method, inspection system and terminal equipment | |
| CN113837526A (en) | Intelligent analysis method and system for transformer substation equipment based on edge calculation | |
| CN118314488B (en) | Extra-high voltage transformer station space-earth multi-scale re-decision target detection method | |
| CN111402552A (en) | Power equipment monitoring and alarming system and method |
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: 20200214 |