CN114626470B - Aircraft skin key feature detection method based on multi-type geometric feature operator - Google Patents
Aircraft skin key feature detection method based on multi-type geometric feature operator Download PDFInfo
- Publication number
- CN114626470B CN114626470B CN202210268285.1A CN202210268285A CN114626470B CN 114626470 B CN114626470 B CN 114626470B CN 202210268285 A CN202210268285 A CN 202210268285A CN 114626470 B CN114626470 B CN 114626470B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- features
- rivet
- feature
- operator
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 16
- 210000001503 joint Anatomy 0.000 claims abstract description 50
- 238000000034 method Methods 0.000 claims abstract description 48
- 230000008447 perception Effects 0.000 claims abstract description 20
- 238000004458 analytical method Methods 0.000 claims abstract description 15
- 238000000605 extraction Methods 0.000 claims abstract description 15
- 238000005070 sampling Methods 0.000 claims description 24
- 230000004927 fusion Effects 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 9
- 238000010586 diagram Methods 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 6
- 230000007246 mechanism Effects 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 238000012847 principal component analysis method Methods 0.000 claims description 3
- 238000011010 flushing procedure Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 238000012797 qualification Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an aircraft skin key feature detection method based on a multi-type geometric feature operator, which comprises the following steps: acquiring three-dimensional point cloud data of a large aircraft skin; constructing an operator graph set containing multiple types of 2D geometric features for each point in the three-dimensional point cloud data; inputting the constructed multi-type 2D geometric feature operator graph set into a trained double-attention and multi-scale perception point cloud classification network, learning the features of rivets and butt joints in a plurality of operators, and classifying three-dimensional point cloud data; extracting rivet contour features in the rivet region and extracting butt seam contour features in the butt seam region by adopting a multi-level model fitting algorithm; and carrying out rivet flushness analysis and butt joint step analysis based on the extracted rivet profile features and the butt joint profile features. The method for detecting the butt joint of the rivets on the surface of the aircraft realizes the accurate and automatic extraction of the rivets with small scale key characteristics and the small butt joint on the surface of the aircraft skin, and improves the assembly quality of the aircraft.
Description
Technical Field
The invention relates to the field of three-dimensional point cloud model detection, in particular to an aircraft skin key feature detection method based on a multi-type geometrical feature operator.
Background
The assembly of large aircraft combines millions of aircraft parts and connectors according to design and technical requirements, and gradually connects the parts, components, large components and the whole machine. Because of the large number of large aircraft parts, various configurations and complex assembly layers and mutual constraint relations, various errors of the assembly parts are mutually overlapped and transferred in the step-by-step assembly process, so that the joint surfaces cannot be tightly combined to generate a butt joint with a length of tens of meters; in addition, due to the special nature of the materials from which large aircraft are made, it is determined that the skin cannot be welded to the fuselage, but rather that millions of rivets are required to secure the skin to the fuselage. The aviation manufacturing industry has strict standard requirements on two key characteristics of butt joint and rivet, and if the two indexes of the butt joint step difference and the rivet Ji Pingdu are not controlled in the design range, the structural strength of the large aircraft can be reduced, and the safety hazard is caused. Therefore, analysis of the exterior skin key features of large aircraft is required.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides an aircraft skin key feature detection method based on a multi-type geometric feature operator, which realizes the accurate and automatic extraction of small-scale key feature rivets and small butt joints on the surface of the aircraft skin and improves the assembly quality of the aircraft.
In order to achieve the above purpose, the invention adopts the following technical scheme: an aircraft skin key feature detection method based on a multi-type geometric feature operator comprises the following steps:
(1) Acquiring three-dimensional point cloud data of a large aircraft skin;
(2) Constructing an operator graph set containing multiple types of 2D geometric features for each point in the three-dimensional point cloud data based on multiple geometric attributes of the three-dimensional point cloud data;
(3) Inputting the multi-type 2D geometric feature operator atlas constructed in the step (2) into a trained double-attention and multi-scale perception point cloud classification network, learning the features of rivets and butt joints in a plurality of operators, and classifying three-dimensional point cloud data into a rivet region, a butt joint region and a non-feature region;
(4) Extracting rivet contour features in the rivet region and extracting butt seam contour features in the butt seam region by adopting a multi-level model fitting algorithm;
(5) And carrying out rivet flushness analysis and butt joint step analysis based on the extracted rivet profile features and the butt joint profile features.
Further, step (1) comprises the following sub-steps:
(1.1) acquiring three-dimensional point cloud data of the aircraft skin for multiple times by using a LeicaATS960 absolute tracker, wherein the three-dimensional point cloud data acquired for multiple times covers a key feature area to be detected on the surface of the aircraft skin;
and (1.2) splicing the three-dimensional point cloud data acquired for multiple times into the three-dimensional point cloud data of the complete large aircraft skin by using an iterative closest point algorithm.
Further, step (2) comprises the following sub-steps:
(2.1) calculating distances from all points in the three-dimensional point cloud data to each point p in the three-dimensional point cloud data, taking the nearest k points as a neighborhood point set, and calculating normal vectors of the points p and all points in the neighborhood point set by using a principal component analysis method; taking a point p as an origin, taking a normal vector of the point p as a Z axis, establishing a local coordinate system, projecting the point p and a neighborhood point set thereof to an XY-plane of the local coordinate system, constructing a minimum square bounding box with edges parallel to an X axis and a Y axis, and dividing the bounding box into n multiplied by n plane grids;
(2.2) calculating the projection height, normal vector, gaussian curvature, average curvature and density of the neighborhood points in each grid to encode the geometric information of the neighborhood points to form a multi-type 2D geometric feature arithmetic diagram, wherein the multi-type 2D geometric feature arithmetic diagram comprises: a height map operator, a normal map operator, a curvature map operator, and a dense map operator.
Further, the dual-attention and multi-scale sensing point cloud classification network in the step (3) is formed by sequentially connecting a multi-type geometrical feature operator set feature extraction module, a dual-attention feature enhancement module and a key feature point classification module; the multi-type geometrical feature operator set feature extraction module extracts features of the multi-type 2D geometrical feature operator graph by using a convolution neural network with shared weights; the dual-attention feature enhancement module consists of a self-attention module and a channel attention SK network module, wherein the self-attention module uses a self-attention mechanism to acquire the weight of each data in the 2D geometrical feature operator features through a convolution layer and a softmax activation function for the features of each type of 2D geometrical feature operators, carries out Hadamard product operation with the corresponding 2D geometrical feature operator features to acquire the self-attention enhancement features of the 2D geometrical feature operator graphs, inputs the self-attention enhancement features into the channel attention SK network module, learns the contribution degree of the self-attention enhancement features of different types of geometrical feature operator graphs to final classification, and outputs multi-operator fusion features; the key feature point classification module is formed by sequentially connecting a multi-scale sensing network module MSP, a convolution layer and a softmax activation function, and the multi-scale sensing network module MSP carries out convolution, pooling and addition operations of different scales on the multi-operator fusion features to obtain multi-scale fusion features; and then, the multi-scale fusion characteristics are subjected to convolution operation and softmax activation function to obtain the probability of each point of the three-dimensional point cloud data on the rivet area, the butt joint area and the non-characteristic area.
Further, the training process of the double-attention and multi-scale perception point cloud classification network in the step (3) specifically comprises the following steps:
(a) Inputting a multi-type 2D geometrical characteristic operator graph set into a multi-type geometrical characteristic operator set characteristic extraction module in the double-attention and multi-scale perception point cloud classification network, and extracting operator characteristics;
(b) Inputting the extracted operator features into a double-attention feature enhancement module in a double-attention and multi-scale perception point cloud classification network, obtaining a weight matrix with the same dimension as the operator features through convolution and softmax operation on each operator feature, carrying out Hadamard product operation on the weight matrix and the operator features to obtain self-attention enhancement features, splicing the self-attention enhancement features of all operators to obtain splicing features, inputting the splicing features into an SK channel attention module, carrying out convolution and softmax operation on channel dimensions to obtain a weight matrix on the channel dimensions, and carrying out Hadamard product operation on the weight matrix and the splicing features to obtain multi-operator fusion features;
(c) Inputting the multi-operator fusion features into a key feature point classification module in a double-attention and multi-scale perception point cloud classification network, predicting the probability of each three-dimensional point cloud data on a rivet region, a butt joint region and a non-feature region, and taking the region with the maximum probability value as the region where the three-dimensional point cloud data is located;
(d) Repeating the steps (a) - (c) until the cross entropy loss function converges, and completing training of the double-attention and multi-scale perception point cloud classification network.
Further, the specific process of extracting the profile features of the rivet in the rivet region in the step (4) is as follows:
(4.1.1) randomly sampling point cloud subsets from all three-dimensional point cloud data of the rivet area, wherein the number of the sampled point cloud subsets is larger than the number of rivets, and fitting each point cloud subset into a round model hypothesis by using a random consistency sampling method, and simultaneously obtaining parameters of each round model hypothesis to form a round model hypothesis set; the parameters of the circular model hypothesis comprise a circle center and a radius;
(4.1.2) for each round model hypothesis, retaining the round model hypothesis when the error of its radius from the actual rivet radius is less than a threshold value;
(4.1.3) comparing the distance between each circular model hypothesis in the reserved circular model hypotheses and the circle centers of other circular model hypotheses according to the spatial characteristics of the discrete distribution of the rivets, and clustering the circular model hypotheses with the distance smaller than 2 times of the radius of the rivets into a rivet model hypothesis set;
(4.1.4) merging three-dimensional point cloud data of rivet areas of all round model hypotheses of each rivet model hypothesis set into one point cloud, deleting the point cloud from the rivet areas, fitting the point cloud into circles by a random consistency sampling method, and taking the fitted circles as a rivet outline feature;
(4.1.5) if the rivet profile characteristics obtained in step (4.4) are less than the total number of rivets to be inspected, repeating steps (4.1.1) - (4.1.4) for non-sampled three-dimensional point cloud data in the rivet region.
Further, the process of extracting the butt seam contour features in the butt seam area in the step (4) specifically includes:
(4.2.1) randomly sampling point cloud subsets from all three-dimensional point cloud data of the butt joint region, wherein the number of the sampled point cloud subsets is larger than the number of the butt joints, and fitting each point cloud subset into a curve model hypothesis by using a random consistency sampling method to form a curve model hypothesis set;
(4.2.2) for each curve model hypothesis, retaining the curve model hypothesis when the distance from a point in the butt-joint region point cloud to the fitted curve is less than a distance threshold and the number of points at which the distance is less than the distance threshold is greater than a set point number threshold;
(4.2.3) clustering a set of curve model hypotheses when the distance from the nearest point of the remaining curve model hypotheses to other curve model hypotheses is smaller than a threshold and the included angle of the curve method phase of the remaining curve model hypotheses to the nearest point of the other curve model hypotheses is smaller than a set threshold;
(4.2.4) merging three-dimensional point cloud data of the butt joint region of all curve model assumptions of each curve model assumption set into one point cloud, deleting the point cloud from the butt joint region, fitting the point cloud into two parallel curves by a random consistency sampling method, and taking the two fitted parallel curves as a butt joint contour feature;
(4.2.5) if the butt seam profile feature obtained in step (4.2.4) is less than the total number of butt seams to be detected, repeating steps (4.2.1) - (4.2.4) for three-dimensional point cloud data not sampled in the rivet region.
Further, the rivet flatness analysis in the step (5) specifically includes: and for any rivet profile feature, acquiring local neighborhood point cloud of the rivet according to geometric parameters of the rivet profile feature, fitting a plane by using a random consistency sampling method, calculating the directed distance between each point in the rivet profile and the plane, and judging that the rivet is qualified in flushness when the maximum directed distance is smaller than a set Ji Pingdu threshold value.
Further, the step difference analysis process in the step (5) specifically includes: for any one of the opposite joint contour features, calculating the directional distance D from each point in two curves of the opposite joint contour to the other curve according to the geometric parameters of the opposite joint contour features, obtaining the local neighborhood point cloud of the point, fitting the local neighborhood point cloud to two approximately parallel planes by using a random consistency sampling method, calculating the directional distance of the two planes as the opposite joint step difference F of the point, and then calculating the opposite joint gap of the pointCalculating the contrast step and contrast gap of all points of the contrast contour feature, and taking the average value +.>And average value of butt gap>When average contrast difference->And average butt gap->When the two are smaller than the set threshold values, the pair of seams is determinedAnd (5) qualified.
Compared with the prior art, the invention has the following beneficial effects: according to the aircraft skin key feature detection method disclosed by the invention, a geometric feature operator set is constructed by using various geometric feature information as input, the contribution degree of final classification of each operator feature is learned by using a double-attention module, various feature information is deeply fused, rivets and butt joints with unobvious features on a large aircraft skin are accurately extracted, and the stability and accuracy of rivet profile and butt joint profile extraction are improved by using a multi-level model fitting method, so that the accurate automatic extraction of small-scale key feature rivets and small butt joints on the surface of the aircraft skin is realized, and the aircraft assembly quality is improved.
Drawings
FIG. 1 is a flow chart of an aircraft skin key feature detection method based on a multi-type geometric feature operator of the present invention;
FIG. 2 is a schematic diagram of a multi-type geometric feature operator construction in accordance with the present invention;
FIG. 3 is a schematic diagram of a dual-attention and multi-scale point cloud classification network according to the present invention;
FIG. 4 is a schematic illustration of a rivet region and a butt-seam region on an aircraft skin;
FIG. 5 is a schematic representation of the detection of key features of an aircraft skin by the present invention.
Detailed Description
The technical scheme of the invention is further described in detail below with reference to the accompanying drawings.
FIG. 1 is a flow chart of the aircraft skin key feature detection method based on the multi-type geometric feature operator, which comprises the following steps:
(1) Acquiring three-dimensional point cloud data of a large aircraft skin; the method specifically comprises the following substeps:
(1.1) because the aircraft skin is large in size, all three-dimensional point cloud data cannot be acquired once during scanning, and complete data are required to be acquired in a mode of repeated acquisition and splicing, specifically, the Leica ATS960 absolute tracker is used for acquiring the three-dimensional point cloud data of the aircraft skin for multiple times, the three-dimensional point cloud data acquired for multiple times cover a key feature area to be detected on the surface of the aircraft skin, and the overlapping rate of the three-dimensional point cloud data acquired in two adjacent times is more than 30%;
and (1.2) splicing the three-dimensional point cloud data acquired for multiple times into the three-dimensional point cloud data of the complete large aircraft skin by using an iterative closest point algorithm.
(2) Constructing an operator graph set containing multiple types of 2D geometric features for each point in the three-dimensional point cloud data based on multiple geometric attributes of the three-dimensional point cloud data so as to acquire multiple types of information and improve the recognition accuracy of the rivet region and the butt joint region; as shown in fig. 2, the method specifically comprises the following substeps:
(2.1) calculating distances from all points in the three-dimensional point cloud data to each point p in the three-dimensional point cloud data, taking the nearest k points as a neighborhood point set, wherein the value of k is 32, and calculating normals of all points in the point p and the neighborhood point set by using a principal component analysis method; the point p is used as an origin, the normal vector of the point p is used as a Z axis, a local coordinate system is established, the influence of the position and the direction of point cloud in the three-dimensional space is avoided, and the complexity of learning tasks is reduced; projecting a point p and a neighborhood point set thereof to an XY-plane of a local coordinate system, constructing a minimum square bounding box with edges parallel to an X axis and a Y axis, dividing the bounding box into n multiplied by n plane grids, and converting the unstructured three-dimensional neighborhood point set into a structured 2D grid through the operation, so that the neighborhood point set can be processed by a deep learning module in the image field;
and (2.2) calculating geometric information such as projection height, normal vector, gaussian curvature, density and the like of each point of a neighborhood point set, dividing the neighborhood point set into different grids according to projection positions of each point of the neighborhood point set on a planar network, taking average height, average normal vector, average Gaussian curvature and average density of all points in each grid as multi-type geometric information of the grid, forming a matrix by each type geometric information of the whole planar grid, and taking all the matrices as multi-type 2D geometric feature calculation graphs, wherein the method comprises the following steps of: a height map operator, a normal map operator, a curvature map operator, and a dense map operator. By extracting geometric information of various types of information, the difference features of the rivet region, the butt joint region and the non-feature region can be better perceived and distinguished, and the recognition accuracy is improved.
(3) Inputting the multi-type 2D geometrical characteristic operator atlas constructed in the step (2) into a trained double-attention and multi-scale perception point cloud classification network, learning the characteristics of rivets and butt joints in a plurality of operators, and classifying three-dimensional point cloud data into rivet areas, butt joint areas and non-characteristic areas.
As shown in fig. 3, the dual-attention and multi-scale sensing point cloud classification network in the invention is formed by sequentially connecting a multi-type geometrical feature operator set feature extraction module, a dual-attention feature enhancement module and a key feature point classification module; the multi-type geometrical feature operator set feature extraction module extracts the features of the multi-type 2D geometrical feature operator graph by using a convolution neural network with shared weight, and the multi-type geometrical feature operator set feature extraction module comprises 4 convolution layers, and a maximum pooling layer is arranged behind each 2 convolution layers so as to reduce the dimension of input data and reduce the computational complexity; the dual-attention feature enhancement module consists of a self-attention module and a channel attention SK network module, wherein the self-attention module uses a self-attention mechanism, the self-attention mechanism is used for the features of each type of 2D geometric feature operators, the weight of each data in the 2D geometric feature operator features is obtained through a convolution layer and a softmax activation function, hadamard operation is carried out on the weights of each data and the corresponding 2D geometric feature operator features to obtain self-attention enhancement features of 2D geometric feature operator graphs, the self-attention enhancement features are input into the channel attention SK network module, the contribution degree of the self-attention enhancement features of different types of geometric feature operator graphs to final classification is learned, a multi-operator fusion feature is output, the network can pay more attention to the important information which is helpful for feature region learning through the use of the dual-attention mechanism, and irrelevant detail information is restrained; the key feature point classification module is formed by sequentially connecting a multi-scale sensing network module MSP, a convolution layer and a softmax activation function, and the multi-scale sensing network module MSP carries out convolution, pooling and addition operations of different scales on the multi-operator fusion features to obtain multi-scale fusion features; and then, the multi-scale fusion characteristics are subjected to convolution operation and softmax activation function to obtain the probability of each point of the three-dimensional point cloud data on the rivet area, the butt joint area and the non-characteristic area. According to the invention, the double-attention and multi-scale sensing point cloud classification network can accurately detect rivets and butt joints with unobvious characteristics in the aircraft skin point cloud data.
The training process of the double-attention and multi-scale perception point cloud classification network in the invention comprises the following steps:
(a) Inputting a multi-type 2D geometrical characteristic operator graph set into a multi-type geometrical characteristic operator set characteristic extraction module in the double-attention and multi-scale perception point cloud classification network, and extracting operator characteristics;
(b) Inputting the extracted operator features into a double-attention feature enhancement module in a double-attention and multi-scale perception point cloud classification network, obtaining a weight matrix with the same dimension as the operator features through convolution and softmax operation on each operator feature, carrying out Hadamard product operation on the weight matrix and the operator features to obtain self-attention enhancement features, splicing the self-attention enhancement features of all operators to obtain splicing features, inputting the splicing features into an SK channel attention module, carrying out convolution and softmax operation on channel dimensions to obtain a weight matrix on the channel dimensions, and carrying out Hadamard product operation on the weight matrix and the splicing features to obtain multi-operator fusion features;
(c) Inputting the multi-operator fusion features into a key feature point classification module in a double-attention and multi-scale perception point cloud classification network, predicting the probability of each three-dimensional point cloud data on a rivet region, a butt joint region and a non-feature region, and taking the region with the maximum probability value as the region where the three-dimensional point cloud data is located;
(d) Repeating the steps (a) - (c) until the cross entropy loss function converges, and completing training of the double-attention and multi-scale perception point cloud classification network.
The cross entropy loss function L in the invention is as follows:
wherein k represents the total number of classification areas; i represents a classification region index; y is i Representing classification region accuracy, y when correct i =1; otherwise, y i ＝0；p i Representing the probability of predicting the i-th category.
(4) The rivet contour features in the rivet region and the butt seam contour features in the butt seam region are extracted by adopting a multi-level model fitting algorithm, and the method can obtain a more stable contour feature extraction result; specifically, the specific process of extracting the rivet profile features in the rivet region is:
(4.1.1) randomly sampling point cloud subsets from all three-dimensional point cloud data of the rivet area, wherein the number of the sampled point cloud subsets is larger than the number of rivets, and fitting each point cloud subset into a round model hypothesis by using a random consistency sampling method, and simultaneously obtaining parameters of each round model hypothesis to form a round model hypothesis set; the parameters of the round model hypothesis in the invention comprise a circle center and a radius;
(4.1.2) for each round model hypothesis, retaining the round model hypothesis when the error of the radius from the actual rivet radius is less than a threshold value, which may be 0.3 times the actual rivet radius;
(4.1.3) according to the spatial characteristics of the discrete distribution of the rivets, the rivet spacing is far greater than the rivet radius, comparing the distance between each circular model hypothesis in the reserved circular model hypotheses and the circle centers of other circular model hypotheses, and clustering the circular model hypotheses with the distance smaller than 2 times of the rivet radius into a rivet model hypothesis set;
(4.1.4) merging three-dimensional point cloud data of rivet areas of all round model hypotheses of each rivet model hypothesis set into one point cloud, deleting the point cloud from the rivet areas, fitting the point cloud into circles by a random consistency sampling method, and taking the fitted circles as a rivet outline feature;
(4.1.5) if the rivet profile characteristics obtained in step (4.1.4) are less than the total number of rivets to be inspected, repeating steps (4.1.1) - (4.1.4) for three-dimensional point cloud data not sampled in the rivet area.
The process for extracting the butt seam contour features in the butt seam region in the invention comprises the following steps:
(4.2.1) randomly sampling point cloud subsets from all three-dimensional point cloud data of the butt joint region, wherein the number of the sampled point cloud subsets is larger than the number of the butt joints, and fitting each point cloud subset into a curve model hypothesis by using a random consistency sampling method to form a curve model hypothesis set;
(4.2.2) for each curve model hypothesis, retaining the curve model hypothesis when the distance from a point in the butt-joint region point cloud to the fitted curve is less than a distance threshold and the number of points at which the distance is less than the distance threshold is greater than a set point number threshold; the distance threshold value is 1 cm in the invention; the point threshold is determined by the length L of the curve model assumption and the average nearest point distance d, the point threshold is generally 0.5 x L/d, after all points with the distance smaller than the distance threshold are obtained, the distance between two points with the farthest distance in the points is taken as the length L of the curve model assumption, and the average nearest point distance d is the average value of the distance from each point to the nearest point in the whole point cloud data;
(4.2.3) clustering a set of curve model hypotheses when the distance from the nearest point of the remaining curve model hypotheses to other curve model hypotheses is smaller than a threshold and the included angle of the curve method phase of the remaining curve model hypotheses to the nearest point of the other curve model hypotheses is smaller than a set threshold; the threshold value of the included angle of the method is set to be 10 degrees;
(4.2.4) merging the three-dimensional point cloud data of the butt joint region of all curve model assumptions of each curve model assumption set into one point cloud, deleting the point cloud from the butt joint region, fitting the point cloud into two parallel curves by a random consistency sampling method, and taking the two fitted parallel curves as a butt joint contour feature;
(4.2.5) if the butt seam profile feature obtained in step (4.2.4) is less than the total number of butt seams to be detected, repeating steps (4.2.1) - (4.2.4) for three-dimensional point cloud data not sampled in the rivet region.
(5) The rivet flush degree analysis and the butt joint step analysis are carried out based on the extracted rivet profile features and the extracted butt joint profile features, and specifically, the rivet flush degree analysis process specifically comprises the following steps: for any rivet profile feature, according to geometric parameters of the rivet profile feature, local neighborhood point clouds of the rivet are obtained, a plane is fitted by a random consistency sampling method, and each rivet in the rivet profile is calculatedAnd the directional distance between each point and the plane, and judging that the rivet is qualified in the degree of flushing when the maximum directional distance is smaller than a set Ji Pingdu threshold, wherein the specific degree of flushing threshold is determined by a detection qualification standard, and the degree of flushing threshold is set to be 1 millimeter in one technical scheme of the invention. The process of the opposite joint step analysis specifically comprises the following steps: for any one of the opposite joint contour features, calculating the directional distance D from each point in two curves of the opposite joint contour to the other curve according to the geometric parameters of the opposite joint contour features, obtaining the local neighborhood point cloud of the point, fitting the local neighborhood point cloud to two approximately parallel planes by using a random consistency sampling method, calculating the directional distance of the two planes as the opposite joint step difference F of the point, and then calculating the opposite joint gap of the pointCalculating the contrast step and contrast gap of all points of the contrast contour feature, and taking the average value +.>And average value of butt gap>When average contrast difference->And average butt gap->And when the two joint positions are smaller than the set threshold values respectively, judging that the joint is qualified. The specific gap step threshold and the gap threshold are determined by the detection qualification standard, and in the technical scheme of the invention, the gap step threshold is 1 millimeter and the gap threshold is 3 millimeters.
FIG. 4 is a schematic illustration of a rivet region and a butt-seam region on an aircraft skin; fig. 5 is a schematic diagram of rivets and butt joint features detected by the method for detecting the key features of the aircraft skin based on the multi-type geometrical feature operators, and by comparing fig. 4 and 5, the method for detecting the key features of the aircraft skin can detect all the rivets and the butt joint features on the aircraft skin.
The above is only a preferred embodiment of the present invention, and the scope of the present invention is not limited to the above embodiment, and all technical solutions belonging to the concept of the present invention are within the scope of the present invention. It should be noted that modifications and adaptations to the invention without departing from the principles thereof are intended to be within the scope of the invention as set forth in the following claims.
Claims (4)
1. An aircraft skin key feature detection method based on a multi-type geometrical feature operator is characterized by comprising the following steps of:
(1) Acquiring three-dimensional point cloud data of a large aircraft skin;
(2) Constructing an operator graph set containing multiple types of 2D geometric features for each point in the three-dimensional point cloud data based on multiple geometric attributes of the three-dimensional point cloud data;
(3) Inputting the multi-type 2D geometric feature operator atlas constructed in the step (2) into a trained double-attention and multi-scale perception point cloud classification network, learning the features of rivets and butt joints in a plurality of operators, and classifying three-dimensional point cloud data into a rivet region, a butt joint region and a non-feature region;
the double-attention and multi-scale perception point cloud classification network is formed by sequentially connecting a multi-type geometrical feature operator set feature extraction module, a double-attention feature enhancement module and a key feature point classification module; the multi-type geometrical feature operator set feature extraction module extracts features of the multi-type 2D geometrical feature operator graph by using a convolution neural network with shared weights; the dual-attention feature enhancement module consists of a self-attention module and a channel attention SK network module, wherein the self-attention module uses a self-attention mechanism to acquire the weight of each data in the 2D geometrical feature operator features through a convolution layer and a softmax activation function for the features of each type of 2D geometrical feature operators, carries out Hadamard product operation with the corresponding 2D geometrical feature operator features to acquire the self-attention enhancement features of the 2D geometrical feature operator graphs, inputs the self-attention enhancement features into the channel attention SK network module, learns the contribution degree of the self-attention enhancement features of different types of geometrical feature operator graphs to final classification, and outputs multi-operator fusion features; the key feature point classification module is formed by sequentially connecting a multi-scale sensing network module MSP, a convolution layer and a softmax activation function, and the multi-scale sensing network module MSP carries out convolution, pooling and addition operations of different scales on the multi-operator fusion features to obtain multi-scale fusion features; then, the multi-scale fusion characteristics are subjected to convolution operation and softmax activation function to obtain the probability of each point of the three-dimensional point cloud data on the rivet area, the butt joint area and the non-characteristic area;
the training process of the double-attention and multi-scale perception point cloud classification network specifically comprises the following steps:
(a) Inputting a multi-type 2D geometrical characteristic operator graph set into a multi-type geometrical characteristic operator set characteristic extraction module in the double-attention and multi-scale perception point cloud classification network, and extracting operator characteristics;
(b) Inputting the extracted operator features into a double-attention feature enhancement module in a double-attention and multi-scale perception point cloud classification network, obtaining a weight matrix with the same dimension as the operator features through convolution and softmax operation on each operator feature, carrying out Hadamard product operation on the weight matrix and the operator features to obtain self-attention enhancement features, splicing the self-attention enhancement features of all operators to obtain splicing features, inputting the splicing features into an SK channel attention module, carrying out convolution and softmax operation on channel dimensions to obtain a weight matrix on the channel dimensions, and carrying out Hadamard product operation on the weight matrix and the splicing features to obtain multi-operator fusion features;
(c) Inputting the multi-operator fusion features into a key feature point classification module in a double-attention and multi-scale perception point cloud classification network, predicting the probability of each three-dimensional point cloud data on a rivet region, a butt joint region and a non-feature region, and taking the region with the maximum probability value as the region where the three-dimensional point cloud data is located;
(d) Repeating the steps (a) - (c) until the cross entropy loss function converges, and completing training of the double-attention and multi-scale perception point cloud classification network;
(4) Extracting rivet contour features in the rivet region and extracting butt seam contour features in the butt seam region by adopting a multi-level model fitting algorithm;
the process of extracting the butt seam contour features in the butt seam region specifically comprises the following steps:
(4.2.1) randomly sampling point cloud subsets from all three-dimensional point cloud data of the butt joint region, wherein the number of the sampled point cloud subsets is larger than the number of the butt joints, and fitting each point cloud subset into a curve model hypothesis by using a random consistency sampling method to form a curve model hypothesis set;
(4.2.2) for each curve model hypothesis, retaining the curve model hypothesis when the distance from a point in the butt-joint region point cloud to the fitted curve is less than a distance threshold and the number of points at which the distance is less than the distance threshold is greater than a set point number threshold;
(4.2.3) clustering a set of curve model hypotheses when the distance from the nearest point of the remaining curve model hypotheses to other curve model hypotheses is smaller than a threshold and the included angle of the curve method phase of the remaining curve model hypotheses to the nearest point of the other curve model hypotheses is smaller than a set threshold;
(4.2.4) merging three-dimensional point cloud data of the butt joint region of all curve model assumptions of each curve model assumption set into one point cloud, deleting the point cloud from the butt joint region, fitting the point cloud into two parallel curves by a random consistency sampling method, and taking the two fitted parallel curves as a butt joint contour feature;
(4.2.5) if the butt seam profile feature obtained in the step (4.2.4) is less than the total number of butt seams to be detected, repeating the steps (4.2.1) - (4.2.4) on the three-dimensional point cloud data not sampled in the rivet region;
(5) Performing rivet flushness analysis and butt joint step analysis based on the extracted rivet profile features and the butt joint profile features;
the rivet flatness analysis process specifically comprises the following steps: for any rivet profile feature, according to geometric parameters of the rivet profile feature, local neighborhood point clouds of the rivet are obtained, a random consistency sampling method is used for fitting a plane, the directed distance between each point in the rivet profile and the plane is calculated, and when the maximum directed distance is smaller than a set Ji Pingdu threshold value, the rivet is judged to be qualified in flushness;
the process of the opposite joint step analysis specifically comprises the following steps: for any one of the opposite joint contour features, calculating the directional distance D from each point in two curves of the opposite joint contour to the other curve according to the geometric parameters of the opposite joint contour features, obtaining the local neighborhood point cloud of the point, fitting the local neighborhood point cloud to two approximately parallel planes by using a random consistency sampling method, calculating the directional distance of the two planes as the opposite joint step difference F of the point, and then calculating the opposite joint gap of the pointCalculating the contrast step and contrast gap of all points of the contrast contour feature, and taking the average value +.>And average value of butt gap>When average contrast difference->And average butt gap->And when the two joint positions are smaller than the set threshold values respectively, judging that the joint is qualified.
2. The method for aircraft skin key feature detection based on multi-type geometric feature operators according to claim 1, wherein the step (1) comprises the following sub-steps:
(1.1) acquiring three-dimensional point cloud data of the aircraft skin for multiple times by using a Leica ATS960 absolute tracker, wherein the three-dimensional point cloud data acquired for multiple times covers a key feature area to be detected on the surface of the aircraft skin;
and (1.2) splicing the three-dimensional point cloud data acquired for multiple times into the three-dimensional point cloud data of the complete large aircraft skin by using an iterative closest point algorithm.
3. The method for aircraft skin key feature detection based on multi-type geometric feature operators according to claim 1, wherein the step (2) comprises the following sub-steps:
(2.1) calculating distances from all points in the three-dimensional point cloud data to each point p in the three-dimensional point cloud data, taking the nearest k points as a neighborhood point set, and calculating normal vectors of the points p and all points in the neighborhood point set by using a principal component analysis method; taking a point p as an origin, taking a normal vector of the point p as a Z axis, establishing a local coordinate system, projecting the point p and a neighborhood point set thereof to an XY-plane of the local coordinate system, constructing a minimum square bounding box with edges parallel to an X axis and a Y axis, and dividing the bounding box into n multiplied by n plane grids;
(2.2) calculating the projection height, normal vector, gaussian curvature, average curvature and density of the neighborhood points in each grid to encode the geometric information of the neighborhood points to form a multi-type 2D geometric feature arithmetic diagram, wherein the multi-type 2D geometric feature arithmetic diagram comprises: a height map operator, a normal map operator, a curvature map operator, and a dense map operator.
4. The method for detecting the key features of the aircraft skin based on the multi-type geometrical feature operators according to claim 1, wherein the specific process of extracting the rivet profile features in the rivet region in the step (4) is as follows:
(4.1.1) randomly sampling point cloud subsets from all three-dimensional point cloud data of the rivet area, wherein the number of the sampled point cloud subsets is larger than the number of rivets, and fitting each point cloud subset into a round model hypothesis by using a random consistency sampling method, and simultaneously obtaining parameters of each round model hypothesis to form a round model hypothesis set; the parameters of the circular model hypothesis comprise a circle center and a radius;
(4.1.2) for each round model hypothesis, retaining the round model hypothesis when the error of its radius from the actual rivet radius is less than a threshold value;
(4.1.3) comparing the distance between each circular model hypothesis in the reserved circular model hypotheses and the circle centers of other circular model hypotheses according to the spatial characteristics of the discrete distribution of the rivets, and clustering the circular model hypotheses with the distance smaller than 2 times of the radius of the rivets into a rivet model hypothesis set;
(4.1.4) merging three-dimensional point cloud data of rivet areas of all round model hypotheses of each rivet model hypothesis set into one point cloud, deleting the point cloud from the rivet areas, fitting the point cloud into circles by a random consistency sampling method, and taking the fitted circles as a rivet outline feature;
(4.1.5) if the rivet profile characteristics obtained in step (4.1.4) are less than the total number of rivets to be inspected, repeating steps (4.1.1) - (4.1.4) for three-dimensional point cloud data not sampled in the rivet area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210268285.1A CN114626470B (en) | 2022-03-18 | 2022-03-18 | Aircraft skin key feature detection method based on multi-type geometric feature operator |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210268285.1A CN114626470B (en) | 2022-03-18 | 2022-03-18 | Aircraft skin key feature detection method based on multi-type geometric feature operator |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114626470A CN114626470A (en) | 2022-06-14 |
| CN114626470B true CN114626470B (en) | 2024-02-02 |
Family
ID=81901862
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210268285.1A Active CN114626470B (en) | 2022-03-18 | 2022-03-18 | Aircraft skin key feature detection method based on multi-type geometric feature operator |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114626470B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115965788B (en) * | 2023-01-12 | 2023-07-28 | 黑龙江工程学院 | Point cloud semantic segmentation method based on multi-view image structural feature attention convolution |
| CN115982864B (en) * | 2023-03-21 | 2023-06-27 | 南京航空航天大学 | Reconstruction method for assembly coordination boundary characteristics of large composite material component |
| CN116778260B (en) * | 2023-08-17 | 2023-11-17 | 南京航空航天大学 | Aviation rivet flushness detection method, device and system based on AdaBoost ensemble learning |
| CN117494534B (en) * | 2024-01-03 | 2024-03-22 | 南京航空航天大学 | Aircraft skin repairing method based on point cloud and finite element analysis |
| CN117788472B (en) * | 2024-02-27 | 2024-05-14 | 南京航空航天大学 | Method for judging corrosion degree of rivet on surface of aircraft skin based on DBSCAN algorithm |
Citations (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108151660A (en) * | 2017-12-29 | 2018-06-12 | 西北工业大学 | The measurement equipment of a kind of aircraft components butt-joint clearance and scale, method and system |
| CN108408080A (en) * | 2018-03-29 | 2018-08-17 | 南昌航空大学 | A kind of aircraft wing body Butt Assembling device, method and system |
| CN109598306A (en) * | 2018-12-06 | 2019-04-09 | 西安电子科技大学 | Hyperspectral image classification method based on SRCM and convolutional neural networks |
| CN109596059A (en) * | 2019-01-07 | 2019-04-09 | 南京航空航天大学 | A kind of aircraft skin gap based on parallel lines structure light and scale measurement method |
| CN109886273A (en) * | 2019-02-26 | 2019-06-14 | 四川大学华西医院 | A kind of CMR classification of image segmentation system |
| CN110276445A (en) * | 2019-06-19 | 2019-09-24 | 长安大学 | Domestic communication label category method based on Inception convolution module |
| CN111028220A (en) * | 2019-12-11 | 2020-04-17 | 南京航空航天大学 | Automatic detection method for levelness of point cloud rivet |
| CN111028221A (en) * | 2019-12-11 | 2020-04-17 | 南京航空航天大学 | Airplane skin butt-joint measurement method based on linear feature detection |
| CN111080629A (en) * | 2019-12-20 | 2020-04-28 | 河北工业大学 | Method for detecting image splicing tampering |
| CN111524129A (en) * | 2020-04-29 | 2020-08-11 | 南京航空航天大学 | Aircraft skin butt joint gap calculation method based on end face extraction |
| CN111524154A (en) * | 2020-04-21 | 2020-08-11 | 南京航空航天大学 | Image-based tunnel segment automatic segmentation method |
| CN111582126A (en) * | 2020-04-30 | 2020-08-25 | 浙江工商大学 | Pedestrian re-identification method based on multi-scale pedestrian contour segmentation fusion |
| KR20200122897A (en) * | 2019-04-19 | 2020-10-28 | 서울대학교산학협력단 | System and method for monitoring the forest gaps using airborne lidar datasets |
| CN111862054A (en) * | 2020-07-23 | 2020-10-30 | 南京航空航天大学 | Rivet contour point cloud extraction method |
| CN111967480A (en) * | 2020-09-07 | 2020-11-20 | 上海海事大学 | Multi-scale self-attention target detection method based on weight sharing |
| CN112053361A (en) * | 2020-10-15 | 2020-12-08 | 南京航空航天大学 | Aircraft skin butt joint detection method based on large-scale point cloud |
| CN112053426A (en) * | 2020-10-15 | 2020-12-08 | 南京航空航天大学 | Deep learning-based large-scale three-dimensional rivet point cloud extraction method |
| CN112446591A (en) * | 2020-11-06 | 2021-03-05 | 太原科技大学 | Evaluation system for student comprehensive capacity evaluation and zero sample evaluation method |
| CN113205055A (en) * | 2021-05-11 | 2021-08-03 | 北京知见生命科技有限公司 | Fungus microscopic image classification method and system based on multi-scale attention mechanism |
| CN113378791A (en) * | 2021-07-09 | 2021-09-10 | 合肥工业大学 | Cervical cell classification method based on double-attention mechanism and multi-scale feature fusion |
| CN113935977A (en) * | 2021-10-22 | 2022-01-14 | 河北工业大学 | Solar cell panel defect generation method based on generation countermeasure network |
-
2022
- 2022-03-18 CN CN202210268285.1A patent/CN114626470B/en active Active
Patent Citations (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108151660A (en) * | 2017-12-29 | 2018-06-12 | 西北工业大学 | The measurement equipment of a kind of aircraft components butt-joint clearance and scale, method and system |
| CN108408080A (en) * | 2018-03-29 | 2018-08-17 | 南昌航空大学 | A kind of aircraft wing body Butt Assembling device, method and system |
| CN109598306A (en) * | 2018-12-06 | 2019-04-09 | 西安电子科技大学 | Hyperspectral image classification method based on SRCM and convolutional neural networks |
| CN109596059A (en) * | 2019-01-07 | 2019-04-09 | 南京航空航天大学 | A kind of aircraft skin gap based on parallel lines structure light and scale measurement method |
| CN109886273A (en) * | 2019-02-26 | 2019-06-14 | 四川大学华西医院 | A kind of CMR classification of image segmentation system |
| KR20200122897A (en) * | 2019-04-19 | 2020-10-28 | 서울대학교산학협력단 | System and method for monitoring the forest gaps using airborne lidar datasets |
| CN110276445A (en) * | 2019-06-19 | 2019-09-24 | 长安大学 | Domestic communication label category method based on Inception convolution module |
| CN111028220A (en) * | 2019-12-11 | 2020-04-17 | 南京航空航天大学 | Automatic detection method for levelness of point cloud rivet |
| CN111028221A (en) * | 2019-12-11 | 2020-04-17 | 南京航空航天大学 | Airplane skin butt-joint measurement method based on linear feature detection |
| CN111080629A (en) * | 2019-12-20 | 2020-04-28 | 河北工业大学 | Method for detecting image splicing tampering |
| CN111524154A (en) * | 2020-04-21 | 2020-08-11 | 南京航空航天大学 | Image-based tunnel segment automatic segmentation method |
| CN111524129A (en) * | 2020-04-29 | 2020-08-11 | 南京航空航天大学 | Aircraft skin butt joint gap calculation method based on end face extraction |
| CN111582126A (en) * | 2020-04-30 | 2020-08-25 | 浙江工商大学 | Pedestrian re-identification method based on multi-scale pedestrian contour segmentation fusion |
| CN111862054A (en) * | 2020-07-23 | 2020-10-30 | 南京航空航天大学 | Rivet contour point cloud extraction method |
| CN111967480A (en) * | 2020-09-07 | 2020-11-20 | 上海海事大学 | Multi-scale self-attention target detection method based on weight sharing |
| CN112053361A (en) * | 2020-10-15 | 2020-12-08 | 南京航空航天大学 | Aircraft skin butt joint detection method based on large-scale point cloud |
| CN112053426A (en) * | 2020-10-15 | 2020-12-08 | 南京航空航天大学 | Deep learning-based large-scale three-dimensional rivet point cloud extraction method |
| CN112446591A (en) * | 2020-11-06 | 2021-03-05 | 太原科技大学 | Evaluation system for student comprehensive capacity evaluation and zero sample evaluation method |
| CN113205055A (en) * | 2021-05-11 | 2021-08-03 | 北京知见生命科技有限公司 | Fungus microscopic image classification method and system based on multi-scale attention mechanism |
| CN113378791A (en) * | 2021-07-09 | 2021-09-10 | 合肥工业大学 | Cervical cell classification method based on double-attention mechanism and multi-scale feature fusion |
| CN113935977A (en) * | 2021-10-22 | 2022-01-14 | 河北工业大学 | Solar cell panel defect generation method based on generation countermeasure network |
Non-Patent Citations (5)
| Title |
|---|
| Qian Xie et al.RRCNet: Rivet Region Classification Network for Rivet Flush Measurement Based on 3-D Point Cloud.《 IEEE Transactions on Instrumentation and Measurement》.2021,摘要、第1节第4段、第3、5-6节、图4. * |
| RRCNet: Rivet Region Classification Network for Rivet Flush Measurement Based on 3-D Point Cloud;Qian Xie et al;《 IEEE Transactions on Instrumentation and Measurement》;摘要、第1节第4段、第3、5-6节、图4 * |
| 基于SVM的三维对缝点云间隙阶差提取方法;张波等;《航空制造技术》;第63卷(第7期);47-54 * |
| 基于双目多线结构光的铆钉齐平度测量方法;王德重等;《航空制造技术》;第64卷;57-65 * |
| 面向结构光对缝测量的光条细化方法;丁祖娇;《中国优秀硕士学位论文全文数据库信息科技辑》;第2018年卷(第3期);第5.2.1-5.2.2节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114626470A (en) | 2022-06-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114626470B (en) | Aircraft skin key feature detection method based on multi-type geometric feature operator | |
| CN111080627B (en) | 2D +3D large airplane appearance defect detection and analysis method based on deep learning | |
| CN111028327B (en) | Processing method, device and equipment for three-dimensional point cloud | |
| CN110287932B (en) | Road blocking information extraction method based on deep learning image semantic segmentation | |
| Movaghati et al. | Road extraction from satellite images using particle filtering and extended Kalman filtering | |
| CN112529015A (en) | Three-dimensional point cloud processing method, device and equipment based on geometric unwrapping | |
| CN116167668B (en) | BIM-based green energy-saving building construction quality evaluation method and system | |
| Xie et al. | RRCNet: Rivet region classification network for rivet flush measurement based on 3-D point cloud | |
| CN109284779A (en) | Object detecting method based on the full convolutional network of depth | |
| CN113506376A (en) | Ground three-dimensional point cloud multi-scale closure error checking and splicing method | |
| CN103295239A (en) | Laser-point cloud data automatic registration method based on plane base images | |
| CN113139453A (en) | Orthoimage high-rise building base vector extraction method based on deep learning | |
| CN111860520A (en) | Large airplane point cloud model self-supervision semantic segmentation method based on deep learning | |
| Xu et al. | Power line extraction from mobile LiDAR point clouds | |
| CN111539422A (en) | Flight target cooperative identification method based on fast RCNN | |
| CN114693932B (en) | Large-scale aircraft large-part point cloud semantic segmentation method | |
| CN107292039B (en) | UUV bank patrolling profile construction method based on wavelet clustering | |
| Dai et al. | Aircraft seam feature extraction from 3D raw point cloud via hierarchical multi-structure fitting | |
| Dey et al. | A robust performance evaluation metric for extracted building boundaries from remote sensing data | |
| Yang et al. | Autonomous UAVs landing site selection from point cloud in unknown environments | |
| Yan et al. | Terrain matching positioning method for underwater vehicles based on curvature discrimination | |
| CN113706496A (en) | Aircraft structure crack detection method based on deep learning model | |
| CN116476070B (en) | Method for adjusting scanning measurement path of large-scale barrel part local characteristic robot | |
| CN117274774A (en) | Yolov 7-based X-ray security inspection image dangerous goods detection algorithm | |
| CN114781056B (en) | Aircraft complete machine shape measuring method based on feature matching |
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 |