CN112464756B - Insulator defect identification-oriented image quantization method - Google Patents
Insulator defect identification-oriented image quantization method Download PDFInfo
- Publication number
- CN112464756B CN112464756B CN202011271571.0A CN202011271571A CN112464756B CN 112464756 B CN112464756 B CN 112464756B CN 202011271571 A CN202011271571 A CN 202011271571A CN 112464756 B CN112464756 B CN 112464756B
- Authority
- CN
- China
- Prior art keywords
- image
- quantization
- insulator
- block
- aerial vehicle
- 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.)
- Active
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
- 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
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to an image quantization method for insulator defect identification, which comprises the following steps: (S1) selecting an insulator shot by unmanned aerial vehicle inspection and a picture of a fault of the insulator, and training images by using a fast RCNN network; (S2) extracting image features by using the fault recognition model trained in the step (S1); (S3) extracting the characteristics of the key areas in the low-layer edge information according to the step (S2); (S4) according to (S3), back-pushing the pooling operation using the pooling characteristics of the feature extraction network; (S5) according to (S4), back-stepping the pooling operation using the convolution characteristics of the feature extraction network; (S6) finally deriving a saliency flag map to guide quantification. Compared with a JPEG compression method without quantization guidance, the method has better defect identification effect on small electric components such as insulators and higher identification accuracy under the same code rate.
Description
Technical Field
The invention relates to an image quantization method, in particular to an image quantization method for insulator defect identification.
Background
The power grids of six large-span provinces of south, north, china, east, north and China are established in China, a large amount of image information acquired by daily inspection operation is unrealistic to perform fault identification only by human eyes, and the high accuracy of fault identification can be realized by fast development of deep learning and machine vision, so that the machine identification replaces part of human eyes to perform fault detection, and the characteristics of uninterrupted fault identification and feedback make the intelligent inspection main pushing direction. In the context of large image data, massive image data needs to be compressed and then transmitted. Meanwhile, with the progress of image processing technology, the unmanned aerial vehicle end can realize a small amount of operation and analyze image characteristics.
At present, the insulator fault is a main part of the power transmission line fault, and a plurality of methods for realizing insulator fault identification by utilizing a deep learning network, such as a method of combining LBP-HF characteristics with an SVM classifier, a CNN classifier, a YOLO V3, a Faster RCNN and the like, are adopted, and because of high identification accuracy, the VGG16 network and the Faster RCNN network are selected to realize fault detection. Conventional image compression methods such as JPEG, JPEG2000, BPG, etc. aim to minimize visual distortion of the human eye. Because of the limitation of the bandwidth of the inspection environment, the image is required to be transmitted only at a lower code rate, but the traditional image compression method can cause obvious blocking effect and image artifact, and the identification accuracy of the encoded image to the small electric components such as the insulator string is obviously reduced when the encoded image is sent into the fault detection model.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide an image quantization method for insulator defect identification, which aims to reduce the influence of compression operation on the fault identification accuracy of small parts such as insulators and the like in the image transmission process and realize that the fault identification accuracy of a restored image is kept at a higher level under a low code rate.
The aim of the invention can be achieved by the following technical scheme:
an image quantization method facing insulator defect identification, which comprises the following steps:
step 1: selecting an insulator and a fault image thereof shot by inspection of the unmanned aerial vehicle, and training a neural network by utilizing image data to obtain a trained fault identification model;
step 2: extracting image features aiming at the trained fault recognition model;
step 3: further extracting key region side features in the low-layer edge information for the image features;
step 4: the pooling operation is reversely pushed by utilizing pooling characteristics of the characteristic extraction network aiming at the characteristics of the key area side in the low-layer edge information;
step 5: aiming at the characteristics of the key region side in the low-layer edge information, performing reverse-pushing pooling operation by utilizing pooling characteristics of the characteristic extraction network, and performing reverse-pushing pooling operation by utilizing convolution characteristics of the characteristic extraction network to obtain a significance mark graph corresponding to the length and width dimensions of the original image;
step 6: and obtaining an image quantization final result after coarse quantization or fine quantization based on the saliency mark graph.
Further, the step 1 specifically includes: selecting an insulator and fault images thereof shot by unmanned aerial vehicle inspection, dividing all image data into a training set and a verification set, and training a Faster RCNN network in a server by using the training set and the verification set to obtain a trained fault recognition model.
Further, the step 2 specifically includes: and extracting the weight and bias of the VGG16 network in the trained fault identification model, carrying out convolution operation on the insulator and the fault image thereof, which are shot by unmanned aerial vehicle inspection, and extracting image characteristics.
Further, the step 3 comprises the following sub-steps:
step 301: pixels at the same position in each channel of the image feature are added to obtain a feature map F which is consistent with the length and width of the image feature and has 1 layer of channels 1 ；
Step 302: calculate the whole feature map F 1 Average value F of (2) avg Map F of the features 1 The middle pixel exceeds the average value F avg The position pixel value of (1) is set to 0, and the rest positions are set to 0, thereby obtaining a feature map F 1 Matrix F with same length and width dimensions 2 Namely, the key region side characteristics in the corresponding low-layer edge information.
Further, the step 4 specifically includes: extracting the characteristic that the network pooling layers are the maximum pooling of stride=2 according to the characteristics in the fault recognition model, and carrying out matrix F corresponding to the characteristics of the key region side in the low-layer edge information 2 The elements in the matrix are assigned the same matrix B n Sequentially push out B n Matrix B of one layer on the corresponding pooling layer n-1 。
Further, the step 5 specifically includes: extracting net according to characteristics in fault recognition modelThe convolution kernels of the complex are all 3×3, and stride=1, and the filling mode is SAME characteristic, based on matrix B n-1 Performing convolution back-pushing process, and calculating to obtain matrix B n-2 I.e. the saliency flag map corresponding to the length-width dimension of the original picture.
Further, the step 6 includes the following sub-steps:
step 601: converting an original image shot by unmanned aerial vehicle inspection from an RGB space to a YUV space, filling the length and width dimensions to multiples of 8, and dividing the original image into a plurality of 8X8 blocks;
step 602: DCT transformation is carried out on the block, the transformed block is divided into a low-frequency component DC and a high-frequency component AC through zigzag scanning, and entropy coding is carried out, so that an original image which is finally processed and shot by unmanned aerial vehicle inspection is obtained;
step 603: and carrying out corresponding coarse quantization or fine quantization on the processed original image of the unmanned aerial vehicle inspection by utilizing the saliency mark graph, and finishing the final result of image quantization.
Further, the step 603 specifically includes:
if the saliency mark map corresponding to each divided 8×8 region block in the processed original image of unmanned aerial vehicle inspection shooting contains a mark 1, finely quantizing the block according to a brightness quantization table;
and if the saliency mark graphs corresponding to the 8 multiplied by 8 area blocks in the processed original image shot by the unmanned aerial vehicle inspection are all marked with 0, coarsely quantizing the blocks.
Further, the block is finely quantized according to the luminance quantization table, and the corresponding description formula is as follows:
in which Q 1 Representing the value of the 8x8 block after fine quantization, DCT Block The Block after DCT transformation is represented, and table0 is represented as a luminance quantization table.
Further, the block is coarsely quantized, and the corresponding description formula is as follows:
where table 1=q×e, Q is quantization step, E is 8×8 unit vector, Q 2 Is the value of the 8x8 block after coarse quantization.
Compared with the prior art, the invention has the following advantages:
(1) Compared with the prior art, after the quantized JPEG compression method is guided to be decoded by the saliency mark graph, the method can realize higher insulator fault identification accuracy under the same code rate compared with the original JPEG compression because the code rate of the background area is small and the code rate of the target area is large.
(2) Different from traditional image compression, the invention provides a quantization method for power inspection by combining the characteristics of machine vision and considering the problem of reduced recognition rate caused by the loss of image detail information of a traditional compression method under low bit rate in power inspection work. At the unmanned aerial vehicle end, feature extraction is realized, classified quantization is conducted according to the feature, key areas of feature extraction are finely quantized, rough quantization is realized in background areas, and recognition accuracy rate is higher than that of an original method under the same code rate.
(3) The invention relates to the problem that the fault recognition accuracy of a small electric component such as an insulator is obviously reduced under a low code rate in the traditional image compression in the data transmission process of the inspection operation of an unmanned aerial vehicle of a power grid, and aims to invent an image quantization method for insulator defect recognition, so that the fault recognition accuracy of a recovered image under the low code rate is still kept at a higher level.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a schematic diagram of a reverse thrust process of maximum pooling in the derivation of a saliency map in the method steps of the present invention;
FIG. 3 is a schematic diagram of a convolution back-thrust process in the derivation of a saliency map in a method step of the present invention;
fig. 4 is a schematic diagram of an image quantization method frame for power inspection.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
DETAILED DESCRIPTION OF EMBODIMENT (S) OF INVENTION
As shown in fig. 1, an image quantization method for insulator defect identification includes the following steps:
(S1) selecting an insulator shot by unmanned aerial vehicle inspection and a fault picture thereof, dividing an image data set into a training set and a verification set, realizing feature extraction by utilizing VGG16, and training a fast RCNN network in a server.
And (S2) extracting the weights and the offsets of the convolution operations of the first layers of the vgg network after training and adjustment by using the fault recognition model trained in the step (S1), and carrying out the convolution operation by using the extracted weights and the offsets at the unmanned aerial vehicle end to realize the extraction of the low-layer edge characteristics of the shot insulator picture and ensure that the low-layer edge characteristics extracted by the fault recognition network are identical with the characteristics extracted by the characteristic extraction operation carried out by the unmanned aerial vehicle end.
And (S3) processing the feature map extracted by the feature extraction network by using the step (S2) to extract the features of the key areas in the low-layer edge information.
Setting a picture edge information feature diagram extracted by a feature extraction network at the unmanned aerial vehicle end as F (i, j, k), adding pixels at the same position in each channel to obtain F with the same length and width dimensions as F and the number of channels being 1 layer 1 The calculation formula is as follows:
wherein F (i, j, k) represents pixel values of the characteristic diagram of the kth channel in the ith row and the jth column, h and w are the height and the width of the characteristic diagram F respectively, and N is the channel number of the characteristic diagram F.
Calculate the whole F 1 Average value F of (2) avg F is to F 1 Middle pixels exceeding F avg The positions of (2) are set to 1, and the remaining positions are set to 0, wherein: 1 represents a feature extraction region, 0 represents a background region, whereby a sum F can be obtained 1 Matrix F of the same length and width dimensions 2 。
(S4) the pooling layers in the feature extraction network are all the maximum pooling of stride=2, and F 2 (i, j) assigning a matrix B having the same number of rows and columns as it n Can push out B in turn n Layer B of one layer n-1 . When the feature extraction network implemented in the unmanned aerial vehicle is two convolution, convolution and pooling operations, the positions of all insulator strings are extracted without excessive low-level edge information, F is the feature output by the second pooling layer, only the positions of the convolution and pooling corresponding to the original image are considered, the numerical values are not considered, as shown in fig. 2, the method is a maximum pooling back-pushing process, B n-1 The calculation formula of (2) is as follows:
B n-1 (2i，2j)＝B n-1 (2i，2j+1)＝B n-1 (2i+1，2j)＝B n-1 (2i+1，2j+1)＝B n (i，j)，i＝0，1，…，h-1；j＝0，1，，1，…，w-1
(S5) As shown in FIG. 3, B n-1 Reversely push out the upper layer B n-2 The convolution back-pushing process is needed, and is shown as the following step B n-1 Reversely push out the upper layer B n-2 The convolution back-pushing process is needed, and the length and width dimensions of the feature map after convolution are unchanged because the convolution kernels of the feature extraction network are 3x3, stride=1 and the filling mode is SAME. When the flag in one 3x3 region is 0, it represents that the corresponding position of the previous layer operation is a background (non-important) region, and when the flag in the region is 1, it cannot represent that all positions of the previous layer operation are important regions, so that B is set first n-2 Is a full 1 matrix with the height and the width of 2h and 2w respectively, when B n-1 When the flag of (u, v) is 0, B n-2 The calculation formula of (2) is as follows:
B n-2 (u-1，v-1)＝B n-2 (u-1，v1)＝B n-2 (u-1，v+1)＝0
B n-2 (u，v-1)＝B n-2 (u，v1)＝B n-2 (u，v+1)＝0
B n-2 (u+1，v-1)＝B n-2 (u+1，v1)＝B n-2 (u+1，v+1)＝0
u＝0，1，…，2h-1；v＝0，1，，1，…，2w-1
and finally pushing out a significance mark graph corresponding to the length and width dimensions of the original picture according to the feature extraction network, and judging whether the region is subjected to coarse quantization or fine quantization according to the position of the interval where 0 and 1 are located.
(S6) conducting quantization on the insulator image according to the saliency mark graph, conducting fine quantization and coarse quantization on the target and background areas according to different marks in the saliency mark graph, converting an original image into YUV space from RGB space, filling the length and width dimensions to multiples of 8, dividing the original image into a plurality of 8x8 blocks, conducting DCT conversion, conducting quantization according to the generated saliency mark graph, conducting zig-zag scanning, dividing the obtained product into a low-frequency component DC and a high-frequency component AC, and finally conducting entropy coding, wherein the basic structure is the same as that of JPEG compression. Wherein the region with the mark 1 in the saliency mark map represents the characteristic region extracted by convolution, the region with the mark 0 represents the background region, taking one 8x8 block in the brightness space Y as an example, if the corresponding saliency mark map in the divided 8x8 region blocks contains the mark 1, finely quantizing the block according to the brightness quantization table to generate Q 1 The calculation formula is as follows:
in which Q 1 Representing the value of the 8x8 block after fine quantization, DCT Block The Block after DCT transformation is represented, and table0 is represented as a luminance quantization table.
If all the region marks in the saliency mark map corresponding to the 8x8 block are 0, coarsely quantizing the saliency mark map to generate Q 2 The calculation formula is as follows:
where table 1=q×e, Q is quantization step, E is 8×8 unit vector, Q 2 Is the value of the 8x8 block after coarse quantization.
While the invention has been described with reference to certain preferred embodiments, it will be understood by those skilled in the art that various changes and substitutions of equivalents may be made and equivalents will be apparent to those skilled in the art without departing from the scope of the invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.
Claims (6)
1. An image quantization method for insulator defect identification is characterized by comprising the following steps:
step 1: selecting an insulator and a fault image thereof shot by inspection of the unmanned aerial vehicle, and training a neural network by utilizing image data to obtain a trained fault identification model;
step 2: extracting image features aiming at the trained fault recognition model;
step 3: further extracting key region side features in the low-layer edge information for the image features;
step 4: the pooling operation is reversely pushed by utilizing pooling characteristics of the characteristic extraction network aiming at the characteristics of the key area side in the low-layer edge information;
step 5: aiming at the characteristics of the key region side in the low-layer edge information, performing reverse-pushing pooling operation by utilizing pooling characteristics of the characteristic extraction network, and performing reverse-pushing pooling operation by utilizing convolution characteristics of the characteristic extraction network to obtain a significance mark graph corresponding to the length and width dimensions of the original image;
step 6: after coarse quantization or fine quantization is performed based on the saliency flag map, an image quantization final result is obtained,
wherein, the step 3 comprises the following sub-steps:
step 301: pixels at the same position in each channel of the image feature are added to obtain a feature map F which is consistent with the length and width of the image feature and has 1 layer of channels 1 ；
Step 302: calculate the whole feature map F 1 Average value F of (2) avg Map F of the features 1 The middle pixel exceeds the average value F avg The position pixel value of (1) is set to 0, and the rest positions are set to 0, thereby obtaining a feature map F 1 Matrix F with same length and width dimensions 2 I.e. the key area side characteristics in the corresponding low-level edge information,
the step 4 specifically includes: extracting the characteristic that the network pooling layers are the maximum pooling of stride=2 according to the characteristics in the fault recognition model, and carrying out matrix F corresponding to the characteristics of the key region side in the low-layer edge information 2 The elements in the matrix are assigned the same matrix B n Sequentially push out B n Matrix B of one layer on the corresponding pooling layer n-1 ，
The step 5 specifically includes: according to the characteristics that the convolution kernels of the feature extraction network in the fault identification model are 3×3, the stride=1, and the filling mode is SAME, the method is based on matrix B n-1 Performing convolution back-pushing process, and calculating to obtain matrix B n-2 I.e., a saliency flag map corresponding to the length-width dimension of the original picture,
the step 6 comprises the following sub-steps:
step 601: converting an original image shot by unmanned aerial vehicle inspection from an RGB space to a YUV space, filling the length and width dimensions to multiples of 8, and dividing the original image into a plurality of 8X8 blocks;
step 602: DCT transformation is carried out on the block, the transformed block is divided into a low-frequency component DC and a high-frequency component AC through zigzag scanning, and entropy coding is carried out, so that an original image which is finally processed and shot by unmanned aerial vehicle inspection is obtained;
step 603: and carrying out corresponding coarse quantization or fine quantization on the processed original image of the unmanned aerial vehicle inspection by utilizing the saliency mark graph, and finishing the final result of image quantization.
2. The image quantization method for insulator defect identification according to claim 1, wherein the step 1 specifically comprises: selecting an insulator and fault images thereof shot by unmanned aerial vehicle inspection, dividing all image data into a training set and a verification set, and training a Faster RCNN network in a server by using the training set and the verification set to obtain a trained fault recognition model.
3. The image quantization method for insulator defect identification according to claim 1, wherein the step 2 specifically comprises: and extracting the weight and bias of the VGG16 network in the trained fault identification model, carrying out convolution operation on the insulator and the fault image thereof, which are shot by unmanned aerial vehicle inspection, and extracting image characteristics.
4. The image quantization method for insulator defect identification according to claim 1, wherein the step 603 specifically includes:
if the saliency mark map corresponding to each divided 8×8 region block in the processed original image of unmanned aerial vehicle inspection shooting contains a mark 1, finely quantizing the block according to a brightness quantization table;
and if the saliency mark graphs corresponding to the 8 multiplied by 8 area blocks in the processed original image shot by the unmanned aerial vehicle inspection are all marked with 0, coarsely quantizing the blocks.
5. The method for image quantization for insulator defect identification according to claim 4, wherein the fine quantization of the block according to the brightness quantization table corresponds to the description formula:
in which Q 1 Representing the value of the 8x8 block after fine quantization, DCT Block The Block after DCT transformation is represented, and table0 is represented as a luminance quantization table.
6. The method for image quantization for insulator defect identification according to claim 4, wherein the block is coarsely quantized according to the following description formula:
where table 1=q×e, Q is quantization step, E is 8×8 unit vector, Q 2 For the value of the 8x8 block after coarse quantization, DCT Block Representing the DCT transformed Block.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011271571.0A CN112464756B (en) | 2020-11-13 | 2020-11-13 | Insulator defect identification-oriented image quantization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011271571.0A CN112464756B (en) | 2020-11-13 | 2020-11-13 | Insulator defect identification-oriented image quantization method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112464756A CN112464756A (en) | 2021-03-09 |
| CN112464756B true CN112464756B (en) | 2023-05-02 |
Family
ID=74837293
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011271571.0A Active CN112464756B (en) | 2020-11-13 | 2020-11-13 | Insulator defect identification-oriented image quantization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112464756B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105976383A (en) * | 2016-05-16 | 2016-09-28 | 国网河南省电力公司电力科学研究院 | Power transmission equipment fault diagnosis method based on limit learning machine image recognition |
| CN106686387A (en) * | 2016-12-30 | 2017-05-17 | 杭州占峰科技有限公司 | Picture compression method for photographing surface of odometer style water meter |
| CN106780438A (en) * | 2016-11-11 | 2017-05-31 | 广东电网有限责任公司清远供电局 | Defects of insulator detection method and system based on image procossing |
| CN107748895A (en) * | 2017-10-29 | 2018-03-02 | 北京工业大学 | UAV Landing landforms image classification method based on DCT CNN models |
| CN108596886A (en) * | 2018-04-17 | 2018-09-28 | 福州大学 | Aerial Images insulator based on deep learning falls piece fault rapid detecting method |
| CN111914720A (en) * | 2020-07-27 | 2020-11-10 | 长江大学 | Method and device for identifying insulator burst of power transmission line |
-
2020
- 2020-11-13 CN CN202011271571.0A patent/CN112464756B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105976383A (en) * | 2016-05-16 | 2016-09-28 | 国网河南省电力公司电力科学研究院 | Power transmission equipment fault diagnosis method based on limit learning machine image recognition |
| CN106780438A (en) * | 2016-11-11 | 2017-05-31 | 广东电网有限责任公司清远供电局 | Defects of insulator detection method and system based on image procossing |
| CN106686387A (en) * | 2016-12-30 | 2017-05-17 | 杭州占峰科技有限公司 | Picture compression method for photographing surface of odometer style water meter |
| CN107748895A (en) * | 2017-10-29 | 2018-03-02 | 北京工业大学 | UAV Landing landforms image classification method based on DCT CNN models |
| CN108596886A (en) * | 2018-04-17 | 2018-09-28 | 福州大学 | Aerial Images insulator based on deep learning falls piece fault rapid detecting method |
| CN111914720A (en) * | 2020-07-27 | 2020-11-10 | 长江大学 | Method and device for identifying insulator burst of power transmission line |
Non-Patent Citations (2)
| Title |
|---|
| Automatic Metallic Surface Defect Detection and Recognition with Convolutional Neural Networks;Xian Tao,et al.;《Appl. Sci.》;20180906;全文 * |
| 基于航拍影像的输电线路绝缘子识别及缺陷检测;危双丰等;《工程勘察》;20200430(第04期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112464756A (en) | 2021-03-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113159173A (en) | Convolutional neural network model compression method combining pruning and knowledge distillation | |
| CN109949217B (en) | Video super-resolution reconstruction method based on residual learning and implicit motion compensation | |
| CN107463932B (en) | Method for extracting picture features by using binary bottleneck neural network | |
| CN110162986B (en) | Reversible information hiding method based on adjacent pixel prediction model | |
| CN110751597A (en) | Video super-resolution method based on coding damage repair | |
| CN109982071B (en) | HEVC (high efficiency video coding) dual-compression video detection method based on space-time complexity measurement and local prediction residual distribution | |
| CN115131675A (en) | Remote sensing image compression method and system based on reference image texture migration | |
| CN112926533A (en) | Optical remote sensing image ground feature classification method and system based on bidirectional feature fusion | |
| CN112738533A (en) | Machine patrol image regional compression method | |
| CN105279742B (en) | A kind of image de-noising method quickly based on piecemeal estimation of noise energy | |
| CN108717690A (en) | A kind of synthetic method of high dynamic range photo | |
| Miyata et al. | Novel inverse colorization for image compression | |
| CN106331730A (en) | Double-compression detection method by using quantification factor same as H.264 video | |
| CN112464756B (en) | Insulator defect identification-oriented image quantization method | |
| CN112911298B (en) | HEVC video dual-compression detection method and system based on boundary effect | |
| CN116152571A (en) | Kitchen waste identification and classification method based on deep learning | |
| CN113298102B (en) | Training method and device for target classification model | |
| Jiajia et al. | Minimum structural similarity distortion for reversible data hiding | |
| CN113822801A (en) | Compressed video super-resolution reconstruction method based on multi-branch convolutional neural network | |
| Zhang et al. | An Improved Quantization Algorithm for Electric Power Inspection | |
| Parameshachari et al. | Singular Value Decomposition (SVD) Based Optimal Image Compression Technique | |
| Alshami et al. | Enhancing quality of lossy compressed images using minimum decreasing technique | |
| CN112508847A (en) | Image quality evaluation method based on depth feature and structure weighted LBP feature | |
| CN111432208A (en) | Method for determining intra-frame prediction mode by using neural network | |
| CN111107360A (en) | Spectrum-space dimension combined hyperspectral image lossless compression method and system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |