CN112365468A - AA-gate-Unet-based offshore wind power tower coating defect detection method - Google Patents

AA-gate-Unet-based offshore wind power tower coating defect detection method Download PDF

Info

Publication number
CN112365468A
CN112365468A CN202011253124.2A CN202011253124A CN112365468A CN 112365468 A CN112365468 A CN 112365468A CN 202011253124 A CN202011253124 A CN 202011253124A CN 112365468 A CN112365468 A CN 112365468A
Authority
CN
China
Prior art keywords
coating
offshore wind
wind power
power tower
tower
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
Application number
CN202011253124.2A
Other languages
Chinese (zh)
Inventor
张堃
郑山建
吴樱樱
涂鑫涛
徐沛霞
刘志诚
冯文宇
黄宇煦
韩宇
朱远璠
张宇豪
祁晖
陆贝洋
沈桠楠
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou Jintian Technology Industry Development Co ltd
Nantong University
Original Assignee
Suzhou Jintian Technology Industry Development Co ltd
Nantong University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Suzhou Jintian Technology Industry Development Co ltd, Nantong University filed Critical Suzhou Jintian Technology Industry Development Co ltd
Priority to CN202011253124.2A priority Critical patent/CN112365468A/en
Publication of CN112365468A publication Critical patent/CN112365468A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion 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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/143—Segmentation; Edge detection involving probabilistic approaches, e.g. Markov random field [MRF] modelling
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8854—Grading and classifying of flaws
    • G01N2021/8874—Taking dimensions of defect into account
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8883—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges involving the calculation of gauges, generating models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20092—Interactive image processing based on input by user
    • G06T2207/20104—Interactive definition of region of interest [ROI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30232—Surveillance

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • Quality & Reliability (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Signal Processing (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Wind Motors (AREA)

Abstract

The invention discloses an AA-gate-Unet-based method for detecting coating defects of an offshore wind power tower, which comprises the following steps: receiving a defect detection request of an offshore wind power tower coating transmitted by a system, and acquiring unmanned aerial vehicle detection video information of the wind power tower coating corresponding to the request; extracting a tower barrel coating image key frame in the video information; uploading the image of the key frame of the tower barrel coating image to a system cloud; predicting the image of the key frame of the offshore wind power tower barrel coating detection image uploaded to the cloud of the system by adopting a prediction algorithm based on HMM and based on Winters three-parameter exponential smoothing to obtain a detection result of the offshore wind power tower barrel coating defect; and when the detection result of the coating of the offshore wind power tower cylinder is in an abnormal state, uploading the image of the defect coating of the offshore wind power tower cylinder to a system, and carrying out corresponding maintenance work of the coating of the offshore wind power tower cylinder. The method comprises the steps of collecting images of a tower drum of the offshore wind turbine by using an unmanned aerial vehicle platform, and segmenting defects of the tower drum by using an improved depth learning model AA-gate-Unet; after accurate segmentation, area value data of various defects can be calculated, so that accurate budget and field detection of fan later maintenance are realized.

Description

AA-gate-Unet-based offshore wind power tower coating defect detection method
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to an AA-gate-Unet-based method for detecting coating defects of an offshore wind power tower.
Background
Wind energy is increasingly gaining attention as a clean renewable energy source in all countries of the world. It is expected that clean renewable energy, represented by wind energy, will also have a more significant and positive role in dealing with climate change and promoting energy transformation. As a clean power generation mode of renewable energy, wind energy has important significance for low carbon, energy conservation and atmospheric environment protection.
Compared with land wind power generation, offshore wind power generation has outstanding characteristics, and offshore wind power is always paid attention by wind power developers due to the advantages of abundant resources, stable wind speed, fewer development interest relevant parties, no land competition with other development projects, large-scale development and the like. From the experience of european predecessors, the offshore wind power industry has experienced over 20 years of development, gradually evolving from a regional energy industry to a global one.
China has abundant offshore wind energy resources and has resource conditions for large-scale development of offshore wind power. According to the detailed investigation of the primary achievement of wind energy resources of the China meteorological office, the installed capacity of wind power in 50 meters above sea level in offshore areas within a water depth line of 5-25 meters in China is about 2 hundred million kilowatts. The water depth is 5-50 m, and the capacity of developed offshore wind at the height of 70m is up to 5 hundred million kW. The scale of wind energy development is increasing, and the daily monitoring demand of wind turbines working in outdoor complex environments is also increasing. The working intensity and the working time of the manual inspection are increased to become an emergency method. And the marine environment is complicated, the weather is moist, and wind power tower cylinder surface coating compares in the wind turbine generator system in other areas the damage condition more easily appears, and these all make the manual work to patrol and examine the work degree of difficulty increase and actually patrol and examine efficiency reduction, and wind turbine generator system's operating condition hardly obtains timely comprehensive feedback. And unit detection mechanism lacks the fairness with the monitoring standard, consequently needs a technical scheme in order to be used for replacing artifical the inspection and provide the check-up for data drive's unit monitoring.
There are many information that can excavate in the unmanned aerial vehicle vision acquires the data, acquires wind turbine generator system's running state from it, patrols and examines the process with the wind farm and becomes more automatic, intelligent, thereby replaces the manual work to patrol and examine better and greatly promotes the efficiency of patrolling and examining, also provides the information check-up that comes from the vision dimension for wind farm data monitoring center, has increased the reliability to wind turbine generator system state monitoring.
Compared with the problem of small storage capacity of the traditional computing technology, the cloud computing technology can be used for infinite storage, the cost of a computer and software can be reduced, and meanwhile, the data reliability is improved, and the compatibility of a document format is improved.
Disclosure of Invention
The invention mainly aims to provide an AA-gate-Unet-based method and device for detecting coating defects of an offshore wind power tower, and aims to solve the technical problems of high cost, low efficiency, poor flexibility and the like of the method for detecting the coating defects of the offshore wind power tower in the prior art.
In order to achieve the above object, the present application provides a method for detecting defects of a coating of an offshore wind turbine tower, comprising the following steps:
s1, receiving an offshore wind power tower coating defect detection request transmitted by a system, and acquiring video information of unmanned detection of a wind power tower coating corresponding to the offshore wind power tower coating defect detection request;
s2, extracting a tower barrel coating image key frame in the video information of the unmanned detection of the wind power tower barrel coating through a deep learning neural network model AA-gate-Unet;
s3, uploading the images of the key frames of the tower barrel coating images to a system cloud;
s4, predicting the image of the tower drum coating image key frame uploaded to the cloud of the system by adopting a prediction algorithm based on HMM and based on Winters three-parameter exponential smoothing to obtain a detection result of the offshore wind power tower drum coating defect;
s5, dividing the detection result of the coating defect of the offshore wind power tower into two states: a normal state and an abnormal state; according to the processing result of the key frame image detected by the coating of the offshore wind power tower cylinder, which is identified by the camera of the unmanned aerial vehicle, different abnormal states are specifically divided into: state 1: defining the condition that the coating of the offshore wind power tower barrel cracks; state 2: the method is defined as the peeling condition of the coating of the offshore wind power tower; state 3: defining the corrosion condition of the coating of the offshore wind power tower;
and S6, when the detection result of the coating defect of the offshore wind power tower is in an abnormal state, uploading the corresponding coating image of the coating defect of the offshore wind power tower to a system, and carrying out corresponding maintenance work of the coating of the offshore wind power tower.
Further, the extraction process in step S2 specifically includes:
(1) extracting a model: a bilateral filter is adopted, and the mathematical model is as follows:
spatial distance function:
Figure BDA0002772239780000031
gray scale change function:
Figure BDA0002772239780000032
wherein (x)i,yi) Represents the coordinates of the current position point, (x)c,yc) Representing the filter center position, gray (x)i,yi) Gray value, gray (x) representing the current coordinate pointi,yi) A gray value representing a center coordinate point of the filter;
(2) adaptive threshold: the T value is changed according to the specific characteristics of the image, and the specific process is represented by the following formula:
Figure BDA0002772239780000033
wherein g (x, y) represents a pixel value located at (x, y) in the image;
(3) and detecting the bilateral of the fan outline in the established image data set by adopting a Hough line detection algorithm, and extracting the region of interest.
Further, the deep learning neural network model AA-gate-Unet described in step S2 has a U-shaped structure, and the feature map is cut and transmitted between the up-sampling and the down-sampling.
Further, the deep learning neural network model AA-gate-Unet described in step S2 includes a left side systolic pathway and a right side diffuse pathway; the left contraction path comprises a plurality of convolution layers and a pooling layer, a plurality of convolution kernels are repeatedly used in the contraction path, and the nonlinear features can be extracted by taking the ReLU as an activation function after convolution.
Further, the AA module in the deep learning neural network model AA-gate-Unet described in step S2 includes:
(1) by inputting the value of (w, h, c)in) The signature of (a) performs a 1 × 1 convolution of the output QKV matrix, which has a size of (w, h,2 × d)k+dv) Wherein w, h,2 x dk+dVThe width, the length and the depth of the matrix are respectively represented;
(2) and then, the QKV matrix is segmented from the depth channel to obtain three matrixes of Query, Key and Value, wherein the depth channel sizes of the three matrixes are dk、dk、dv；
(3) A multi-head attention mechanism structure is adopted, and the Query matrix, the Key matrix and the Value matrix are respectively divided into N equal matrixes from a depth channel for subsequent calculation, so that the model learns characteristic information in different subspaces;
(4) the weight matrix is calculated by performing matrix multiplication by using two matrixes of Flat _ Q, Flat _ K, and relative position embedding calculation is added on the weight matrix, so that the relative position information of each point on the characteristic diagram is obtained by performing weight calculation on the Q matrix in the length and width directions, and the transformation of the characteristic position is prevented, and the final effect of the model is reduced.
Further, the gate module in the deep learning neural network model AA-gate-Unet described in step S2 is to add an attribute enhanced convolution module in the down-sampling of the neural network model Unet, and add attentionate in the sampling path on the right layer of the Unet to weight the fusion features.
Further, the defect classification category is predicted in advance by using the hidden markov chain in step S4.
Further, in step S4, the defect classification category is used as a to-be-predicted value, and is set as a hidden chain of the HMM, and the GPS height difference and the change in the yaw angle of the north indicator are collected by continuous calculation, and are recorded as an observation chain of the HMM, and the observation data of the image is set.
Further, in step S4, the CHMM is selected for the prediction of the segmentation class.
Further, after the step S4, the method further includes:
when the segmentation class sequence of the next picture is predicted, the class in the activation function softmax in the last 1 × 1 convolution layer of the network is updated in real time by the parameter, and then:
Figure BDA0002772239780000051
wherein i represents the class of the segmentation index, C represents the total class number, Si represents the ratio of the index of the current element to the indexes of all elements, Vi represents the output of a preceding-stage output unit of the model, and the C value is input in real time through the prediction of the segmentation class.
The invention also discloses an AA-gate-Unet-based device for detecting the coating defects of the offshore wind power tower, which comprises the following components:
the request receiving module is used for receiving an offshore wind power tower coating defect detection request and acquiring the unmanned aerial vehicle detection video information of the wind power tower coating corresponding to the offshore wind power tower coating defect detection request;
the video extraction module is used for extracting a coating defect image key frame in the coating defect monitoring video information of the offshore wind power tower;
the coating detection module is used for obtaining a defect detection result of the coating of the offshore wind power tower cylinder according to a processing result of the defect coating key frame image detection;
and the tower tube identification module is used for extracting the coating defect picture information of the offshore wind power tower tube and determining the position information of the offshore wind power tower through the identified tower tube information.
Has the advantages that: the invention has the following beneficial effects:
(1) the surface images of the fan tower drums are acquired by using an unmanned aerial vehicle, the types of surface defects contained in each fan tower drum are determined by an image analysis means, and the areas of the defects are automatically counted, so that the image acquisition efficiency is improved;
(2) aiming at the environmental characteristics of an offshore wind farm, analyzing the main structure and flight control characteristics of the unmanned aerial vehicle, and specially making a set of fan tower barrel video acquisition scheme for a wind tower barrel; aiming at a video acquired by an unmanned aerial vehicle, screening a video frame acquired by the unmanned aerial vehicle through fan tower barrel modeling, designing a preprocessing algorithm aiming at imaging characteristics in an image, and further extracting a tower barrel interested region in the image;
(3) carefully analyzing the complex imaging characteristics in the image, providing a novel AA-gate-Unet model for image segmentation, segmenting different types of defects, and designing a more reasonable loss function aiming at the image characteristics in the topic in the novel network model;
(4) and aiming at the segmentation result, combining with hardware parameters of the unmanned aerial vehicle when the picture is acquired, predicting a segmentation result sequence by using a Markov chain theory, and changing inherent parameters in the model according to the predicted segmentation result sequence to improve the accuracy.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a schematic diagram of a neural network AA-gate-Unet structure provided in an embodiment of the present invention;
FIG. 2 is a schematic diagram of an AA architecture provided by an embodiment of the present invention;
FIG. 3 is a schematic diagram of a gate structure provided in an embodiment of the present invention;
FIG. 4 is a schematic diagram of an HMM prediction algorithm provided by an embodiment of the present invention;
FIG. 5 is a state transition diagram of a hidden Markov chain according to an embodiment of the present invention;
fig. 6 is a diagram of a hidden markov prediction probability structure according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.
At present, the inspection of the fan is generally carried out by manual aerial photography, the defect types are identified by naked eyes, and the manual mode is obviously not in accordance with the reality under the condition that the scale of the offshore fan is larger and larger. The deep learning is a product brought by the continuous improvement of computer hardware computing power in recent years, and the robustness of the filter stick performance of the algorithm is far superior to that of the traditional algorithm. The realization of accurate automatic processing of images by using models in deep learning is a necessary trend of current technical development. The invention is based on a deep learning model, applies the image segmentation model to the surface defect detection of the sea surface fan tower, replaces the existing manual detection method, provides a set of accurate, efficient and stable fan tower surface defect detection algorithm, and verifies that the algorithm meets the detection requirements of actual production in practical application.
The invention discloses an AA-gate-Unet-based offshore wind power tower coating defect detection method, which comprises the following steps:
s1, video acquisition is carried out on the offshore wind power tower cylinder through the unmanned aerial vehicle, a single fan tower cylinder picture database is established by screening video frames, and a fan tower cylinder region is extracted, and the method specifically comprises the following steps:
the unmanned aerial vehicle flight control strategy is researched, and a set of complete fan tower barrel video capable of being collected is designed. And mathematical modeling is carried out on the tower, database index information is constructed by utilizing sensor information such as a GPS (global positioning system) and a north arrow on the unmanned aerial vehicle, video frames are screened to establish a single-seat fan tower tube image database, and the problem of data redundancy is solved. And a set of preprocessing algorithm is designed according to the image imaging characteristics, so that the area of the tower drum of the fan can be accurately extracted. And establishing a key-value data structure of a dictionary-like system through the real-time height of the GPS sensor and the deflection angle of the screw instrument, and screening the video frames in the shooting repeated area through a method of corresponding the real-time height and the real-time angle to the real-time video frames.
S2, detecting and extracting the defects of the fan tower barrel area image through a deep learning neural network model AA-gate-Unet, which specifically comprises the following steps:
(1) extracting a model: a bilateral filter. The bilateral filter is more excellent in retention of edge information in an image than a gaussian filter. The mathematical principle of the bilateral filter is mainly to use a Gaussian function to represent the gray scale change relationship and to multiply the distance of a Gaussian function representing the spatial relationship. The main mathematical model is as follows:
spatial distance function:
Figure BDA0002772239780000081
gray scale change function:
Figure BDA0002772239780000082
wherein (x)i,yi) Represents the coordinates of the current position point, (x)c,yc) Representing the filter center position, gray (x)i,yi) Gray value, gray (x) representing the current coordinate pointi,yi) Representing the grey value of the central coordinate point of the filter.
(2) Adaptive threshold: that is, the empirical value is not fixed, and the value T is changed according to the specific characteristics of the image. There are many ways to generate T, which can be determined by the form of the gradation, gradient, morphology, etc. of the pixels in the local region, and T (x, y) can be used. The adaptive threshold may be represented by the following equation:
Figure BDA0002772239780000083
where g (x, y) represents a pixel value located at (x, y) in the image.
The flight incidence angle of attack constantly changes along with the fuselage gesture when unmanned aerial vehicle gathers the video, and the relative position with the sun also changes when its camera shoots always, so the distribution of natural light has very big difference in the image at different moments. Compared with threshold binarization, the adaptive binarization selects the region threshold according to the pixel value characteristics of the block region, and has robustness for the pictures with larger difference. Here, when the area is a rectangle with a size of 100 × 100, the best effect is obtained by adaptive binarization using the area pixel gray level mean value.
(3) The method adopts a Hough line detection algorithm to detect the bilateral of the fan contour in the established image data set, and extracts the region of interest.
The Hough line detection principle is as follows: in a rectangular coordinate system, two variables are generally used to represent a straight line in a two-dimensional space, and the equation is:
y＝kx+b
the value of k and b can be from negative infinity to positive infinity, the accurate value range is difficult to express in a computer through a program language, and the straight line in the two-dimensional space is more difficult to realize through an exhaustive mode. In Hough line detection, a two-dimensional rectangular coordinate system is converted into a two-dimensional polar coordinate, and a straight line can be represented by an equation:
r＝x cosθ+y sinθ
where r is the perpendicular distance of the line to the pole, and x, y are the intersection points of the lines on the x-axis and y-axis, respectively. Passing a fixed point (x) under the coordinate system0,y0) The straight line of (d) can be expressed as:
rθ＝x0 cosθ+y0 sinθ
from this equation, it can be inferred that each pair (r)θAnd theta) can represent a pass point(x0,y0) Is measured. For a certain point in polar coordinates, all straight lines passing through the vertex can be expressed as about (r)θEquation r of θ)θ＝x0cosθ+y0sin θ. Analyzing the middle points of all polar coordinates in Hough line detection, and determining the parameter pair (r) represented by the intersection point when the curve passing through the same point reaches a certain threshold valueθAnd θ) is a straight line in the source image. And determining the coordinates of the edge of the tower of the wind turbine through Hough line detection. And selecting boundary points of the profile of the tower drum of the fan to intercept the ROI.
In this embodiment, the deep learning neural network model AA-gate-Unet specifically includes:
the whole Unet structure is shown in figure 1 and is in a U-shaped structure, and the cutting and transmission of the feature map are established between the up-sampling and the down-sampling, so that more original information can be learned between different layers. The U-net uses all convolutions in the VAiled mode to guarantee the segmentation result. Structurally, U-net consists of a left contraction path and a right diffusion path. The left contraction path is formed by adding a plurality of convolution layers and pooling layers, so that the space dimension of the features can be effectively reduced, and the number of the features can be increased. The contraction path repeatedly uses a plurality of convolution kernels, ReLU is adopted as an activation function after convolution to guarantee that nonlinear features can be extracted, and meanwhile a plurality of Max boosting operations are used to enable training parameters to be reduced remarkably. The expansion path is an up-sampling process, as opposed to the contraction path, implemented using deconvolution. And a cropping and copying channel is used between the contraction path and the expansion path, and the low-level feature map cropping of the contraction path is copied into the expansion path and the corresponding high-level feature map fusion.
AA module (attentionagment): the concrete structure of the Attention authority is shown in FIG. 2, which essentially obtains a series of key-value pair mappings through queries.
First, the input size is (w, h, c)in) The signature of (a) performs a 1 × 1 convolution of the output QKV matrix, which has a size of (w, h,2 × d)k+dv) Wherein w, h,2 x dk+dVThe width, length and depth of the matrix are indicated, respectively. And the QKV matrix is segmented from the depth channel to obtain Q: (The depth channel sizes of the three matrixes of Query, K (Key) and V (value) are dk、dk、dv. Next, a Multi-Head Attention mechanism (Multi-Head Attention) structure is adopted, Q, K, V three matrices are respectively divided into N equal matrices from the depth channel for subsequent calculation (set to 2 in this embodiment), and this Multi-Head Attention mechanism (Multi-Head Attention) expands the original single Attention calculation into a plurality of calculations which are smaller and independent in parallel, so that the model can learn feature information in different subspaces.
The flattened matrix of the divided Q, K, V generates three matrixes of Flat _ Q, Flat _ K, Flat _ V (namely Q, K, V keeps the depth channel unchanged and compresses the depth channel to 1 dimension from the length direction and the width direction), wherein the sizes of the first two matrixes are (w × h, d)k) The latter matrix size is (w x h, d)v). Next, the attribute evaluation stores the original Self-attribute, and performs matrix multiplication using two matrices of Flat _ Q, Flat _ K to calculate a weight matrix, and adds calculation of Relative position embedding (Relative position embedding) to the weight matrix, and performs weight calculation of length and width directions on the Q matrix to obtain Relative position information of each point on the feature map, thereby preventing the final effect of the model from being reduced due to the transformation of the feature position. The related position information in the length direction and the width direction is obtained by inner products of the Q matrix and the weight matrixes H and W respectively and is recorded as ShAnd SwAnd the weight matrixes H and W are obtained by training the model and have the size of (wh, wh, 1). Then, the resulting three matrices are added and multiplied by a scaling factor
Figure BDA0002772239780000111
To prevent the calculation result from being too large.
And then processing by using a softmax function to obtain a final characteristic weight matrix. And finally, multiplying the weight matrix by the V matrix, and performing 1 × 1 convolution operation on the result reshape to the original length and width to obtain a final attention feature matrix O. The Attention feature matrix O and the normal convolution process are spliced (concat) according to the depth direction, and then the result of the Attention augmentation can be obtained. The formula for the calculation of the attention characterization matrix O is as follows:
Figure BDA0002772239780000112
gate attention module: in addition to adding an attention enhanced convolution module to the downsampling, adding attentionate to the Unet right layer upsampling path weights the fused features. The structure is shown in fig. 3. x is the number oflG is a feature in the next layer in the upsampling path, as part of the feature that the downsampling path of the same layer copies and cuts into the upsampling path. g, xlRespectively making their characteristic dimensions be changed from original F by means of convolution operationg,FlConversion to F of the same dimensionint. G, x of the same dimensionlAddition for ReLU activation and the generated data were calculated using a 1 × 1 convolution.
Activation using the sigmoid function then guarantees its non-linearity to generate a matrix representing the pixel weight magnitude. Resampler herein represents a resampling operation for which the calculated data is associated with xlEnsuring the dimension to be consistent. Extracting the weight matrix of AAconv, wherein the dimension of the weight matrix MHA (X) in AAconv is dv because xlIs a feature clipped into the upsampling in the downsampling in the copyandcrop path, and the dimension size of the feature should be FxWe change its dimension to F by convolving itx-dvSplicing with MHA (X) matrix, the dimension of the spliced matrix just satisfies the original characteristic dimension F in the middle transmission channelx. By convolution, concatenation operation, not only in dimension with xlFeature fusion can be performed by keeping consistency, weight is generated by using lower layer feature information, meanwhile, weight information in fusion features is further strengthened by using a weight matrix generated in the step of enhancing convolution, and a region with larger attention weight can be highlighted.
S3, uploading the images of the key frames of the tower barrel coating images to a system cloud; the images of the key frames of the tower drum coating images are uploaded to a cloud computing of the system, so that computing cost and time can be greatly reduced, system performance is improved, and reliability of the system is enhanced.
And S4, predicting the extracted defect image by adopting a prediction algorithm based on an HMM (hidden Markov model), so as to obtain a defect type corresponding to each defect, and thus completing the identification of the coating defect type of the wind power tower.
In this embodiment, the collected sea surface fan videos are subjected to frame screening, so as to complete a single-base fan tower tube picture set. According to the fan video frame attribute information, a sea surface fan tower tube picture set with information is established. The invention uses the GPS height of the position where the unmanned aerial vehicle is positioned when shooting, the deflection angle of the north indicator, the segmented defect area and the image depth as the picture set information.
The method is used for classifying the problems of the paint surface of the tower drum of the fan, and has four problems of seal ring aging failure, mechanical flaw, corrosion, mechanical damage, oil leakage and the like. The four problems have certain probability at the position on the tower drum of the fan, the problem type in the next picture is predicted according to the problem type and the generated height of the tower drum, and the reverse inspection of the problem part segmentation of the next picture is realized.
In order to perform reverse checking on the segmentation result, the hidden Markov chain is adopted to perform advance prediction on four classification categories. Since the classification type is used as a to-be-predicted value, the classification type is set as a hidden chain of an HMM, and the height difference of gps and the deflection angle change of the north-pointing instrument can be collected by continuous calculation, and are recorded as an observation chain of the HMM, and the prediction flow is shown in fig. 4. At this time, the observation data of the image needs to be set, and it is proposed herein to set a difference score function to generate the observation data, as shown in the following formula:
Figure BDA0002772239780000131
wherein h isi-1Representing the height, alpha, of the i-1 th image in the data seti-1Representing the north arrow angle P of the unmanned plane in the i-1 imagei-1Representing the mean value of the gray levels of the i-1 st image. The scoring function comprehensively measures the difference of gray scale, height and north arrow deflection between two pictures. Instant scoreThe closer to 0, the defect types contained in the two pictures should be consistent. The result of the ScoreX function is divided into four grades, and the four grades respectively represent the superposition degree of the defect types. That is, the score is [0,0.1) as the first grade, the score is [0.1,0.5) as the second grade, [0.5,1) as the third grade, and ≧ 1 as the fourth grade. The larger the score value is, the higher the defect superposition degree is, and the observation value x is obtained by continuously calculating a ScoreX function.
The segmentation class status is represented as:
S＝{S1,S2,...,Sm},m＝4
calculating a sequence x by using the historical segmentation class sequence Y and frame information corresponding to the historical segmentation class sequence Y, calculating model parameters by using a maximum judicial method, and recording a mathematical expression of the initial probability of the segmentation class as:
π＝{π1,π2,...,πm},m＝4
wherein piiIndicates that the picture division type is in the state SiThe initial probability of (c).
Through statistical analysis, any state S can be obtainediNumber of occurrences is NiMatched directional state S via one-step transferiTo the other direction state SjThen the state transition probability is:
Figure BDA0002772239780000132
all the division type states SiThe one-step transition probability is listed in a matrix form, and then an m × m one-step transition probability matrix is obtained:
Figure BDA0002772239780000141
in the invention, the decisive influence of different heights on the distribution of the segmentation types on the surface of the tower drum is considered, and the defect types of the segmented tower drum can not change under the same height in most cases; under different heights, the probability of the defect variety change of the tower barrel can be greatly improved. In this embodiment, it is determined whether two frames are in a same height relationship according to the information calculation value of the next frame, and then different probability transition matrices are used in different height relationships. Namely, there is a contour transition probability matrix a and a non-contour probability matrix B. From this a hidden markov chain can be determined, the state transition process of which is shown in figure 5.
Since the calculation x belongs to random variables, CHMM is selected for the prediction of segmentation types. Through a large number of experiments, the segmentation state S is foundiThe corresponding frame information calculation value x obeys Gaussian distribution, and the state SiThe number of data points of the corresponding frame information calculation value x is recorded as N'i。
Since there is a one-to-one correspondence between the historical segmentation state sequences Y and the corresponding X, N isi＝N’i. Given the frame information calculation value x, the segmentation status SiMean value μ of the corresponding gaussian density function of xiSum variance
Figure BDA0002772239780000142
Comprises the following steps:
Figure BDA0002772239780000143
Figure BDA0002772239780000144
due to different states SiCorresponding x calculated mean value muiSum variance
Figure BDA0002772239780000145
Different, so at the same distance xkThe probability of occurrence of the state is different for different directions. Given a calculated value xkThe occurrence probability is:
Figure BDA0002772239780000146
according to the Viterbi algorithm, the embodiment only needs to make
Figure BDA0002772239780000147
The probability is maximum, and the probability prediction of the surface defects of the next frame of the wind turbine tower is realized. Meanwhile, the segmentation class with the highest probability is selected as the segmentation result of the next frame, and the structure diagram is shown in fig. 4.
In the step, a novel AAt-Unet model is generated by using Unet fusion Self-attention, Focalloss is used for improving a loss function according to the imaging characteristics of a sea surface fan, a Markov chain prediction algorithm is designed for a segmentation result for reverse detection, the accuracy is improved, and the whole flow chart is shown as 6.
When the next picture segmentation class sequence can be predicted, the class parameters in the activation function softmax in the last 1 × 1 convolutional layer of the network are updated in real time, and there are:
Figure BDA0002772239780000151
where i represents the segmentation index class, C represents the total class number, and Si represents the ratio of the index of the current element to the indices of all elements. Vi represents the output of the model preceding stage output means, and the accuracy of the division is further improved by inputting the C value in real time by predicting the type of division.
S5, dividing the detection result of the coating defect of the offshore wind power tower into two states: a normal state and an abnormal state; according to the processing result of the key frame image detected by the coating of the offshore wind power tower cylinder, which is identified by the camera of the unmanned aerial vehicle, different abnormal states are specifically divided into: state 1: defining the condition that the coating of the offshore wind power tower barrel cracks; state 2: the method is defined as the peeling condition of the coating of the offshore wind power tower; state 3: the corrosion condition of the coating of the offshore wind power tower is defined.
And S6, when the detection result of the coating defect of the offshore wind power tower is in an abnormal state, uploading the corresponding coating image of the coating defect of the offshore wind power tower to a system, and carrying out corresponding maintenance work of the coating of the offshore wind power tower.
The invention also discloses an AA-gate-Unet-based device for detecting the coating defects of the offshore wind power tower, which comprises the following components:
the request receiving module is used for receiving an offshore wind power tower coating defect detection request and acquiring the unmanned aerial vehicle detection video information of the wind power tower coating corresponding to the offshore wind power tower coating defect detection request;
the video extraction module is used for extracting a coating defect image key frame in the coating defect monitoring video information of the offshore wind power tower;
the coating detection module is used for obtaining a defect detection result of the coating of the offshore wind power tower cylinder according to a processing result of the defect coating key frame image detection;
and the tower tube identification module is used for extracting the coating defect picture information of the offshore wind power tower tube and determining the position information of the offshore wind power tower through the identified tower tube information.
The method comprises the steps of collecting images of a tower drum of the offshore wind turbine by using an unmanned aerial vehicle platform, and segmenting defects of the tower drum by using an improved depth learning model AA-gate-Unet; after accurate segmentation, area value data of various defects can be calculated, so that accurate budget and field detection of fan later maintenance are realized.
Although the present invention has been described with reference to a preferred embodiment, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (11)

1. The method for detecting the coating defects of the offshore wind power tower based on AA-gate-Unet is characterized by comprising the following steps: the method comprises the following steps:
s1, receiving an offshore wind power tower coating defect detection request transmitted by a system, and acquiring video information of unmanned detection of a wind power tower coating corresponding to the offshore wind power tower coating defect detection request;
s2, extracting a tower barrel coating image key frame in the video information of the unmanned detection of the wind power tower barrel coating through a deep learning neural network model AA-gate-Unet;
s3, uploading the images of the key frames of the tower barrel coating images to a system cloud;
s4, predicting the image of the tower drum coating image key frame uploaded to the cloud of the system by adopting a prediction algorithm based on HMM and based on Winters three-parameter exponential smoothing to obtain a detection result of the offshore wind power tower drum coating defect;
s5, dividing the detection result of the coating defect of the offshore wind power tower into two states: a normal state and an abnormal state; according to the processing result of the key frame image detected by the coating of the offshore wind power tower cylinder, which is identified by the camera of the unmanned aerial vehicle, different abnormal states are specifically divided into: state 1: defining the condition that the coating of the offshore wind power tower barrel cracks; state 2: the method is defined as the peeling condition of the coating of the offshore wind power tower; state 3: defining the corrosion condition of the coating of the offshore wind power tower;
and S6, when the detection result of the coating defect of the offshore wind power tower is in an abnormal state, uploading the corresponding coating image of the coating defect of the offshore wind power tower to a system, and carrying out corresponding maintenance work of the coating of the offshore wind power tower.
2. The AA-gate-Unet-based offshore wind turbine tower coating defect detection method of claim 1, wherein: the extraction process in step S2 specifically includes:
(1) extracting a model: a bilateral filter is adopted, and the mathematical model is as follows:
spatial distance function:
Figure FDA0002772239770000021
gray scale change function:
Figure FDA0002772239770000022
wherein (x)i,yi) Represents the coordinates of the current position point, (x)c,yc) Representing the filter center position, gray (x)i,yi) Gray value, gray (x) representing the current coordinate pointi,yi) A gray value representing a center coordinate point of the filter;
(2) adaptive threshold: the T value is changed according to the specific characteristics of the image, and the specific process is represented by the following formula:
Figure FDA0002772239770000023
wherein g (x, y) represents a pixel value located at (x, y) in the image;
(3) and detecting the bilateral of the fan outline in the established image data set by adopting a Hough line detection algorithm, and extracting the region of interest.
3. The AA-gate-Unet-based offshore wind turbine tower coating defect detection method of claim 1, wherein: the deep learning neural network model AA-gate-Unet described in step S2 is in a U-shaped structure, and the cutting and transmission of the feature map is established between the up-sampling and the down-sampling.
4. The AA-gate-Unet-based offshore wind turbine tower coating defect detection method of claim 1, wherein: the deep learning neural network model AA-gate-Unet described in step S2 includes a left side contraction path and a right side diffusion path; the left contraction path comprises a plurality of convolution layers and a pooling layer, a plurality of convolution kernels are repeatedly used in the contraction path, and the nonlinear features can be extracted by taking the ReLU as an activation function after convolution.
5. The AA-gate-Unet-based offshore wind turbine tower coating defect detection method of claim 1, wherein: the AA module in the deep learning neural network model AA-gate-uet described in step S2 includes:
(1) by inputting the value of (w, h, c)in) The signature of (a) performs a 1 × 1 convolution of the output QKV matrix, which has a size of (w, h,2 × d)k+dv) Wherein w, h,2 x dk+dVThe width, the length and the depth of the matrix are respectively represented;
(2) and then, the QKV matrix is segmented from the depth channel to obtain three matrixes of Query, Key and Value, wherein the depth channel sizes of the three matrixes are dk、dk、dv；
(3) A multi-head attention mechanism structure is adopted, and the Query matrix, the Key matrix and the Value matrix are respectively divided into N equal matrixes from a depth channel for subsequent calculation, so that the model learns characteristic information in different subspaces;
(4) the weight matrix is calculated by performing matrix multiplication by using two matrixes of Flat _ Q, Flat _ K, and relative position embedding calculation is added on the weight matrix, so that the relative position information of each point on the characteristic diagram is obtained by performing weight calculation on the Q matrix in the length and width directions, and the transformation of the characteristic position is prevented, and the final effect of the model is reduced.
6. The AA-gate-Unet-based offshore wind turbine tower coating defect detection method as claimed in claim 1, wherein the gate module in the deep learning neural network model AA-gate-Unet in step S2 is obtained by adding an attention enhancement convolution module in the down-sampling of the neural network model Unet and adding attentionate in the sampling path on the right layer of Unet to weight the fusion features.
7. The AA-gate-Unet-based offshore wind turbine tower coating defect detection method of claim 1, wherein: the defect classification category is predicted in advance by using a hidden markov chain in step S4.
8. The AA-gate-Unet-based offshore wind turbine tower coating defect detection method of claim 1, wherein: in step S4, the defect classification type is used as a to-be-predicted value, and is set as a hidden chain of the HMM, the GPS height difference and the change in the yaw angle of the north indicator are collected by continuous calculation, and are recorded as an observation chain of the HMM, and the observation data of the image is set.
9. The AA-gate-Unet-based offshore wind turbine tower coating defect detection method of claim 1, wherein: in step S4, the CHMM is selected for prediction of the segmentation class.
10. The AA-gate-Unet-based offshore wind turbine tower coating defect detection method of claim 1, wherein: the step S4 is followed by:
when the segmentation class sequence of the next picture is predicted, the class in the activation function softmax in the last 1 × 1 convolution layer of the network is updated in real time by the parameter, and then:
Figure FDA0002772239770000041
wherein i represents the class of the segmentation index, C represents the total class number, Si represents the ratio of the index of the current element to the indexes of all elements, Vi represents the output of a preceding-stage output unit of the model, and the C value is input in real time through the prediction of the segmentation class.
11. The utility model provides a marine wind power tower section of thick bamboo coating defect detecting device based on AA-gate-Unet which characterized in that: the method comprises the following steps:
the request receiving module is used for receiving an offshore wind power tower coating defect detection request and acquiring the unmanned aerial vehicle detection video information of the wind power tower coating corresponding to the offshore wind power tower coating defect detection request;
the video extraction module is used for extracting a coating defect image key frame in the coating defect monitoring video information of the offshore wind power tower;
the coating detection module is used for obtaining a defect detection result of the coating of the offshore wind power tower cylinder according to a processing result of the defect coating key frame image detection;
and the tower tube identification module is used for extracting the coating defect picture information of the offshore wind power tower tube and determining the position information of the offshore wind power tower through the identified tower tube information.
CN202011253124.2A 2020-11-11 2020-11-11 AA-gate-Unet-based offshore wind power tower coating defect detection method Pending CN112365468A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011253124.2A CN112365468A (en) 2020-11-11 2020-11-11 AA-gate-Unet-based offshore wind power tower coating defect detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011253124.2A CN112365468A (en) 2020-11-11 2020-11-11 AA-gate-Unet-based offshore wind power tower coating defect detection method

Publications (1)

Publication Number Publication Date
CN112365468A true CN112365468A (en) 2021-02-12

Family

ID=74515922

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011253124.2A Pending CN112365468A (en) 2020-11-11 2020-11-11 AA-gate-Unet-based offshore wind power tower coating defect detection method

Country Status (1)

Country Link
CN (1) CN112365468A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112950584A (en) * 2021-03-01 2021-06-11 哈尔滨工程大学 Coating surface defect identification method based on deep learning
CN112967266A (en) * 2021-03-23 2021-06-15 武汉大学 Laser directional energy deposition area calculation method of full convolution neural network
CN115082419A (en) * 2022-07-14 2022-09-20 江苏诺阳家居科技有限公司 Blow-molded luggage production defect detection method
CN115082435A (en) * 2022-07-21 2022-09-20 浙江霖研精密科技有限公司 Defect detection method based on self-supervision momentum contrast
CN115122154A (en) * 2022-08-30 2022-09-30 南通市海门区海纳机械有限公司 Operation detection and control method for metal component machining numerical control milling machine
CN115631197A (en) * 2022-12-21 2023-01-20 中科慧远视觉技术(北京)有限公司 Image processing method, device, medium, equipment and system

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108959245A (en) * 2018-06-08 2018-12-07 浙江大学 The quality determining method of grid equipment defect text
CN109376660A (en) * 2018-10-26 2019-02-22 天宇经纬（北京）科技有限公司 A kind of target monitoring method, apparatus and system
CN109738512A (en) * 2019-01-08 2019-05-10 重庆大学 Nondestructive detection system and method based on multiple physical field fusion
CN109819208A (en) * 2019-01-02 2019-05-28 江苏警官学院 A kind of dense population security monitoring management method based on artificial intelligence dynamic monitoring
CN111127699A (en) * 2019-11-25 2020-05-08 爱驰汽车有限公司 Method, system, equipment and medium for automatically recording automobile defect data
CN111523535A (en) * 2020-04-01 2020-08-11 南通大学 Circle-like object recognition counting detection algorithm based on machine vision and deep learning

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108959245A (en) * 2018-06-08 2018-12-07 浙江大学 The quality determining method of grid equipment defect text
CN109376660A (en) * 2018-10-26 2019-02-22 天宇经纬（北京）科技有限公司 A kind of target monitoring method, apparatus and system
CN109819208A (en) * 2019-01-02 2019-05-28 江苏警官学院 A kind of dense population security monitoring management method based on artificial intelligence dynamic monitoring
CN109738512A (en) * 2019-01-08 2019-05-10 重庆大学 Nondestructive detection system and method based on multiple physical field fusion
CN111127699A (en) * 2019-11-25 2020-05-08 爱驰汽车有限公司 Method, system, equipment and medium for automatically recording automobile defect data
CN111523535A (en) * 2020-04-01 2020-08-11 南通大学 Circle-like object recognition counting detection algorithm based on machine vision and deep learning

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
JIANXIN ZHANG 等: "Attention Gate ResU-Net for Automatic MRI Brain Tumor Segmentation", 《IEEEACCESS》 *
OZAN OKTAY 等: "Attention U-Net:Learning Where to Look for the Pancreas", 《1ST CONFERENCE ON MEDICAL IMAGING WITH DEEP LEARNING (MIDL 2018)》 *
YAN LV 等: "Attention Guided U-Net With Atrous Convolution for Accurate Retinal Vessels Segmentation", 《IEEEACCESS》 *
余胜威 等: "《MATLAB图像滤波去噪分析及其应用》", 30 September 2015, 北京航空航天大学出版社 *
刘瑞祯 等: "《OpenCV教程 基础篇》", 30 June 2007, 北京航空航天大学出版社 *
张堃 等: "大视场大规模目标精确检测算法应用研究", 《仪器仪表学报》 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112950584A (en) * 2021-03-01 2021-06-11 哈尔滨工程大学 Coating surface defect identification method based on deep learning
CN112967266A (en) * 2021-03-23 2021-06-15 武汉大学 Laser directional energy deposition area calculation method of full convolution neural network
CN112967266B (en) * 2021-03-23 2024-02-06 湖南珞佳智能科技有限公司 Laser directional energy deposition area calculation method of full convolution neural network
CN115082419A (en) * 2022-07-14 2022-09-20 江苏诺阳家居科技有限公司 Blow-molded luggage production defect detection method
CN115082435A (en) * 2022-07-21 2022-09-20 浙江霖研精密科技有限公司 Defect detection method based on self-supervision momentum contrast
CN115122154A (en) * 2022-08-30 2022-09-30 南通市海门区海纳机械有限公司 Operation detection and control method for metal component machining numerical control milling machine
CN115122154B (en) * 2022-08-30 2023-08-18 广东昭明电子集团股份有限公司 Operation detection and control method of numerical control milling machine for machining metal components
CN115631197A (en) * 2022-12-21 2023-01-20 中科慧远视觉技术(北京)有限公司 Image processing method, device, medium, equipment and system

Similar Documents

Publication Publication Date Title
CN112365468A (en) AA-gate-Unet-based offshore wind power tower coating defect detection method
Bartler et al. Automated detection of solar cell defects with deep learning
CN111047554B (en) Composite insulator overheating defect detection method based on instance segmentation
CN113538391B (en) Photovoltaic defect detection method based on Yolov4 and thermal infrared image
CN111339882B (en) Power transmission line hidden danger detection method based on example segmentation
CN111368690B (en) Deep learning-based video image ship detection method and system under influence of sea waves
CN111598098B (en) Water gauge water line detection and effectiveness identification method based on full convolution neural network
CN113313107B (en) Intelligent detection and identification method for multiple types of diseases on cable surface of cable-stayed bridge
CN115170816A (en) Multi-scale feature extraction system and method and fan blade defect detection method
Xiaoxun et al. Research on crack detection method of wind turbine blade based on a deep learning method
CN115240069A (en) Real-time obstacle detection method in full-fog scene
CN111767826A (en) Timing fixed-point scene abnormity detection method
CN117058396A (en) Fan blade defect area rapid segmentation method and system based on artificial intelligence
CN116994161A (en) Insulator defect detection method based on improved YOLOv5
Zhao et al. Ocean ship detection and recognition algorithm based on aerial image
CN115049856A (en) Fan blade fault detection method and system based on improved YOLOv5
Tian et al. Feature fusion–based preprocessing for steel plate surface defect recognition
Liu et al. Channel-Spatial attention convolutional neural networks trained with adaptive learning rates for surface damage detection of wind turbine blades
Cao et al. IDS-Net: Integrated Network for Identifying Dust State of Photovoltaic Panels
Cao et al. Detection and classification of surface defects of magnetic tile based on SE-U-Net
CN117115175B (en) Online monitoring and early warning method and system for oil spilling
Wang et al. Power Plant Liquid Leakage Detection Based on Key-frame Quality Enhancement
Liu et al. An Algorithm for Recognition of Foreign Objects in Transmission Lines with Small Samples
CN118247245A (en) Automatic detection method for dam underwater defects based on underwater robot and improved YOLOv8
Elyanboiy et al. Advanced intelligent fault detection for solar panels: incorporation of dust coverage ratio calculation

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

Application publication date: 20210212

RJ01 Rejection of invention patent application after publication