CN117011222A - Cable buffer layer defect detection method, device, storage medium and equipment - Google Patents
Cable buffer layer defect detection method, device, storage medium and equipment Download PDFInfo
- Publication number
- CN117011222A CN117011222A CN202310577087.8A CN202310577087A CN117011222A CN 117011222 A CN117011222 A CN 117011222A CN 202310577087 A CN202310577087 A CN 202310577087A CN 117011222 A CN117011222 A CN 117011222A
- Authority
- CN
- China
- Prior art keywords
- cable
- module
- feature
- defect detection
- tomography slice
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000007547 defect Effects 0.000 title claims abstract description 153
- 238000001514 detection method Methods 0.000 title claims abstract description 123
- 238000002591 computed tomography Methods 0.000 claims abstract description 167
- 238000012549 training Methods 0.000 claims abstract description 67
- 238000002372 labelling Methods 0.000 claims abstract description 45
- 238000013528 artificial neural network Methods 0.000 claims abstract description 10
- 238000000034 method Methods 0.000 claims description 51
- 230000006870 function Effects 0.000 claims description 20
- 238000005259 measurement Methods 0.000 claims description 15
- 238000007781 pre-processing Methods 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 15
- 238000013507 mapping Methods 0.000 claims description 13
- 238000013527 convolutional neural network Methods 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 10
- 238000012216 screening Methods 0.000 claims description 10
- 238000012937 correction Methods 0.000 claims description 9
- 230000004927 fusion Effects 0.000 claims description 8
- 230000003044 adaptive effect Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 7
- 238000011176 pooling Methods 0.000 claims description 7
- 230000001629 suppression Effects 0.000 claims description 7
- 238000012952 Resampling Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 5
- 238000012360 testing method Methods 0.000 description 7
- 230000006872 improvement Effects 0.000 description 5
- 238000013526 transfer learning Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 3
- 238000002679 ablation Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
- G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/766—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using regression, e.g. by projecting features on hyperplanes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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
- G06T2207/10072—Tomographic images
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Probability & Statistics with Applications (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
Abstract
The invention provides a cable buffer layer defect detection method, which comprises the steps of constructing and optimizing an original model based on a cascade mask region convolution neural network, and generating a cable computer tomography slice defect detection model; training the cable computer tomography slice defect detection model by utilizing the plurality of three-dimensional computer tomography slice image data and the corresponding labeling files to obtain a weight file with optimal performance; and detecting the cable buffer layer computer tomography slice image by using the cable computer tomography slice defect detection model containing the weight file with the optimal performance, and identifying the cable buffer layer defect. The invention solves the problem of lower model detection precision of the target detection model in the prior art, and has the effect of improving the detection precision of the cable CT slice defect detection model.
Description
Technical Field
The present invention relates to the field of deep learning target detection technology, and in particular, to a method, an apparatus, a storage medium, and a device for detecting a cable buffer layer defect.
Background
The high voltage cable buffer layer ablation failure is one of the main factors of the high voltage cable failure. The defect of the buffer layer of the high-voltage cable is usually detected by adopting Computer Tomography (CT) detection imaging, and the defect existing in the defect is generally identified by manually detecting an imaging picture by CT detection. In recent years, the development of deep learning breaks through a plurality of difficult visual problems, improves the level of image cognition, accelerates the progress of related technologies in the field of target detection, and becomes a mainstream research direction when the image feature automatic learning method in the deep learning is applied to the defect detection of industrial CT images.
The deep learning-based target detection model is mainly divided into two major categories, namely one-stage and two-stage, wherein the one-stage target detection model directly predicts the category and the position of a target by using a convolutional neural network, and realizes classification and regression in one step. In addition, when the common target detection model calculates the intersection ratio (IoU) of the suggested candidate frame and the real labeling frame, the suggested candidate frame is generally divided into positive and negative samples according to a single threshold value set for the intersection ratio, the number of positive samples is generally far greater than that of negative samples, the positive and negative samples are sampled in a test stage, so that the ratio of the positive and negative samples meets a certain ratio, but in the test stage, the quality of the suggested candidate frame is lower and the detection precision is lower due to the fact that the comparison of the real labeling frame is not available. Therefore, a model with a cascade structure is needed to break through the bottleneck of high-voltage cable defect detection precision under a single threshold, so that samples of the model after each regression can be closer to the real position of the high-voltage cable defect, and the model is suitable for the distribution of different suggested candidate frames.
The object detection model in the prior art has the problem of lower model detection precision.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention provides a cable buffer layer defect detection method, which solves the problem that a target detection model in the prior art has lower model detection precision.
In a first aspect, the present invention provides a method for detecting a defect of a cable buffer layer, comprising: acquiring a plurality of three-dimensional computed tomography slice image data containing cable buffer layer defects and corresponding labeling files; constructing and optimizing an original model based on a cascade mask region convolution neural network, and generating a cable computer tomography slice defect detection model; training the cable computer tomography slice defect detection model by utilizing the plurality of three-dimensional computer tomography slice image data and the corresponding labeling files to obtain a weight file with optimal performance; and detecting the cable buffer layer computer tomography slice image by using the cable computer tomography slice defect detection model containing the weight file with the optimal performance, and identifying the cable buffer layer defect.
Further, training the cable computer tomography slice defect detection model by using the plurality of three-dimensional computer tomography slice image data and the corresponding labeling files to obtain a weight file with optimal performance, wherein the training comprises the following steps: feature extraction is carried out on the three-dimensional computed tomography slice image data to generate a feature map; generating a suggestion candidate frame for the feature map; the suggested candidate frames and the corresponding labeling files pass through a three-stage cascade detector to obtain three corresponding defect categories and boundary frame regression parameters under a set intersection ratio threshold; the three-stage cascade detector consists of a target classifier and a boundary box regressor; and screening out a high-probability target frame by using a non-maximum suppression algorithm according to the defect category and the boundary frame regression parameter, and obtaining a final training result to obtain a weight file with optimal performance.
Further, generating a suggested candidate frame for the feature map by utilizing a multi-scale detection algorithm feature pyramid network algorithm; the feature pyramid network algorithm comprises: classifying the layers without changing the size of the feature map into a stage, wherein the features output by the last layer of each stage form a feature layer pyramid; convolving each stage by adopting a convolution check of 1 multiplied by 1, so that the channel numbers of the stages are consistent; the small feature images of the top layer are amplified to the same size as the feature images of the previous stage in a top-down up-sampling mode; respectively fusing the upsampling result and the feature map with the same size generated from bottom to top by an addition method; convolving each fusion result by adopting a convolution check of 3*3; counting the area and the length-width ratio of the rectangular frame of the labeling file through a mean value clustering algorithm, and setting reference frames with different areas to respectively correspond to different characteristic layers; the reference frames for each area are provided with various aspect ratios, and the frames are traversed in a sliding manner on the feature layer to generate suggested candidate frames.
Further, the boundary regression box regressor is defined as follows:where x represents a sub-image block and b represents a sample Distribution, f represents bounding box regressor, f 1 、f 2 、f 3 Three different cross ratio thresholds are set, f 1 The output of (2) is f 2 Input of f 2 The output of (2) is f 3 Is input of { f 1 ，f 2 ，f 3 Optimizing for resampling distribution at different stages.
Further, training the cable computed tomography slice defect detection model by using the plurality of three-dimensional computed tomography slice image data and the corresponding annotation files, further comprising: mapping the suggested candidate frames to the feature map to obtain corresponding feature matrixes, and uniformly adjusting the feature matrixes to a specified size through feature projection.
Optionally, acquiring the plurality of three-dimensional computed tomography slice image data containing the cable buffer defect and the corresponding annotation file includes: scanning the cable by using a special offset center correction computer tomography measuring workpiece to obtain a projection sinogram; extracting a projection edge according to the projection sinogram, and primarily measuring a rotation center offset value; reconstructing a plurality of images within a certain range according to the preliminarily measured rotation center offset value, analyzing the image quality according to an image Point Spread Function (PSF), and accurately measuring the rotation center offset value; controlling a motion system to drive a turntable to move a distance corresponding to the offset value of the accurate value according to the offset value of the rotation center, and correcting the offset of the rotation center; performing three-dimensional image reconstruction according to the corrected projection sinogram scanned by the computer tomography measuring workpiece to obtain three-dimensional computer tomography slice image data; and obtaining a labeling file of the three-dimensional computed tomography slice image data containing the cable buffer layer defect.
Optionally, acquiring the plurality of three-dimensional computed tomography slice image data containing the cable buffer defect and the corresponding annotation file includes: acquiring a plurality of three-dimensional computed tomography slice image raw data containing cable buffer layer defects; preprocessing the original data of the three-dimensional computed tomography slice images to obtain the data of the three-dimensional computed tomography slice images containing the defects of the cable buffer layer; the preprocessing comprises gray stretching, high dynamic range image enhancement and image denoising; the gray scale stretching adopts piecewise linear gray scale stretching, the high dynamic range image enhancement algorithm adopts a self-adaptive non-sharpening mask method algorithm, and the image denoising adopts Gaussian denoising; and obtaining the annotation files corresponding to the image data of the plurality of three-dimensional computed tomography slices.
Further, the calculation formula of the adaptive non-sharpening mask method algorithm is as follows: y (n, m) =x (n, m) +λ·z (n, m), where x (n, m) is an original image, z (n, m) is a high frequency image, y (n, m) is an output image, and λ represents an adaptive function.
Further, the ResNeXt-101 convolutional neural network or the ResNet-50 convolutional neural network comprises a convolutional layer, a pooling layer and an activation layer, wherein the convolutional layer performs feature extraction from the plurality of three-dimensional computed tomography slice image data to generate a feature map; the pooling layer is used for removing redundant information, reducing the quantity of parameters and expanding the receptive field; the activation layer convolves the output layer with an activation function to increase the nonlinearity of the output layer.
Optionally, training the cable computed tomography slice defect detection model by using the plurality of three-dimensional computed tomography slice image data and the corresponding labeling files, and before obtaining the weight file with the optimal performance, further including: utilizing the training weight of the image data set as the pre-training weight of the cable computer tomography slice defect detection model; and updating parameters of the defect detection model by using a gradient descent optimization algorithm.
In a second aspect, the present invention provides a cable buffer layer defect detection apparatus, comprising: the acquisition module is used for acquiring a plurality of three-dimensional computed tomography slice image data containing cable buffer layer defects and corresponding labeling files; the building module is used for building and optimizing the original model based on the cascade mask region convolution neural network to generate a cable computer tomography slice defect detection model; the training module is used for training the cable computer tomography slice defect detection model by utilizing the plurality of three-dimensional computer tomography slice image data and the corresponding labeling files to obtain a weight file with optimal performance; and the detection module is used for detecting the cable buffer layer computer tomography slice image by using the cable computer tomography slice defect detection model containing the weight file with the optimal performance and identifying the cable buffer layer defect.
Optionally, the training module includes: the feature extraction module is used for carrying out feature extraction on the plurality of three-dimensional computed tomography slice image data to generate a feature map; a suggested candidate frame module for generating suggested candidate frames for the feature map; the cascade regression module is used for enabling the suggested candidate frames and the corresponding annotation files to pass through a three-stage cascade detector to obtain three defect types and boundary frame regression parameters corresponding to the set intersection ratio threshold; the three-stage cascade detector consists of a target classifier and a boundary box regressor; and the screening module is used for screening out a high-probability target frame by utilizing a non-maximum suppression algorithm according to the defect category and the boundary frame regression parameter, obtaining a final training result and obtaining a weight file with optimal performance.
Optionally, the suggestion candidate box module generates a suggestion candidate box for the feature map using a multi-scale detection algorithm feature pyramid network algorithm, including: the feature pyramid module is used for classifying the layers which do not change the size of the feature map into a stage, and features output by the last layer of each stage form a feature layer pyramid; a 1×1 convolution module, configured to use a 1×1 convolution kernel to perform convolution on each stage, so that the number of channels of the stages is consistent; the up-sampling module is used for amplifying the small feature images of the top layer to the same size as the feature images of the previous stage in a top-down up-sampling mode; the transverse connection module is used for respectively fusing the up-sampling result and the feature images with the same size generated from bottom to top by an addition method;
3*3 convolution module, which is used to convolve each fusion result by adopting 3*3 convolution check; the reference frame mapping module is used for counting the area and the length-width ratio of the rectangular frames of the labeling file through a mean value clustering algorithm, and setting the reference frames with different areas to respectively correspond to different characteristic layers; and the traversing module is used for setting various length-width ratios of the reference frames of each area, sliding and traversing on the feature layer, and generating suggested candidate frames.
Optionally, the training module further comprises: and the suggested candidate frame mapping module is used for mapping the suggested candidate frames to the feature map to obtain corresponding feature matrixes, and uniformly adjusting the feature matrixes to a specified size through feature projection.
Optionally, the acquiring module includes: the scanning module is used for scanning the cable by using a special offset center correction computer tomography measuring workpiece to obtain a projection sinogram; the preliminary measurement rotation center offset value module is used for extracting projection edges according to the projection sinogram and preliminary measuring rotation center offset values; the accurate measurement rotation center offset module is used for reconstructing a plurality of images in a certain range according to the initial measurement rotation center offset, analyzing the image quality according to the image point spread function and accurately measuring the rotation center offset; the offset value correction module is used for controlling the motion system to drive the turntable to move by a distance corresponding to the offset value of the accurate value according to the offset value of the rotation center, which is accurately measured, so as to correct the offset of the rotation center; the three-dimensional reconstruction module is used for carrying out three-dimensional image reconstruction according to the corrected projection sinogram of the computer tomography measurement workpiece scanning to obtain three-dimensional computer tomography slice image data; the labeling file acquisition module is used for acquiring a labeling file of the three-dimensional computed tomography slice image data containing the cable buffer layer defect.
Optionally, the acquiring module includes: the original data acquisition module is used for acquiring original data of a plurality of three-dimensional computed tomography slice images containing cable buffer layer defects; the preprocessing module is used for preprocessing the original data of the three-dimensional computed tomography slice images to obtain the data of the three-dimensional computed tomography slice images containing the defects of the cable buffer layer; the preprocessing comprises gray stretching, high dynamic range image enhancement and image denoising; the gray scale stretching adopts piecewise linear gray scale stretching, the high dynamic range image enhancement algorithm adopts a self-adaptive non-sharpening mask method, and the image denoising adopts Gaussian denoising; and the annotation file acquisition module is used for acquiring the annotation files corresponding to the plurality of three-dimensional computed tomography slice image data.
Optionally, the apparatus further comprises: the pre-training module is used for utilizing the training weight of the image data set as the pre-training weight of the cable computer tomography slice defect detection model; and the updating module is used for updating parameters of the defect detection model by using a gradient descent optimization algorithm.
In a third aspect, the present invention provides a computer readable storage medium storing at least one instruction for execution by a processor to perform the steps of the above method.
In a fourth aspect, the present invention provides a computer device comprising a processor and a memory; the memory stores at least one instruction for execution by the processor of the steps of the method described above.
The technical principle of the invention is as follows: and acquiring a plurality of three-dimensional computed tomography slice image data containing cable buffer layer defects and corresponding labeling files. And constructing and optimizing an original model based on the cascade mask region convolution neural network to generate a cable computer tomography slice defect detection model. Training the cable computer tomography slice defect detection model by utilizing the three-dimensional computer tomography slice image data and the corresponding labeling files to obtain the optimal weight file, and finally detecting the cable buffer layer computer tomography slice image by using the cable computer tomography slice defect detection model containing the optimal weight file to identify the cable buffer layer defect.
Compared with the prior art, the invention has the following beneficial effects: by introducing the cascade structure, the problem of lower accuracy of the target detection model based on a single threshold is solved, and the problem of lower model detection accuracy of the target detection model in the prior art is solved.
Drawings
FIG. 1 is a flow chart of a method for detecting cable buffer layer defects according to an embodiment of the present invention;
FIG. 2 is a training flow chart of a cable computed tomography slice defect detection model according to another embodiment of the present invention;
FIG. 3 is a flowchart of a feature pyramid network algorithm in accordance with another embodiment of the present invention;
FIG. 4 is a flowchart of acquiring a plurality of three-dimensional computed tomography slice image data and corresponding annotation files according to another embodiment of the invention;
FIG. 5 is a flowchart of a method for detecting a cable buffer layer defect according to another embodiment of the present invention;
fig. 6 is a schematic structural diagram of a computer device according to another embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings and examples.
As shown in fig. 1, an embodiment of the present invention provides a method for detecting a cable buffer layer defect, which includes the following steps:
s1, acquiring a plurality of three-dimensional computed tomography slice image data containing cable buffer layer defects and corresponding labeling files.
S2, constructing and optimizing an original model based on the cascade mask region convolution neural network, and generating a cable computer tomography slice defect detection model.
And S3, training the cable computer tomography slice defect detection model by utilizing the plurality of three-dimensional computer tomography slice image data and the corresponding labeling files to obtain a weight file with optimal performance.
And S4, detecting the cable buffer layer computer tomography slice image by using a cable computer tomography slice defect detection model containing the weight file with the optimal performance, and identifying the cable buffer layer defect.
The detailed working procedure of this embodiment is: and acquiring a plurality of three-dimensional Computed Tomography (CT) slice image data containing cable buffer layer defects and corresponding labeling files. And constructing and optimizing an original model based on the cascade Mask region convolutional neural network, namely, based on the cascade Mask RCNN original model, generating a cable CT slice defect detection model, and adopting a ResNeXt-101 convolutional neural network or a ResNet-50 convolutional neural network as a backbone network. Training the cable CT slice defect detection model by using the plurality of three-dimensional CT slice image data and the corresponding labeling files to obtain the weight file with the optimal performance, and finally detecting the cable buffer layer CT slice image by using the cable CT slice defect detection model with the weight file with the optimal performance to identify the cable buffer layer defect. Because the model introduces a cascade structure, the problem that the accuracy of the target detection model based on a single threshold is low is solved, and the effect that the defect of the cable buffer layer can be automatically identified and the detection accuracy of the detection model is higher is produced.
In some embodiments, the pixel size of the plurality of three-dimensional CT slice image data is about 1000×1000, wherein the defect type is only one, the plurality of three-dimensional CT slice image data and the corresponding annotation file are randomly divided into a training set and a test set according to a ratio of 8:2, and 10% of the data are randomly divided from the training set as a verification set. And training the cable CT slice defect detection model by using a training set, and selecting a weight file with optimal performance by using a verification set for the training result of each round. And then testing on a test set by using a model containing a weight file with optimal performance, evaluating the performance of the model, and finally detecting the CT slice image of the cable buffer layer by using a cable CT slice defect detection model with good performance after evaluation and obviously improving the performance, so as to identify the defect of the cable buffer layer. The cable CT slice defect detection model and other target detection algorithms are respectively trained and tested by using the training set and the testing set, and the accuracy rate AP and the recall rate RP are used as evaluation indexes to compare experimental results. As shown in the table:
| Method | AP | AR |
| STCT | 0.843 | 0.907 |
| transfer learning STCT | 0.876 | 0.935 |
| STCT+RCT | 0.439 | 0.529 |
Analysis of the above experimental results may give: compared with the traditional two-stage target detection model, the cable CT slice defect detection model of the embodiment has obvious improvement on the AP and AR of the ablation defect detection of the cable buffer layer.
Further, the ResNeXt-101 convolutional neural network or the ResNet-50 convolutional neural network comprises a convolutional layer, a pooling layer and an activation layer, wherein the convolutional layer performs feature extraction from the plurality of three-dimensional computed tomography slice image data to generate a feature map; the pooling layer is used for removing redundant information, reducing the quantity of parameters and expanding the receptive field; the activation layer convolves the output layer with an activation function to increase the nonlinearity of the output layer. Therefore, the ResNeXt-101 convolutional neural network or the ResNet-50 convolutional neural network is utilized, the structure is simple, the modularization is realized, the number of super-parameters required to be manually adjusted is small, and the calculated amount is small.
In the embodiment of the present invention, as shown in fig. 2, the step S3 of training the cable computed tomography slice defect detection model by using the plurality of three-dimensional computed tomography slice image data and the corresponding labeling files, and obtaining the weight file with the optimal performance includes the following steps:
and S31, extracting features of the three-dimensional computed tomography slice image data to generate a feature map.
S32, generating a suggestion candidate frame for the feature map.
S33, the suggested candidate frames and the corresponding labeling files pass through a three-stage cascade detector to obtain three corresponding defect categories and boundary frame regression parameters under a set intersection ratio threshold; the three-stage cascade detector consists of a target classifier and a bounding box regressor.
And S34, screening out a high-probability target frame by using a non-maximum suppression algorithm according to the defect category and the boundary frame regression parameter, and obtaining a final training result to obtain a weight file with optimal performance.
The detailed working procedure of this embodiment is: and carrying out feature extraction on the plurality of three-dimensional computed tomography slice image data to generate a feature map, and generating a suggestion candidate frame for the feature map. In target detection, the cross-over (IoU) threshold is often used to define positive/negative. The threshold value used to train the detector defines the quality of the detector. Although the usual threshold value of 0.5 results in noise detection, if the threshold value is large, the detection performance tends to be degraded. There are two reasons for this paradox of high quality detection: (1) overfitting, due to the disappearance of positive samples for large thresholds; (2) Inferred time quality mismatch between detector and test hypothesis. To address this problem, a multi-stage target detection architecture, namely a cascade RCNN, is proposed that consists of a series of detectors trained to raise the IoU threshold, the detectors being trained sequentially, using the output of the detector as the training set for the next detector. This resampling progressively improves the quality of the hypothesis, ensures that the positive training set of all detectors is equal in size, and minimizes overfitting. The implementation of cascaded RCNN achieves the most advanced performance on the COCO dataset. The annotation file is made as a COCO dataset. The three-stage cascade detector includes a target classifier and a bounding box regressor optimized for a corresponding IoU threshold. And the three-level cascade detector obtains three corresponding defect categories and bounding box regression parameters under the set IoU threshold according to the suggested candidate boxes and the corresponding labeling files. And finally, screening out a high-probability target frame by utilizing a non-maximum suppression algorithm according to the defect category and the boundary frame regression parameter, and obtaining a final training result to obtain a weight file with optimal performance. Based on the further improvement, the three-stage cascade detector is adopted, so that the effect of minimizing the overfitting and improving the detection precision of the model is generated.
Further, the bounding box regressor is defined as follows:wherein x represents a sub-image block, b represents sample distribution, f represents a bounding box regressor, f 1 、f 2 、f 3 Three different cross ratio thresholds are set, f 1 The output of (2) is f 2 Input of f 2 The output of (2) is f 3 Is input of { f 1 ，f 2 ，f 3 Optimizing for resampling distribution at different stages. Compared with the traditional image classification, the target detection not only needs to realize the classification of the target, but also solves the problem of positioning the target, namely, acquiring the position information of the target in the original image. All the bounding regression boxes have to do is to use some mapping relation so that the mapped target boxes of the suggested candidate boxes are infinitely close to the real target boxes.
In some embodiments, the total loss function L defining the training model includes two parts: the calculation formula of the total loss function is as follows:
in the above formula: l (L) cls Representing objectsClassified loss function, L loc Loss function representing bounding box regression, { b t The sample distribution of the different training phases t is represented and b is present t ＝f t-1 (x t-1 ,b t-1 )，h t Representing the target classifier, f t Represents a boundary regressor, g represents the corresponding x t Mu represents a compromise coefficient, X t Represents x t Corresponding label, x t Representing an image block.
Optionally, the step S3 further includes mapping the suggested candidate frame to the feature map to obtain a corresponding feature matrix, and uniformly adjusting the feature matrix to a specified size through feature projection. The feature projection (ROI alignment) is provided in a mask region convolution neural network, so that the problem of region mismatch caused by twice quantization in the ROIPooling operation is well solved, and the accuracy of a detection model can be improved by replacing ROIPooling with the ROI alignment in a detection task.
In another embodiment of the present invention, as shown in fig. 3, for the step S32, a suggested candidate box is generated for the feature map by using a multi-scale detection algorithm feature pyramid network algorithm, where the feature pyramid network algorithm includes the following steps:
s320, classifying the layers without changing the size of the feature map into a stage, wherein the features output by the last layer of each stage form a feature layer pyramid.
S321, convolving each stage by adopting a convolution check of 1×1 so that the channel numbers of the stages are consistent.
And S322, enlarging the small feature map of the top layer to the same size as the feature map of the previous stage in a top-down up-sampling mode.
S323, fusing the up-sampling result and the feature map with the same size generated from bottom to top by an addition method.
S324, convolving each fusion result by adopting a convolution check of 3*3.
S325, counting the area and the length-width ratio of the rectangular frame of the labeling file through a mean value clustering algorithm, and setting the reference frames with different areas to respectively correspond to different characteristic layers.
S326, setting multiple length-width ratios of the reference frames of each area, sliding and traversing on the feature layer, and generating suggested candidate frames.
The detailed working procedure of this embodiment is: utilizing a multi-scale detection algorithm feature pyramid network algorithm (FPN algorithm) to improve RPN to generate a suggestion candidate frame for the feature map, as shown in the following table:
firstly, in the forward propagation process of the ResNeXt-101 network, classifying the layers without changing the size of the feature map into one stage (stage), forming a feature layer pyramid by the features output by the last layer of each stage, firstly, in the bottom-up process, adopting downsampling of a set step length, recording the output of the last residual block of each stage as { C1, C2, C3, C4, C5}, namely 5 stages of the FPN network, and acquiring global context features. In order to strengthen the original characteristics, attention mechanisms are introduced into the model, in { C3, C4 and C5} of ResNeXt-101, attention weights are obtained by adopting 1X 1 convolution and softmax, the importance degree of each channel is obtained, and the number of the characteristic layer channels of all levels is kept consistent. And secondly, carrying out nonlinear processing by adopting a ReLU activation function after convolution and normalization of 1 multiplied by 1. The method comprises the steps of amplifying a small feature image of a top layer to the same size as a feature image of a previous stage in a top-down sampling mode, and then aggregating global context features onto features of each position by utilizing addition to form a long-distance dependency relationship, namely fusing an up-sampling result and feature images with the same size generated from bottom to top respectively through an addition method, namely transversely connecting. And finally, convolving each fusion result through a convolution check of 3 multiplied by 3 to obtain a final characteristic layer of P= { P2, P3, P4, P5}. Feature layer P after feature fusion is subjected to statistical marking through a mean value clustering algorithm (K-means algorithm) The area and the length-width ratio of a rectangular marking frame in the note file are set to five areas {32 ] 2 ,64 2 ,128 2 ,256 2 ,512 2 The reference frames (anchors) of the } are respectively corresponding to the { P2, P3, P4, P5, P6} five feature layers, wherein P5 is independently downsampled by 0.5 times to form a P6 feature layer, the P6 feature layer corresponds to an area 162, each anchor of the areas is provided with seven aspect ratios {1:10,1:5,1:2,1:1,2:1,5:1,10:1}, and the generated anchors slide through the feature layers to generate suggested candidate frames. Based on the above further improvement, when the previous RPN performs object detection, the RPN acts on the last layer, this is not problematic in detecting a large object, but there is some problem in detecting a small object, because for a small object, when convolution pooling is performed to the last layer, the semantic information is not yet actually available, and because the cross connection in the FPN fuses the bottom layer feature information with high resolution with the high layer feature information with high semantics, the processed low layer feature and the processed high layer feature are accumulated, so that the purpose is that the low layer feature can provide more accurate position information, fuse the multi-layer feature information, and output at different features, resulting in the effects of reducing the calculation amount, improving the calculation speed, and improving the accuracy of object detection.
In another embodiment of the present application, as shown in fig. 4, the step S1 includes the following steps:
s11, scanning the cable by using a special offset center correction computer tomography measuring workpiece to obtain a projection sinogram.
And S12, extracting projection edges according to the projection sinogram, and primarily measuring a rotation center offset value.
S13, reconstructing a plurality of images in a certain range according to the preliminarily measured rotation center offset value, analyzing the image quality according to an image point spread function, and accurately measuring the rotation center offset value.
S14, controlling a motion system to drive a turntable to move a distance corresponding to the offset value of the accurate value according to the offset value of the rotation center, and correcting the offset of the rotation center.
And S15, performing three-dimensional image reconstruction according to the corrected projection sinogram of the computed tomography measurement workpiece scanning, and obtaining three-dimensional computed tomography slice image data.
S16, acquiring a labeling file of the three-dimensional computer tomography slice image data containing the cable buffer layer defect.
The detailed working procedure of this embodiment is: the rotation center offset value is a pixel offset of a projection position of the rotation center of the turntable on the detector relative to the center position of the detector. In Computed Tomography (CT) scanning systems, accurate measurement of the projection position (center of projection) of the center of rotation on the detector is very important. Even small errors in the measured and actual projection centers distort the reconstructed image and create artifacts. The application adjusts the parameters of the equipment to determine the rotation center in the process of acquiring CT slice image data; extracting a projection edge according to the projection sinogram, and primarily measuring a rotation center offset value; reconstructing a plurality of images within a certain range according to the preliminarily measured rotation center offset value, analyzing the image quality according to an image point spread function (PSF function), and accurately measuring the rotation center offset value; controlling a motion system to drive a turntable to move a distance corresponding to the offset value of the accurate value according to the offset value of the rotation center, and correcting the offset of the rotation center; and carrying out three-dimensional image reconstruction according to the corrected projection sinogram scanned by the CT measurement workpiece to obtain three-dimensional CT slice image data. Based on the further improvement, the PSF function is utilized to screen the high-quality image so as to accurately measure the rotation center offset value and correct the rotation center offset value, so that the calculation time is greatly reduced and the image quality is ensured in the process of not determining one rotation offset value reconstruction every time after each scanning, thereby providing the high-quality image for model training and further improving the accuracy of model training.
Optionally, step S1 obtains a plurality of three-dimensional computed tomography slice image data containing cable buffer defects and corresponding annotation files, including: acquiring a plurality of three-dimensional computed tomography slice image raw data containing cable buffer layer defects; preprocessing the original data of the three-dimensional computed tomography slice images to obtain the data of the three-dimensional computed tomography slice images containing the defects of the cable buffer layer; the preprocessing comprises gray stretching, high dynamic range image enhancement and image denoising; the gray scale stretching adopts piecewise linear gray scale stretching, the high dynamic range image enhancement algorithm adopts a self-adaptive non-sharpening mask method algorithm, and the image denoising adopts Gaussian denoising; and obtaining the annotation files corresponding to the image data of the plurality of three-dimensional computed tomography slices. In this way, the cable computer tomography slice image is preprocessed to obtain the enhanced and normalized image, so that the image quality of model training is higher, and the model training accuracy is further improved.
Further, the calculation formula of the adaptive non-sharpening mask method algorithm is as follows:
y (n, m) =x (n, m) +λ·z (n, m), where x (n, m) is an original image, z (n, m) is a high frequency image, y (n, m) is an output image, and λ represents an adaptive function. The idea of the adaptive non-sharpening mask method (CLAHE algorithm) is to take a larger value at the image details to increase the contrast, while in the original image the contrast is better to take a smaller value to ensure the contrast of the whole image.
As shown in fig. 5, the embodiment of the invention further provides another cable buffer layer defect detection method, which includes the following steps:
s51, acquiring a plurality of three-dimensional computed tomography slice image data containing cable buffer layer defects and corresponding labeling files;
s52, constructing and optimizing an original model based on a cascade mask region convolution neural network, and generating a cable computer tomography slice defect detection model;
s53, utilizing the training weight of the image data set as the pre-training weight of the cable computer tomography slice defect detection model;
s54, updating parameters of the defect detection model by using a gradient descent optimization algorithm.
S55, training the cable computer tomography slice defect detection model by utilizing the plurality of three-dimensional computer tomography slice image data and the corresponding labeling files to obtain a weight file with optimal performance;
S56, detecting the cable buffer layer computer tomography slice image by using the cable computer tomography slice defect detection model containing the weight file with the optimal performance, and identifying the cable buffer layer defect.
The detailed working procedure of this embodiment is: when the cable Computer Tomography (CT) slice defect detection model is trained by using the plurality of three-dimensional computer tomography slice image data and the corresponding labeling files, an image dataset (ImageNet dataset) training weight can be used as a pre-training weight of the cable CT slice defect detection model, and a gradient descent optimization algorithm (Nadam algorithm) is used for updating parameters of the defect detection model. A method of transfer learning is employed. Based on the further improvement, the method of transfer learning and meta learning is adopted, so that the identification accuracy of the cable CT slice defect detection algorithm on a small sample data set can be improved, the generalization capability of a transfer learning model is improved through the difference of different reconstruction data, the pre-training models of STCT and RCT are fused, and the mask area convolution neural network model based on meta learning is used, so that the effect of improving the algorithm performance of transfer learning is achieved.
An embodiment of the present invention provides a cable buffer layer defect detection device, including: the acquisition module is used for acquiring a plurality of three-dimensional computed tomography slice image data containing cable buffer layer defects and corresponding labeling files; the building module is used for building and optimizing the original model based on the cascade mask region convolution neural network to generate a cable computer tomography slice defect detection model; the training module is used for training the cable computer tomography slice defect detection model by utilizing the plurality of three-dimensional computer tomography slice image data and the corresponding labeling files to obtain a weight file with optimal performance; and the detection module is used for detecting the cable buffer layer computer tomography slice image by using the cable computer tomography slice defect detection model containing the weight file with the optimal performance and identifying the cable buffer layer defect.
In some embodiments, the training module comprises: the feature extraction module is used for carrying out feature extraction on the plurality of three-dimensional computed tomography slice image data to generate a feature map; a suggested candidate frame module for generating suggested candidate frames for the feature map; the cascade regression module is used for enabling the suggested candidate frames and the corresponding annotation files to pass through a three-stage cascade detector to obtain three defect types and boundary frame regression parameters corresponding to the set intersection ratio threshold; the three-stage cascade detector consists of a target classifier and a boundary box regressor; and the screening module is used for screening out a high-probability target frame by utilizing a non-maximum suppression algorithm according to the defect category and the boundary frame regression parameter, obtaining a final training result and obtaining a weight file with optimal performance.
In some embodiments, the suggested candidate box module generates suggested candidate boxes for the feature map using a multi-scale detection algorithm feature pyramid network algorithm, comprising: the feature pyramid module is used for classifying the layers which do not change the size of the feature map into a stage, and features output by the last layer of each stage form a feature layer pyramid; a 1×1 convolution module, configured to use a 1×1 convolution kernel to perform convolution on each stage, so that the number of channels of the stages is consistent; the up-sampling module is used for amplifying the small feature images of the top layer to the same size as the feature images of the previous stage in a top-down up-sampling mode; the transverse connection module is used for respectively fusing the up-sampling result and the feature images with the same size generated from bottom to top by an addition method; 3*3 convolution module, which is used to convolve each fusion result by adopting 3*3 convolution check; the reference frame mapping module is used for counting the area and the length-width ratio of the rectangular frames of the labeling file through a mean value clustering algorithm, and setting the reference frames with different areas to respectively correspond to different characteristic layers; and the traversing module is used for setting various length-width ratios of the reference frames of each area, sliding and traversing on the feature layer, and generating suggested candidate frames.
In some embodiments, the training module further comprises: and the suggested candidate frame mapping module is used for mapping the suggested candidate frames to the feature map to obtain corresponding feature matrixes, and uniformly adjusting the feature matrixes to a specified size through feature projection.
In some embodiments, the acquisition module comprises: the scanning module is used for scanning the cable by using a special offset center correction computer tomography measuring workpiece to obtain a projection sinogram; the preliminary measurement rotation center offset value module is used for extracting projection edges according to the projection sinogram and preliminary measuring rotation center offset values; the accurate measurement rotation center offset module is used for reconstructing a plurality of images in a certain range according to the initial measurement rotation center offset, analyzing the image quality according to the image point spread function and accurately measuring the rotation center offset; the offset value correction module is used for controlling the motion system to drive the turntable to move by a distance corresponding to the offset value of the accurate value according to the offset value of the rotation center, which is accurately measured, so as to correct the offset of the rotation center; the three-dimensional reconstruction module is used for carrying out three-dimensional image reconstruction according to the corrected projection sinogram of the computer tomography measurement workpiece scanning to obtain three-dimensional computer tomography slice image data; the labeling file acquisition module is used for acquiring a labeling file of the three-dimensional computed tomography slice image data containing the cable buffer layer defect.
In some embodiments, the acquisition module comprises: the original data acquisition module is used for acquiring original data of a plurality of three-dimensional computed tomography slice images containing cable buffer layer defects; the preprocessing module is used for preprocessing the original data of the three-dimensional computed tomography slice images to obtain the data of the three-dimensional computed tomography slice images containing the defects of the cable buffer layer; the preprocessing comprises gray stretching, high dynamic range image enhancement and image denoising; the gray scale stretching adopts piecewise linear gray scale stretching, the high dynamic range image enhancement algorithm (HDR image enhancement algorithm) adopts a self-adaptive non-sharpening mask method, and the image denoising adopts Gaussian denoising; and the annotation file acquisition module is used for acquiring the annotation files corresponding to the plurality of three-dimensional computed tomography slice image data.
In some embodiments, the apparatus further comprises: the pre-training module is used for utilizing the training weight of the image data set as the pre-training weight of the cable computer tomography slice defect detection model; and the updating module is used for updating parameters of the defect detection model by using a gradient descent optimization algorithm.
The present invention provides a computer readable storage medium storing at least one instruction for execution by a processor to perform the steps of the above method. Wherein the storage medium may be a volatile or nonvolatile computer readable storage medium.
Based on the same technical conception, a computer device is also provided. Referring to fig. 6, a schematic structural diagram of a computer device according to an embodiment of the present invention includes a processor, a memory, and a bus. The memory is used for storing execution instructions, and comprises a memory and an external memory; the memory is also called an internal memory, and is used for temporarily storing operation data in the processor and data exchanged with an external memory such as a hard disk, and the processor exchanges data with the external memory through the memory. The memory is specifically used for storing and executing a program logic code corresponding to a scanning control method of the ray source and/or a filtered back projection image reconstruction algorithm according to the embodiment of the invention, and is controlled and executed by the processor. That is, when the computer device is running, the processor and the memory communicate via the bus, such that the processor executes the application code stored in the memory, which in turn is used to control the complete reconstruction of the scan control method and/or the filtered backprojection image reconstruction algorithm of one of the embodiments. The processor may be an integrated circuit chip having signal processing capabilities. The processor may be a general-purpose processor, including a central processing unit, a network processor, etc.; but also digital signal processors, application specific integrated circuits, field programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The disclosed methods, steps, and logic blocks in the embodiments of the present invention may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The memory may be, but is not limited to, random access memory, read-only memory, programmable read-only memory, erasable read-only memory, electrically erasable read-only memory, etc.
It will be appreciated that the architecture illustrated in fig. 6 is not intended to constitute a particular limitation of computer devices. In actual use, the computer device may include more or fewer components than shown, or may combine certain components, or split certain components, or a different arrangement of components. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.
Finally, it is noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered by the scope of the claims of the present invention.
Claims (19)
1. A method for detecting a cable buffer layer defect, comprising:
acquiring a plurality of three-dimensional computed tomography slice image data containing cable buffer layer defects and corresponding labeling files;
constructing and optimizing an original model based on a cascade mask region convolution neural network, and generating a cable computer tomography slice defect detection model;
Training the cable computer tomography slice defect detection model by utilizing the plurality of three-dimensional computer tomography slice image data and the corresponding labeling files to obtain a weight file with optimal performance;
and detecting the cable buffer layer computer tomography slice image by using the cable computer tomography slice defect detection model containing the weight file with the optimal performance, and identifying the cable buffer layer defect.
2. The method for detecting cable buffer layer defect according to claim 1, wherein training the cable computed tomography slice defect detection model by using the plurality of three-dimensional computed tomography slice image data and the corresponding annotation files, and obtaining the weight file with optimal performance comprises:
feature extraction is carried out on the three-dimensional computed tomography slice image data to generate a feature map;
generating a suggestion candidate frame for the feature map;
the suggested candidate frames and the corresponding labeling files pass through a three-stage cascade detector to obtain three corresponding defect categories and boundary frame regression parameters under a set intersection ratio threshold; the three-stage cascade detector consists of a target classifier and a boundary box regressor;
And screening out a high-probability target frame by using a non-maximum suppression algorithm according to the defect category and the boundary frame regression parameter, and obtaining a final training result to obtain a weight file with optimal performance.
3. The method of claim 2, wherein generating suggested candidate boxes for the feature map is generating suggested candidate boxes for the feature map using a multi-scale detection algorithm feature pyramid network algorithm;
the feature pyramid network algorithm comprises:
classifying the layers without changing the size of the feature map into a stage, wherein the features output by the last layer of each stage form a feature layer pyramid;
convolving each stage by adopting a convolution check of 1 multiplied by 1, so that the channel numbers of the stages are consistent;
the small feature images of the top layer are amplified to the same size as the feature images of the previous stage in a top-down up-sampling mode;
respectively fusing the upsampling result and the feature map with the same size generated from bottom to top by an addition method;
convolving each fusion result by adopting a convolution check of 3*3;
counting the area and the length-width ratio of the rectangular frame of the labeling file through a mean value clustering algorithm, and setting reference frames with different areas to respectively correspond to different characteristic layers;
The reference frames for each area are provided with various aspect ratios, and the frames are traversed in a sliding manner on the feature layer to generate suggested candidate frames.
4. The method for detecting cable buffer fault as recited in claim 2, wherein said bounding box regressor is defined as follows:
wherein x represents a sub-image block, b represents sample distribution, f represents a bounding box regressor, f 1 、f 2 、f 3 Three different cross ratio thresholds are set, f 1 The output of (2) is f 2 Input of f 2 The output of (2) is f 3 Is input of { f 1 ，f 2 ，f 3 Optimizing for resampling distribution at different stages.
5. The method of claim 2, wherein training the cable computed tomography slice defect detection model using the plurality of three-dimensional computed tomography slice image data and corresponding annotation files, further comprises:
mapping the suggested candidate frames to the feature map to obtain corresponding feature matrixes, and uniformly adjusting the feature matrixes to a specified size through feature projection.
6. The method for detecting cable buffer defect according to claim 1, wherein obtaining a plurality of three-dimensional computed tomography slice image data containing cable buffer defects and corresponding markup files comprises:
Scanning the cable by using a special offset center correction computer tomography measuring workpiece to obtain a projection sinogram;
extracting a projection edge according to the projection sinogram, and primarily measuring a rotation center offset value;
reconstructing a plurality of images in a certain range according to the preliminarily measured rotation center offset value, analyzing the image quality according to an image point spread function, and accurately measuring the rotation center offset value;
controlling a motion system to drive a turntable to move a distance corresponding to the offset value of the accurate value according to the offset value of the rotation center, and correcting the offset of the rotation center;
performing three-dimensional image reconstruction according to the corrected projection sinogram scanned by the computer tomography measuring workpiece to obtain three-dimensional computer tomography slice image data;
and obtaining a labeling file of the three-dimensional computed tomography slice image data containing the cable buffer layer defect.
7. The method for detecting cable buffer defect according to claim 1, wherein obtaining a plurality of three-dimensional computed tomography slice image data containing cable buffer defects and corresponding markup files comprises:
acquiring a plurality of three-dimensional computed tomography slice image raw data containing cable buffer layer defects;
Preprocessing the original data of the three-dimensional computed tomography slice images to obtain the data of the three-dimensional computed tomography slice images containing the defects of the cable buffer layer; the preprocessing comprises gray stretching, high dynamic range image enhancement and image denoising;
the gray scale stretching adopts piecewise linear gray scale stretching, the high dynamic range image enhancement algorithm adopts a self-adaptive non-sharpening mask method, and the image denoising adopts Gaussian denoising;
and obtaining the annotation files corresponding to the image data of the plurality of three-dimensional computed tomography slices.
8. The method for detecting cable buffer layer defect according to claim 7, wherein the calculation formula of the adaptive non-sharpening mask method is as follows: y (n, m) =x (n, m) +λ·z (n, m), where x (n, m) is an original image, z (n, m) is a high frequency image, y (n, m) is an output image, and λ represents an adaptive function.
9. The cable buffer layer defect detection method of claim 1, wherein the ResNeXt-101 convolutional neural network or the ResNet-50 convolutional neural network comprises a convolutional layer, a pooling layer and an activation layer, and the convolutional layer performs feature extraction from the plurality of three-dimensional computed tomography slice image data to generate a feature map; the pooling layer is used for removing redundant information, reducing the quantity of parameters and expanding the receptive field; the activation layer convolves the output layer with an activation function to increase the nonlinearity of the output layer.
10. The method for detecting cable buffer layer defect according to claim 1, wherein training the cable computed tomography slice defect detection model by using the plurality of three-dimensional computed tomography slice image data and corresponding labeling files, and before obtaining the weight file with optimal performance, further comprises:
utilizing the training weight of the image data set as the pre-training weight of the cable computer tomography slice defect detection model;
and updating parameters of the defect detection model by using a gradient descent optimization algorithm.
11. A cable buffer layer defect detection apparatus, comprising:
the acquisition module is used for acquiring a plurality of three-dimensional computed tomography slice image data containing cable buffer layer defects and corresponding labeling files;
the building module is used for building and optimizing the original model based on the cascade mask region convolution neural network to generate a cable computer tomography slice defect detection model;
the training module is used for training the cable computer tomography slice defect detection model by utilizing the plurality of three-dimensional computer tomography slice image data and the corresponding labeling files to obtain a weight file with optimal performance;
And the detection module is used for detecting the cable buffer layer computer tomography slice image by using the cable computer tomography slice defect detection model containing the weight file with the optimal performance and identifying the cable buffer layer defect.
12. The cable buffer defect detection apparatus of claim 11 wherein the training module comprises:
the feature extraction module is used for carrying out feature extraction on the plurality of three-dimensional computed tomography slice image data to generate a feature map;
a suggested candidate frame module for generating suggested candidate frames for the feature map;
the cascade regression module is used for enabling the suggested candidate frames and the corresponding annotation files to pass through a three-stage cascade detector to obtain three defect types and boundary frame regression parameters corresponding to the set intersection ratio threshold; the three-stage cascade detector consists of a target classifier and a boundary box regressor;
and the screening module is used for screening out a high-probability target frame by utilizing a non-maximum suppression algorithm according to the defect category and the boundary frame regression parameter, obtaining a final training result and obtaining a weight file with optimal performance.
13. The cable buffer layer defect detection apparatus of claim 12, wherein the suggestion candidate box module generates a suggestion candidate box for the feature map using a multi-scale detection algorithm feature pyramid network algorithm, comprising:
The feature pyramid module is used for classifying the layers which do not change the size of the feature map into a stage, and features output by the last layer of each stage form a feature layer pyramid;
a 1×1 convolution module, configured to use a 1×1 convolution kernel to perform convolution on each stage, so that the number of channels of the stages is consistent;
the up-sampling module is used for amplifying the small feature images of the top layer to the same size as the feature images of the previous stage in a top-down up-sampling mode;
the transverse connection module is used for respectively fusing the up-sampling result and the feature images with the same size generated from bottom to top by an addition method;
3*3 convolution module, which is used to convolve each fusion result by adopting 3*3 convolution check;
the reference frame mapping module is used for counting the area and the length-width ratio of the rectangular frames of the labeling file through a mean value clustering algorithm, and setting the reference frames with different areas to respectively correspond to different characteristic layers;
and the traversing module is used for setting various length-width ratios of the reference frames of each area, sliding and traversing on the feature layer, and generating suggested candidate frames.
14. The cable buffer defect detection apparatus of claim 12 wherein the training module further comprises:
And the suggested candidate frame mapping module is used for mapping the suggested candidate frames to the feature map to obtain corresponding feature matrixes, and uniformly adjusting the feature matrixes to a specified size through feature projection.
15. The cable buffer defect detection apparatus of claim 11, wherein the acquisition module comprises:
the scanning module is used for scanning the cable by using a special offset center correction computer tomography measuring workpiece to obtain a projection sinogram;
the preliminary measurement rotation center offset value module is used for extracting projection edges according to the projection sinogram and preliminary measuring rotation center offset values;
the accurate measurement rotation center offset module is used for reconstructing a plurality of images in a certain range according to the initial measurement rotation center offset, analyzing the image quality according to the image point spread function and accurately measuring the rotation center offset;
the offset value correction module is used for controlling the motion system to drive the turntable to move by a distance corresponding to the offset value of the accurate value according to the offset value of the rotation center, which is accurately measured, so as to correct the offset of the rotation center;
the three-dimensional reconstruction module is used for carrying out three-dimensional image reconstruction according to the corrected projection sinogram of the computer tomography measurement workpiece scanning to obtain three-dimensional computer tomography slice image data;
The labeling file acquisition module is used for acquiring a labeling file of the three-dimensional computed tomography slice image data containing the cable buffer layer defect.
16. The cable buffer defect detection apparatus of claim 11, wherein the acquisition module comprises:
the original data acquisition module is used for acquiring original data of a plurality of three-dimensional computed tomography slice images containing cable buffer layer defects;
the preprocessing module is used for preprocessing the original data of the three-dimensional computed tomography slice images to obtain the data of the three-dimensional computed tomography slice images containing the defects of the cable buffer layer; the preprocessing comprises gray stretching, high dynamic range image enhancement and image denoising; the gray scale stretching adopts piecewise linear gray scale stretching, the high dynamic range image enhancement algorithm adopts a self-adaptive non-sharpening mask method, and the image denoising adopts Gaussian denoising;
and the annotation file acquisition module is used for acquiring the annotation files corresponding to the plurality of three-dimensional computed tomography slice image data.
17. The cable buffer defect detection apparatus of claim 11, wherein the apparatus further comprises:
The pre-training module is used for utilizing the training weight of the image data set as the pre-training weight of the cable computer tomography slice defect detection model;
and the updating module is used for updating parameters of the defect detection model by using a gradient descent optimization algorithm.
18. A computer-readable storage medium, characterized by: the storage medium stores at least one instruction for execution by a processor to implement a cable buffer layer defect detection method according to any one of claims 1-10.
19. A computer device, characterized by: the computer device includes a processor and a memory; the memory stores at least one instruction for execution by the processor to implement a cable buffer fault detection method as claimed in any one of claims 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310577087.8A CN117011222A (en) | 2023-05-19 | 2023-05-19 | Cable buffer layer defect detection method, device, storage medium and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310577087.8A CN117011222A (en) | 2023-05-19 | 2023-05-19 | Cable buffer layer defect detection method, device, storage medium and equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117011222A true CN117011222A (en) | 2023-11-07 |
Family
ID=88573450
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310577087.8A Pending CN117011222A (en) | 2023-05-19 | 2023-05-19 | Cable buffer layer defect detection method, device, storage medium and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117011222A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117253197A (en) * | 2023-11-20 | 2023-12-19 | 国网天津市电力公司培训中心 | Power cable buffer layer state monitoring method, system and equipment |
| CN118150613A (en) * | 2024-03-26 | 2024-06-07 | 世秀新材料科技(浙江)有限公司 | Detection system of waterborne polyurethane coating resin |
-
2023
- 2023-05-19 CN CN202310577087.8A patent/CN117011222A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117253197A (en) * | 2023-11-20 | 2023-12-19 | 国网天津市电力公司培训中心 | Power cable buffer layer state monitoring method, system and equipment |
| CN118150613A (en) * | 2024-03-26 | 2024-06-07 | 世秀新材料科技(浙江)有限公司 | Detection system of waterborne polyurethane coating resin |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111784671B (en) | Pathological image focus region detection method based on multi-scale deep learning | |
| CN111814867B (en) | Training method of defect detection model, defect detection method and related device | |
| CN110189255B (en) | Face detection method based on two-stage detection | |
| CN111524478B (en) | Apparatus and method for detecting failure | |
| CN111325739B (en) | Method and device for detecting lung focus and training method of image detection model | |
| CN117011222A (en) | Cable buffer layer defect detection method, device, storage medium and equipment | |
| CN102096917B (en) | Automatic eliminating method for redundant image data of capsule endoscope | |
| CN110619618A (en) | Surface defect detection method and device and electronic equipment | |
| US20130230229A1 (en) | Method for brightness level calculation of the digital x-ray image for medical applications | |
| CN111275660B (en) | Flat panel display defect detection method and device | |
| CN112669323B (en) | Image processing method and related equipment | |
| CN111798416A (en) | Intelligent glomerulus detection method and system based on pathological image and deep learning | |
| CN109815931B (en) | Method, device, equipment and storage medium for identifying video object | |
| CN116012291A (en) | Industrial part image defect detection method and system, electronic equipment and storage medium | |
| CN112365497A (en) | High-speed target detection method and system based on Trident Net and Cascade-RCNN structures | |
| CN115909006B (en) | Mammary tissue image classification method and system based on convolution transducer | |
| CN114627106A (en) | Weld defect detection method based on Cascade Mask R-CNN model | |
| CN116682112B (en) | Polysaccharide test data storage and digitizing method | |
| CN113763384B (en) | Defect detection method and defect detection device in industrial quality inspection | |
| CN116109558A (en) | Method for screening fetal chromosomal abnormalities based on neck semitransparent thickness measurement | |
| CN115439654A (en) | Method and system for finely dividing weakly supervised farmland plots under dynamic constraint | |
| CN113793357A (en) | Bronchopulmonary segment image segmentation method and system based on deep learning | |
| CN117173145A (en) | Method, device, equipment and storage medium for detecting surface defects of power equipment | |
| CN113256670A (en) | Image processing method and device, and network model training method and device | |
| CN111626379A (en) | X-ray image detection method for pneumonia |
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 |